Claire Selltiz - Research Methods in Social Relations

RESEARCH METHODS IN SOCIAL RELATIONS
CLAIRE SELLTIZ MARIE JAHODA
MORTON DEUTSCH STUART W. COOK
EDITORIAL READERS
ISIDOR CHEIN HAROLD M. PROSHANSKY
PUBLISHED FOR THE SOCIETY FOR THE
PSYCHOLOGICAL STUDY OF SOCIAL ISSUES
Holt, Rinehart and Winston

New York ~ Chicago - San Francisco
Toronto
RESEARCH
METHODS
IN SOCIAL
RELATIONS
Revised
i.
One-Volume Edition
ACKNOWLEDGMENTS
Grateful acknowledgment is made to the following for the use of copyrighted

materials:
University of Minnesota Press for the paragraphs from Interracial Housing:

A Psychological Evaluation of a Social Experiment by Morton Deutsch and Mary
Evans Collins. 1951
Harper & Brothers for quotations from The Authoritarian Personality by T.
\V. Adorno, Else Frenkel-Brunswik, Daniel J. Levinson, and R. Nevitt Sanford.
From On The Witness Stand by Hugo Miinsterberg. Copyright 1908, 1923 by
Doubleday & Co., Inc. Reprinted by permission of the publisher.
Robert F. Bales and Addison-Wesley Publishing Company, Inc., for the re-
production of a chart from Interaction Process Analysis by Robert F. Bales.
Free Press for a passage from The Focused Interview by Robert K. Merton, M.
Fiske, and P. L. Kendall.
Harper & Brothers for permission to quote from Public Opinion and the Indi-
vidual by Gardner Murphy and Rensis Likert. .
The University of Illinois Press for a diagram from The Measurement of Mean-
ing by Charles E. Osgood, G. J. Suci, and P. H. Tannenbaum.
Personnel Psychology, Inc., for permission to quote from an article, "The Validity
of Direct and Indirect Questions in Measuring Job Satisfaction," by Joseph Weitz
and R. C. Nuckols, 1953,6.
Teachers College, Columbia University for a passage from Measurement of Fair-
ruindedness by Goodwin B. Watson, Teachers College, Columbia University
Contributions to Education No. 176. _
Cambridge University Press for an extract from Scientific Explanation: A Study
of the Function ot Theory, Probability and Law in Sci~nce, by R. B. Braithwaite.
The University of Chicago Press for a paragraph from "Fundamentals of Con-
cept Formation in Empirical Science" by Carl G. Hempel, International.Encyclo-
pedia of Unified Science, Vol. II, No.7.
Yale University Press for a selection from Communication and Persuasion: Psy-
chological Studies ot Opinion Change, by Carl 1. Hovland, 1. K. Janis, and H. H.
Kelley.
Columbia University Press for material from Negro Intelligence and Selective
Migration by Otto Klineberg.
The authors also are indebted to Professor" Sir Ronald A. Fisher, Cambridge, and
to Messrs. Oliver and Boyd Ltd., Edinburgh, for permission to reprint an extract
from their book The Design of Experiments.
January, 1967
Copyright 1951, 1959
by Holt, Rinehart and Winston, Inc.
Library of Congress Catalog Card Number: 59-8714
27803-0119
Printed in the United States of America
PREFACE TO THE FIRST EDITION
ALTHOUGH THERE HAVE BEEN many technical publications on specific

aspects of research methods for the study of social relations, few books
bring together on an introductory level the considerations which enter
into every step of the research process. This is what we have set out to
do.
Research methods can be presented in many different ways. The
special emphasis of our presentation arises from a growing concern
among social scientists that their work should contribute to the solu-
tion of practical problems as they arise in the contemporary world.
Experience has demonstrated that research conducted without con-
cern for immediate application is neither easily nor promptly put to
use. Research concerned with immediate application requires through-
out the research process a collaborative effort between social scientists
and those who are to act upon their findings. Such collaboration
creates problems of its own, for which neither partner is fully prepared
by his specific training. This book, therefore, includes a treatment of
such problems.
Wherever possible we have chose~ the illustrative material from
an area in which such collaboration is particularly needed and where
it has already been attempted with some success: the area of preju-
dice. Hostilities between persons of different racial, religious, or na-
tional background present one of the major problems of our times.
To the extent to which we learn to understand the factors which pro-
mote or obstruct harmonious relations amon-g such groups, we may
hope to contribute to the solution of this' problem.
The book is diiided into two parts. Part I, Basic Processes, deals
consecutively with the major steps of a scientific inquiry into social
relations, as welll,as with the interrelationships between the steps.
Part II, Selected Techniques, deals in.more technical detail with some
specific methodological problems.
v
Vl PREFACE
In Part I we have aimed at two groups of readers-those who are
preparing to conduct social research and those who are to use its re-
sults. The former will find in Part I an introduction to the methods of
social research. The latter reader will find in Part I an orientation to
research procedures which will, we hope, prevent him from expecting
either too little or too much from research. If he expects too little, he
will be disinclined to collaborate, to the detriment of further research;
if he expects too much, collaboration must end in disappointment.
Parts I and II together are organized as a text for courses in research
methods. We have experimented with their use as a text in a graduate
course on research methods at New York University and hope that
others will have equally encouraging experiences with its use. In addi-
tion, the social scientist here and abroad who has not specialized in con-
ducting research in social relations will, we hope, find useful the sys-
tematic account of the research process in Part I and the more technical
elaborations of some methodological problems in Part II.
This book is in many ways the outcome of group effort. The idea
of producing it arose ,in a group; its production was financed by several
grolWs; it had the editorial guidance of a group; and it was produced by
agroup~
The Society for the Psychological Study of Social Issues (S.P .S.S.I.),
under whose auspices this book is presented, conducts much of its
scientific and professional activities through committees; one of them,
the Committee on Intergroup Relations, in 1948 received from Gordon
a
W. Allport a suggestion to produce book on the measurement of
prejudice. A subcommittee was established under the chairmanship
of Robert Chin to investigate whether such a book was needed and
whether its production was feasible. Both questions were answered in
the affirmative, and the subcommittee prepared a first outline of the
material to be covered in such a publication. The following member~
of the subcommittee participated actively in the preparation of the
outline: Gordon Allpo'rt, Robert Chin,Harold Kelley, Bernard Kramer,
Bernard Kutner, M. Brewster Smith, Nancy Starbuck, and Babette S.
Whipple.
After the Council of S.P .S.S.I. decided to adopt the project and
support it financially, an editorial committee was appointed consist-
ing of the following members: Hadley Cantril (Princeton), Robert
PREFACE vii
Chin (Boston University), Stuart Cook, chairman (New York Uni-
versity), Eugene Hartley (City College of New York), Samuel Flower-
man (American Jewish Committee), Herbert Hyman (National
Opinion Research Center), Donald McGranahan (United Nations),
Irwin Shannon (National Conference of Christians and Jews), Leo
Srole (Anti-Defamation League of B'nai B'rith), Frank Trager (Anti-
Defamation League of B'nai B'rith), and Goodwin Watson (Teachers
College, Columbia University). The editorial committee reviewed
the outline on several occasions and selected the contributors. The
reading of the final manuscript was delegated by the Council of
S.P.S.S.I. to Theodore Newcomb and Goodwin Watson.
The first planning conferences among the editors resulted in a
recommendation for a shift in focus, based on two considerations. First,
the measurement of prejudice is not fundamentally different from the
measurement of other social relations; secondly, a discussion of meas-
urement techniques alone runs the risk of being misleading unless it is
placed into the broader context of the entire research process. With
the concurrence of the editorial committee, this shift was made: the
plan was modified from ondor a book of techniques for the measure-
ment of prejudice to one on research methods in the study of social re-
lations with especial emphasis on prejudice. This led to the conception
of Part I in its present form. This part was produced by the three au-
thors. Although Marie Jahoda and Morton Deutsch were primarily
responsible for writing the drafts of the various chapters, cooperation
among all three was so close and continuous that it would be impossible
to specify the contribution of anyone individual. The chapters in
Part II were written, with few exceptions, especially for this book by
experts in various special fields.
In the conviction that the production of a book on methods of re-
search in the study of social relations would serve the interests of many
other organizations, the Council of S.P.S.S.1. approached several other
groups to enlist financial support for the undertaking. The production
of the book might not have been possible without the prompt and gen-
erous response of the Anti-Defamation League of B'nai B'rith. This
was followed by a ~ant from the National Conference of Christians
and Jews and a contract with the United Nations Educational, Scien-
tific and Cultural Organization.
PREFACE
We are indebted to a number of colleagues for their generous con-
tribution of time and thought. Gordon Allport has never lost touch
with the development of this perhaps somewhat wayward child of his
imagination. He has carefully and constructively criticized our early
efforts and suggested many helpful avenues of approach. Robert Chin
and M. Brewster Smith collaborated in a two-day conference with the
editors during which the original outline was transformed into the one
underlying this book. M. Brewster Smith and Harold Kelley have given
intensive assistance in reviewing drafts of most of the chapters in Part
1. Both have made many helpful critical suggestions. Isidor Chein has
reviewed several chapters and in addition contributed much to the final
editorial work. Claire Selltiz has collaborated with us in the last revision
of Part 1. She has, through her suggestions, much improved the general
organization and clarity of presentation. To her and to Alicia Hemler
we are also indebted for much editorial assistance with the final manu-
script of the entire book. Robert Lee has given valuable assistance in the
preparation of the bibliography. The secretarial personnel of the Re-
search Center for Human Relations, especialfy Miss Sally Cohen, Mrs.
Virginia Revere, and Mrs. Anita Walkley, have borne patiently with
the typing and retyping of many revisions of the text. Their cheerful co-
operation has helped to reduce the tensions which so often arise in the
attempt to me~t a deadline and coordinate the work of several con-
tributors.
We acknowledge with thanks the permission of the following or-
ganizations for reprinting material originally produced by them: The
United Nations for permission to re'print a condensed version of Th,e
Main Types and Causes of Discrimination in Appendix B; The Bur.eau
of Applied Social Research, Columbia University, for permission to
reprint a condensed version of Training Guide on Constructing Ques-
tionnaires and Interview Schedules in Chapt~r 12; and the Princeton
Press for permission to reprint a slightly modified version of Chapter I
of Measureinent and Prediction in Chapter 21.
Research Center for Human Relations

at New York University J
MARIE AHODA
March 1951 MORTON DEUTSCH
STUART W. COOK
PREFACE TO THE REVISED EDITION
IN REVISING Research Methods in Social Relations, we have had two

m,ain goals: to bring the book up to date, and to organize it as a one-
volume text suitable for use in undergraduate courses as well as in in-
troductory graduate courses in research methods in social psychology
and sociology.
In the years since the first edition was prepared, there has been a con-
siderable increase in the number of colleges offering undergraduate
courses in research methods in the social sciences. Teachers who have
used the earlier edition in such courses have reported that students had
difficulty with some parts of Volume One and much of Volume Two.
It seemed to us that the trouble in Volume One sprang from occasional
complications of style and from an occasional taking-for-granted that
readers were familiar with certain concepts, rather than from any for-
midable difficulty in the subject matter. On the /
other hand, some
sections of Volume Two were highly technical and demanded, for their
understanding, background beyond what could be expected of an
undergraduate student or what could be provided within the text. In
the present version we have tried to clarify the material that was in
Volume One by simplifying the language and by expanding some of
the discussion; and we have incorporated some material from Volume
Two that seems to be within the grasp of students without extensive
technical background. Although we have tried to make the volume
more easily understandable, we h;ve not "talked down," nor have we
omitted the discussion of any concepts or methods that seem to us
important for a b'asic understanding of research methods in social
psychology and soc'iology,
In bringing the book up to date, we have taken account of new de-
velopments in bo~h concepts and methods. For example, since the
first edition was written, there have been new formulations of the con-
cepts of validity and reliability of measuring instruments; the discus,
ix
x PREFACE TO THE REVISED EDITION
sion of these concepts has taken account of these new formulations. At

the same time, new measuring techniques have come into general use
-for example, the semantic differential and the Q-sort-and brief ex-
planations of these techniques have been added. In addition, we have,
of course, included illustrations from research carried out since the
earlier writing.
One other feature of the revision should be mentioned: to the il-
lustrations used in the first edition, which were taken largely from the
area of intergroup relations and prejudice, we have added examples
from research dealing with other social phenomena, in order to make
more clear the applicability of the discussion to a whole range of prob-
lems of social relations.
As with the first edition, this revision has been very much a group
enterprise. It has again been carried out under the auspices of the So-
ciety for the Psychological Study o,f Social Issues. Major responsibility
for the revision has been carried by Claire Selltiz, in close cooperation
with the authors of the earlier edition. Again, we owe thanks to many
colleagues for their help. Isidor Chein, ary Evans Collins, John
Harding, Robert R. Holt, Irwin Katz, Robert S. Lee, Harold M.
Proshansky, Lillian C. Robbins, and M. Brewster Smith have each
rC?d and commented on one or more chapters. There is not space to
acknowledge in detail the contributions of each 'of them. We want,
however, to express especial gratitude to Drs. Chein, Harding, and
Proshansky, and to Mrs. Robbins, for their many helpful suggestions.
Mrs. Robbins and Drs. Chein and P,roshansky-the two latter in their
capacity as the review committee for S.P.S,S.I ....:....have read the entire
manuscript. Dr. Chein drafted parts of Chapters 4, 6, 8, and 10, and
wrote Appendix B; Dr. Harding drafted parts of Chapters 1, 2,6, an9
10; and Dr. Proshansky drafted parts of Chapters 7 and 8. Arthur Korn-
hauser, Donald V. McGranahan, 'Paul B. Sheatsley, William Foote
Whyte, and Alvin Zander kindly gave us permission to incorporate in
this volume material they had contributed to Volume Two 6f the
earlier edition. In addition to the help of these professional colleagues,
we had the assistance of a number of undergraduate students, who
read several chapters and not only pointed out passages that were
ambiguous or unnecessarily complicated, but made many helpful sug-
gestions. They were: Patricia Buchalter, Marilyn Charney, Jane Wiener
PREFACE TO THE REVISED EDITION xi
Einhorn, Ira Latto, and Stephen Silverman, of the Washington Square
College of New York University, and Alan Guskin and Naomi Litt,
of Brooklyn Co1lege. Mary Insinna and Gloria Greaves have typed and
re-typed the various drafts with patience and care. To all of these col-
laborators we express our thanks.
February 1959
CLAIRE SELLTIZ
MARIE JAHODA
MORTON DEUTSCH
STUART W. COOK
CONTENTS
1. The Research Process 1

The Importance of Knowing How Research Is Done, 5
Major Steps in Research, 8
An Illustration, 9
of
Organization the Text, 23
2. Selection and Formulation of a Research Problem 25

Selecting a Topic for Research, 26
Formulating the Research Problem, 30
Formulating Hypotheses, 35
Defining Concepts, 41
EstablisQing Working Definitions, 42
Relating the Findings to Other Knowledge,#
Summary, 47
3. Research Design: I
Exploratory and Descriptive Studies 49
Formulative or Exploratory Studies, 51
Descriptive:Studies,65
Summary, 78
4. Research Design: II
Studies Testing Causal Hypotheses 79
The Logic Qf Testing Hypotheses about Causal Relation-
ships,80i
Causal Inference from Experiments, 91
xiii
xiv CONTENTS
Causal Inference from Other Study Designs, 127

Summary, 142
5. Some General Problems of Measurement 145

Variations in Scores on Measuring Instruments, 149
The Validity of Measurements, 154
The Reliability of Measurements, 166
Scales of Measurement, 186
Summary, 198
6. Data Collection: I
Observational Methods 199
Unstructured Observation, 207
Structured Observation, 221
7. Data Coliection: II
Questionnaires and Interviews 235
Comparison of Interview and Questionnaire,238
Question Content, 243
Types of Interviews and Questionnaires, 255
The Sociometric Method, 268, .
Visual Aids il). Interviewing, 272
A Concluding Note, 276
8. Data Collection: III

Projective and Other Indirect Methods 279
Projective Methods, 280
Structured Disguised Tests of Social Attitudes, 299
Substitute Measures, 310
A Note on Validation. ~.ll
CONTENTS XV
9. The Use of Available Data as Source Material 315
Statistical Records, 316
Personal Documents, 323
Mass Communications, 330
Summary, 342
10. Placing Individuals on Scales 343

Rating Scales, 345
Questionnaires that Form Scales, 357
Some Modifications of Scaling Techniques, 378
A Concluding Note, 383
11. Analysis and Interpretation 385

The Influence of Anticipated Analysis and Interpretation
on Previous Research Steps, 386
The Establishment of Categories, 391
Coding: The Categorization of Data, 401
Tabulation,406 v
Statistical Analysis of Data, 409
Inferring Causal Relations, 422 /
The Use of Nonquantified Data in Analysis and Inter-
pretation, 432
12. The Research Report 441

What the Report Should Contain, 443
Modifications for Shorter Reports, 449
The Style of the Report, 451
13. The Applic,ation of Social Research 455

Concern with Application During the Research, 457
Presentation
I,
of Action-Oriented Research, 466
Extending the Area of Application, 472
Research:md SoCial Policy, 474
XVI CONTENTS
14. Research and Theory 479

The Function of Theory, 480
Theory as a Basis for Research, 487
The Contribution of Research to the Development of
Theory, 492
Interrelation of Theory and Research-A Summary, 498
APPENDICES
A. Estimating the Time and Personnel Needed for a

Study 502
B. An Introduction to Sampling 509
Some Basic Definitions and Concepts,_ 509
Nonprobability Sampling, 576
Probability Sampling, 521
Combinations of Probability and Nonprobability Sam-
pIing, 535
Special Applications of Nonprobability Sampling, 537
C. Questionnaire Construction and Interview Procedure 546

Outline of Procedures in Questionnaire Construction, 546
Guide for Questionnaire Construction, 552
The Art of Interviewing, 574
Bibliography 589
Index 607
RESEARCH METHODS IN SOCIAL RELATIONS
1
THE RESEARCH PROCESS
The Importance of Knowing How Research Is Done'
Major Steps in Research

/
An Illustration
Organization of the Text
I: PROPERT~ OF I
~ the Kansas S.ate Universil)' oj ~
i
I Agriculture & Applzed Science I
I TC M India. I
-
It cannot be that axioms established by argumentation can
suffice for the discovery of new works, since the subtlety of
nature is greater many times over than the subtlety of
argument. FRANCIS BACON
T HE PURPOSE OF RESEARCH is to discover answers to questions

through the application of scientific procedures. These pro-
cedures have been developed in order to increase the likelihood that the
information gathered will be relevant to the question asked and will be .
reliable and unbiased. To be sure, there is no guarantee that any given
research undertaking actually will produce r<:;levant, reliable, and un-
biased illformation.: But scientific research procedures are more likely
to do so than any other method known to man.
Research always starts from a question or a problem of some sort:
Why is the sun visible for more hours each day in summer than in
. winter? Why do apples regularly fall to the ground instead of floating
off into space? How does it happen that a book appears to be about
the same size whether it is fifteen feet away Or two feet away, even
though the image on our retina is m4ch smaller when it is farther away?
Is the desire ~o own private property inherent in "human nature';?
Are the efforts of a settlement house to reduce juvenile delinquency in
its neighborhood having the desired effect? Will sponsorship of a quiz
program on television increase the sales of a ,given brand ,of cigarettes
more than sponsorship of a telecast of a baseba11 game? Is it true that
property values fall when a Negro family moves into a formerly alI-
white neighborhood? Will training foremen in "human relations'~ lead
to a reduction in absenteeism on the part of workers?
The nature of the question varies. Starting from observation of
an event or a series of events, one may ask whether events of this sort
always follow the same pattern, or whether there are circumstances in
2
THE RESEARCH PROCESS 3
which the outcome may be different. Or one may seek an explanation
of the process by which certain conditions lead to certain outcomes.
Questions may take the form "What happens when ...?" or "What
would happen if . . .?" or "Which is the more efficient way of ac-
complishing ...?" They may be concerned with filling in a gap in
knowledge, or with testing a hypothesis, or with checking whether
some proposition which is generally believed is in fact tenable.
In order to be answerable by research, questions must have one
characteristic in common: They must be such that observation or
experimentation in the natural world (including, in the case of the
social sciences, the b~havior of human beings) can provide the needed
information. Many questions of choice or decision cannot be answered
on the basis of information alone, since they involve values as well
as information. "Should the government establish a compulsory system
of health insurance?" is a question of this type. The answer depends
not only on factual information, such as the incidence of disease, the
cost of medical care, etc.; it also involves values about individual initia-
tive, the "welfare state," etc. However, it is often possible to transform
what seems to be a question of value into a question of fact. For ex-
ample, opposition to compulsory health insurance may be based on
values having to do with the freedo~ of patients to choose their
doctors, the freedom of doctors to practice without bureaucratic in-
terference, etc. But whether health insurance does or does not limit
the freedom of patients to choose their doctors, does or does not in-
troduce bureaucratic controls over medical pra.ctice, are questions of
fact, and therefore open to answer by research. Here, however,
further values have been specified-namely, freedom to choose one's
doctor and freedom from bureaucratic control-and the question of
the desirability of a given course of action becomes transformed
into whether or not it is likely to lead to the valued outcomes. This
latter question c~n be answered by research. But the further values
imply other questions-for example, "Is the freedom of patients to
choose their doctbrs so important that it outweighs all other considera-
tions?" This question, in turn, can be answered by research only if some
other value is spe,~ified, so that the importance of freedom of choice in
reaching this other goal can be determined.
There are other questions which could be answe~ed on the basis
4 THE RESEARCH PROCESS
of information alone but cannot be answered by research at the present

time because adequate procedures for gathering the relevant informa-
tion have not yet been devised. Issues calling for universally applicable
scales of psychological measurement are a case in point.
Many questions, however, can be approached by scientific
methods at the present time. This does not mean that research will
always emerge with an answer, let alone a definitive answer. Research
is oriented toward seeking answers; it mayor may not find them.
Characteristically, modern science, and especially social science, is an
unfinished process. To quote Jaspers (1950) : "\Vhereas ancient science
had the appearance of something completed, to which the notion of
progress was not essential, modern science progresses into the infinite."
More often than not, social research results in the raising of new ques-
tions or the reformulation of old ones.
The reasons for asking research questio.ns are of two general kinds:
inteIIectual, based on the desire to know or understand for the satis-
faction of knowing or understanding; and practical, based on the desire
to know for the sake of being able to do something better or more
efficiently. The'investigations to which these two types of question
lead-sometimes labeled "pure" (or "basic") and "applied" research-
are at times discussed as if they were so~ehow opposed or mutually
exclusive, and frequently as if one were better than the other. Such an
approach is misleading. Historically, the scientific enterprise has been
concerned both with knowledge for its own sake and with knowledge
'for what it can contribute to practical concerns. This double emphasis
is perhaps especially appropriate in ,the case of social science. On the
one hand, its responsibility as a science is to develop a body of principles
that make possible the understanding and prediction of the whole
range of human interactions. On the .other hand, because of its
social orientation, it is increasingly being looked to for practical
guidance in solving immediate problems of human relations. Even, this
statement-HOn the one hand ... on the other"-suggests that the
two aspects are distinct. But in the long run, neither goal can be fully
realized without the other. Only as 'general principles are developed
can social science offer sound guidance for immediate action, and only
to the extent that it can make predictions about the results of action
in specific situations does it justify its claim of 1?roviding an adequate
THE RESEARCH PROCESS
systematic body of knowledge about social interactions. Moreover,

the sta~ting point of a study does not necessarily determine what the
nature of its contribution will be. Both in the social sciences and in
the physical sciences, research on practical problems may lead to the
discovery of basic principles, and "basic" research often yields knowl-
edge that has immediate practical usefulness. And whether the motivat-
ing purpose of a given investigation is primarily intellectual or primarily
practical, the requirements of sound research procedure are essentially
the same.
The general scientific enterprise, then, must be concerned both
with the development of general principles and with practical problems
in specific.situations. This does not mean, however, that every study
must have both goals. An investigation may be concerned with a
practical problem or with "pure" research, or with some combination
of these interests. Not only may any given piece of research quite
properly be concerned with only one aspect or the other; any given
investigator may choose to concentrate more or less exclusively on one'
or the other aspect. The condition that is required for the healthy
development of science is that each aspect be adequately represented
in the collective research enterprise.
Throughout this book, we have tried to keep both emphases in
mind. This means that, in addition to concerning ourselves with the
basic requirements of good research, we have devoted considerable
attention to ways of making research relevant to the .solution of social
problems and to increasing the likelihood that its findings will actually
be used.
The Importance of Knowing How Research Is Done
, Even though "research cannot provide any Ena1 answers to the

questions with which it is concerned, there has been a constant effort
to devise procedures that wiII increase the probable accuracy of research
answers. Why is Jt important to be familiar with these procedures?
For the student ~ho is preparing for a career of carrying out research
in social psychology or sociology, the answer is obvious: Research
techniques are the tools of his trade. He needs not only to develop
skill in using them but also to understand the logic behind them.
But it is not only the student who intends to carry out research
who needs to know about research methods. The positions for which
social science students are likely to be preparing themselves-teaching.
administration in government or business, community consultation,
social work-increasingly can for the ability to evaluate and to use
research results~ to judge whether a study has been carried out in such
a way that one can have reasonable confidence in its findings and
whether its findings are applicable to the specific situation at hand.,
Even if one does not expect to make specific use of research find-
ings in his job, in our scientific age all of us are in many ways "con-
~umers" of research results. To use them intelligently, we need to be
'lble to judge the adequacy of the methods by which they have been
obtained. As a student, for example, you will find that many of t~e
"facts" presented in your courses rest on the results of research. But
you may discover that the "facts" reported by one study are quite dif-
ferent from those produced by another study. on the same point~ One
investigator, for example, may report that children who are weaned'
early grow up to be more independent and better adjusted than those
who are nursed for a longer time; another investigator turns up with'
just the opposite finding. Or several studies may conclude that wheQ
Negroes and whites live near each other, each group is likely to become
more favorably inclined toward the other; but other studies may
conclude that interracial hostility is likely to be especially intense in
neighborhoods where Negroes and whites live in close proximIty. In
order to be able to make a tentative judgment about which conclusion
merits more confidence, you need to be able to judge the adequacy of
the studies. Later sections of this book will .consider the criteria ,of
good research in detail. Here we may suggest simply that you wi1J
want to ask such questions as: How do the' investigators define their
terms? Are they really both talking about the same things, or have
they used the same words for different phenomena? Was the evidence
they gathered re1evant to the problem? Were there any obvious sources
of bias in the way the data were gathered? Were there different con~
ditions in the studies that might account for the difference in findings?
Even in the course of daily living, the average citizen increasingly
THE RESEARCH PROCESS 7
needs to be able to evaluate research in order to make intelligent
decisions. This is perhaps most clear at the present time with respect
to medical research and decisions based on it: Should I have my child
inoculated with "flu" vaccine? Should I vote for fluoridation of the
local water supply? Should I stop smoking in order to lessen the risk
of getting lung cancer?1 With the rapid increase in social science
research, it seems likely that the average citizen will increasingly be
presented with social science findings. At the present time, he has
relatively little occasion to evaluate these findings as a basis for his own
actions. But the person who knows how research is carried out is better
able to judge the probable accuracy of opinion polls or election predic-
tions and to view with appropriate skepticism the claims that "9
doctors out of 10 approve ..."
Y,' Besides all these practical advantages of familiarity with research
methods, there is the satisfaction of acquiring a new intellectual tool.

And it is a tool with much broader uses than the specific purposes for
which it was devised. It can become a way of looking at the world, of
judging everyday experience~ The person who really understands the
basic elements of research method is in a position to ask, with respect
to every statement he reads or hears: What is the basis for that view?
Is it supported by evidence? Under what conditions is it likely to hold
true? Of course, he will not necessarily questioii' all statements in this
way. As has already been pointed out, not all matters are appropriately
considered in this way. If a friend admires a beautiful sunset, or ex-
presses a preference for spending his vacation in the mountains, it is
'irrelevant to question the factual basis or the "objective truth" of his
opinion. It may often be inappropriate to change the tone of a social
situation by demanding evidence for a statement lightly made. But if
one reads that public opinion, as reflected in mail to members of
Congress, is opposed to foreign aid, or that a.1egislator has proposed
I
1 Any individual',s decisions will be determined not nnly by the facts as dis-
covered by research but also by his values, which determine the import of th(
facts for him. With respect to health matters, since there is nearly universal agree
ment on the desirability of good health, the importance of values as well as facl
in determining action is often overlooked. However, the importance of values in
social issues is indickted by considering the question: "Should there be racial
integration?" The fact that integration is conductive to friendly association be-
tween Negroes and whites may be taken by some as a basis for supporting segrega-
tion and by others as a basis for supporting integration.
curing juvenile delinquency by fining the parents of delinquents; or if

one hears a neighbor say that he will not sell his home to a Jewish
family because it would spoil the neighbo'rhood-in such situations
one may well want to raise, at least in his own mind, such questions as:
How do you know that? What are the facts on which you base your
conclusion? Is the interpretation warranted by the facts?
.
Maior Steps in Research
The obiect of this book is to describe in detail the procedures neces-
sary to discover answers to questions ,through research. But since
concern with detail often obscures perception of the whole, it is well,
before embarking on the examination of specifi.c procedures, to point
out some over-all aspects of the research process. , .
'The research process cOIisists of a number of closely related activi-
ties that overlap continuously rather than following a strictly prescribed
sequence. So interdependent are these acti0ties that the first step of
a research project largely determines the nature of the last. If sub-
sequent procedures have not been taken into account in the early
stages, serious difficulties may arise and prevent the completion of a
study. Frequently these difficulties cannot be remedied at the time
they become apparent because they are rooted in the earlier procedures.
They can be avoided only by keeping in mind, at each step of the re-
search process, the requirements of subsequent steps.
To be sure, as research proceeds from the conception of a. theme
for a study through the gathering of data to the production of a report
and the application of the findings, the focus of attention will ne~es
sarily shift from one activity to the next. This shift reflects a difference
in emphasis, however, rather than' an exclusive concentration on, one
step. A mechanically consecutive sequence of procedures, in which one
research step is entirely completed before the next is begun, is rarely,
if ever, the experience of social scientists.
The usual pattern of reporting research creates an oversimplified
expectation of what is involved in doing research. Customarily, a report
on completed research, when it appears as an article in a technical
journal, resembles, with minor modifications, t~ following model:
MAJOR STEPS IN RESEARCH 9
1. A statement of purpose is made in the form of formulating the
problem;
2. A description of the study design is given;
3. The methods of data collection are specified;
4. The results are presented;
5. Frequently, there follows a section on conclusions and inter-
pretation.
Whatever the individual variations from this model, published
research strongly suggests the existence of a prescribed sequence of
procedures, each step presupposing the completion of the preceding
one. Although this model is entirely justified in the interest of economy
of scientific reporting, it must not be mistaken for a model of the
research process, which differs from it in two respects: (1) The re-
search process almost never follows the neat sequential pattern of
activities suggested in the organization of research reports; and (2)_
the process involves many additional activities which are rarely men-
tioned in published studies.
Some of these additional activities are related to the scientific re-
quirements of the study; others to its practical demands. The ap-
parently simple reporting of the methods of data collection, for ex-
ample, summarizes decisions about the kinds of data needed and the
most efficient way of collecting them, and the activities carried out in
the development and pretesting of the data-collection instruments. In
addition to these steps, related to the scientific requirements of the
'study, there are other, more "practical," demands: the budget must be
planned; funds must be obtained and administered; personnel must
be allocated and, in some cases, specially trained; the setting within
which the data are, to be collected must be explored and the coopera-
tion of the, people 'in it must be gained; etc. In addition, if the study
is one designed to solve an immediate, practical problem, the antici-
pated application of the findings must be considered from the outset.
An Illustration
I,
The manner in which each step influences, and is influenced by,
others is perhaps best demonstrated by a brief case history of a research
project. As an illustration, we shall use a study of interracial housing

projects carried out by Deutsch and Collins (1951). To a considerable
extent we shall use the authors' own words, as they appear in the report
of the study. But since, like all reports, this one suggests that the study
was carried out in a neat series of separate steps, we shall intersperse
comments about activities carried on in the course of the study which
are not mentioned in the formal report. The account of this study will
give an over-all view of the research process; each of the steps within
the process will be discussed in detail in later chapters.
The investigators were members of a research group that had as
a primary concern the study of relations between members of different
racial and religious groups within the United States. As a further
specification of its general area of interest, the group had chosen to
concentrate on the study of situations in which members of different
groups are in face-to-face contact. Clearly, these decisions mapped o~ut
a very general area of interest; the focus had to be specified much more
sharply before a study could be undertaken.
In this case, the next step was somewhat atypical; it was the
selection of a general setting for the research, even before the problem
to be studied had been identified. This setting was to be large-scale
interracial housing. Since no member of the research group had any
real familiarity with such housing, the group had little basis for select-
ing a specific problem to be studied within that setting. Accordingly,
they drew up a proposal which called for research to be carried out
in two stages: (1) six months to be devoted to an exploratory study,
consisting of interviews-with housing experts, 1m. the basis of which
a specific question would be selected for more systematic study; (2) a
year to be spent in a systematic study of the research question selected
on the basis of the exploratory study. A foundation interested in' re-
search in the field of intergroup relations granted them the necessary
funds.
One member of the research team took primary responsibility for
the exploratory study. Preliminary work involved reading material
about public and private interracial housing, compiling a list of hous-
ing experts whose experience and insights might make a valuable
contribution, and arranging to interview them. In conferences with
the director of the research group and with other colleagues, plans for
AN ILLUSTRATION 11
these interviews were drawn up. Next, the investigator visited housing
projects and housing officials in the eastern, midwestern, and far-
western United States. Forty-two interviews, with managers of housing
projects, race relations officials of public housing agencies, and other
persons with relevant experience in housing, were carried out. The
interviews lasted from two to eight hours. They covered the respond-
ents' views about factors important in influencing relations between
Negro arid white tenants, problems on which research was especially
needed, and suggestions about the feasibility of research on various
problems.
These interviews strengthened the belief that the housing setting
would be a useful one for study of the effects of personal association
between members of different racial groups, pointed to the choice of
public housing because of the scarcity of private interracial housing
projects, and suggested a number of research questions, one of which
the investigators selected for the systematic study. They explain both
their choice of setting and their choice of a specific question on grounds
of implications for social action, theoretical interest, and opportunities
for research:
Public housing has existed for more than ten years. It has
had a chance to develop standard patterns and variations with
respect to racial occupancy. The common pattern is complete
segregation-Negroes and whites live in separate housing proj-
ects-but there are important exceptions. These exceptions
and the variations among them, in effect, provide a natural social
i experiment which permits those engaged in carefully controlled
social research to gather valuable information about condi-
tions which makeI for wholesome race relations. Such research
can, by providing scientifically grounded knowledge in the place
of current uncertainties, aid policy makers in their efforts to
live up to the ori'nciples of the democratic ethos in their official
functioning.
Not only may such research be socially useful ... but it

might also have ~heoretical significance. Most of the studies of
attempts to change prejudice have, by and large, been limited
to influences which were relatively minor in relation to other
influences in the subject's social milieu . . . . there is reason to

believe that the housing community . . . provides one of the
few opportunities for interracial contact of a sufficiently intimate
and extended nature to result in large attitudinal change. In light
of the crudity of most attitude measurement techniques, changes
of relatively large magnitude are necessary to make possible iden-
tification and measurement of significant theoretical relation-
ships.
With a census of problems and issues in the field of inter-

racial housing before us, our task was to select a problem for
more definitive research. Many studies were suggested by our
survey [of housing officials}: the effects of different policy de-
cisions, the effects of different management procedures, the
effects of different kinds of tenant activities. An of these ane!
many more would be useful. From these possibilities, we selectfd
for study the impact of different occupancy patterns: the in-
tegrated interracial pattern (families are assigned to apartments
without <tonsideration of race) and the segregated bi-racial pat-
tern (Negro and white families live in the same project but are
assigned to different buildings or to different parts of the proj-
ect).
We make no claim that the occupancy pattern is the only
important influence on racial relations in projects that house
both white and Negro families. Quite on the contrary, our sur
vey indicated that the state of racial relations in a project Would
be affected by many factors . I :the neighborhood in which
the project is located, the racial composition of the tenants, the?
attitudes of the management staff, project facilities, etc. The
effects of the occupancy pattern would, of necessity, be colored
by the influence of these other factors. . , .
Yet our decision to investigate the .effects of the occupancy
pattern, as our first systematic study in this area, was not a mat-
ter of whim. We had several reasons for the choice. From. our
survey of housing officials and from our theoretical expecta-
tions . . . it was apparent that the occupancy pattern would
very probably"prove to be one of the most crucial influences on
race relations in housing projects. . . .
I t is not often that social research executed withouCThe
AN ILLUSTRATION 13
instigation of an administrator will be useful in affecting admin-
istrative decisions. Nevertheless, a further important reason for
our interest in Ithe occupancy pattern is our belief that, under
the present circumstances, something can be done about it and
that research can offer guidance in the doing. Unlike many of
the factors which affect the state of race relations in a housing
project, the occupancy pattern is directly determined by an ad-
ministrative decision, a decision which is responsive primarily
to the "social climate" rather than to economic and physical
limitations . . . . With the political atmosphere such as it is,
with the alternative occupancy policies each having its pros and
cons with respect to political feasibility, these decisions are partic-
ularly apt tobe influenced by knowledge of the consequences of
the different pat,terns for tenant relations.
Actually, it had not seemed feasible within the exploratory study
to examine in detail the practical possibilities for research on each of
the topics suggested; the investigators therefore did some further
"scouting" before they settled on the effects of occupancy patterns as
the focus of the second stage of the research. The purpose of this
scouting was to determine whether it would be possible to find ap
propriate projeots in which to carry out research on this topic and to
secure permission from housing officials to do so. Preliminary inquiries
suggested that it would be feasible; accordingly, the decision was made
to study the effects of occupancy patterns.
For much of the next fourteen months, both investigators worked
on the study full time. Their efforts were supplemented from time to
time by conferences with a consultant and with colleagues, by a staff
of interviewers, a group of coders, an organization which specialized
in the machine processing of data, a secretary. But before these more
specialized

services:I were needed, it was necessary to formulate the
research question more precisely, and to consider its relation to ex
isting knowledge. '
Our statement of the problem . . . needed much addi-
tional formulation before it could serve as a guide to research.
In effect, we ~ad to develop hypotheses about the possible
effects of the 9ccupancy pattern from our knowledge of the
essential differences between the two types of project we were
studying and from a knowledge of basic socio-psychological

principles. Otherwise research efforts would be dissipated on the
investigation of factors not likely to be related to the occupancy
pattern.
. . . In the development of hypotheses, it is always simpler
to deal with imaginary, idealized phenome'na rather than with
events as they occur in the real wor1d. So, for the sake of conven-
ience, let us conjure up two projects exactly alike except for
their occupancy pattern. In one project, Negro and white fam-
ilies are assigned to apartments without consideration of race;
in the other project, though Negro and white families live in
the same project, they are assigned to different buildings or to
different parts of the project. . . .
From the point of view of race relations, what . . . are
the essential differences between the two projects? . . . it
seems to us that the two types of projects differ mainly w~th
respect to (1) the physical and functional proximity of Negro
and white families, (2) the social norms regarding racial rela-
tions implicit in the policy decision of the occupancy pattern
by an official public authority, and (5) the relationship of the
project to the broader community.
These are the differences, but what are their effectsi. The
answer to this question requires a further specific~tjon of interest.
Effects upon what? Our original interest directs us to inquire
about the effects of the occupa~cy pattern upon (1) social re-
lations across racial lines, (2) the social standards for behavior
with people of the other race, (3) the general pattern of social
relations in the project, and 1 (4) interracial attitudes.
As it became clear what the research question was to be and' ~h~
type of evidence would be relevant tp the answer, the investigators
began to consider how to col1ect this evidence most economically and
with least chance of being led to an incorrect conclusion-i.e., they
began to develop their research design. The essential feature of the
design was a comparison of the responses of residents in the two types
of project. The development of a satisfactory research design! is likely
to be particularly difficult when the research is to be carried out in a
real-life setting. In the case of this study, it required that the two types
of project be alike in all relevant _respects except occupancy pattern,
AN ILLUSTRATION 15
and that the tenants in the two projects should have comparable initial
attitudes. I t was not easy to satisfy these conditions.
In the abstract, the research design called for by the hypoth-
eses is relatively uncomplicated. It simply requires a compar-
ative study of a number of segregated and integrated interracial
housing projects which are equated in all relevant respects lother
than the occupancy pattern. Practically speaking, however, the
phrase "equated in all relevant respects" introduces enormous
complications and difficulties. As any housing administrator
will point out, "No two housing projects are,alike."
We designed our study to overcome as many as possible of

the difficulties which would otherwise distort our findings. First
of all, we carefully selected the segregated and integrated inter-
racial projects we were to study so they were as equivalent as pos-
sible in all relevant respects other than the occupancy pattern.
Second, we decided not to limit our study to one project of each
type; we stretched our funds so as to study two of each kind. And
third, in our method of investigation we carefully collected data
about factors other than the occupancy pattern, to determine
whether our results could be explained in terms of these other
/
factors.
Compromises with the ideal of projects "equated in all other rele-
vant respects" had to be made. In the course of the preliminary scout-
ing, it had been learned that the number of integrated projects was
limited. At the time of the study, there were some fifty-odd cities
throughout the Unityd States that had public housingprojects officially
described as integrated; however, in most of these projects the great
majority (over 90 pet cent) of the families were of one'race. It seemed
reasonable to SUppOSy that the effects of occupancy pattern would show
up more clearly in projects where the numbers of white and Negro fam-
ilies were more eveniy balanced. But such projects existed in less than
ten cities. While the research question was being sharpened and the
design of the studYI,worked out, the search for appropriate projects
continued. It was finally decided that New York City presented the
best setting for selection of integrated projects; neighboring Newark
provided matching segregated bi-racial projects. But interviewing

could not be conducted in the projects without the consent of the
housing officials. Although officials in both cities were interested and
cooperative, several conferences were needed to complete arrangements
with them.
The fact that the integrated and the segregated projects were in
different cities introduced a problem; so too did certain characteristics
I)f the projects. New York and Newark are different. Since the research
funds did not provide for studying the attitudes of a sample of citizens
of each city not living in public housing projects, it was not possible
. to work out a study.design that would make it possible to 'judge to
what extent observed differences between tenants in the two types of
project might reflect differences in atmosphere in the two cities rather
than the impact of occupancy pattern. Further, it developed that the
projects that could best be matched On other grounds all had a ~high
proportion of Negro tenants-from 40 to 70 per cent. This meant that
the findings could properly be generalized only to projects with a sim-
ilarly high proportion of Negroes. Finally,_the projects in Newark were
of a type known as area-segregated, with Negro and white families living
in different sections of the project. Although this had the advantage
of giving a dear-cut pattern of segregation, it meant that the'\ndings
might not hold for projects with less extreme patterns of segregation,
such as those in which separate buildings occupied by Negro families
and by white families are scattered throughout the project.
During the same period that the projects were being selected, the
.staff considered the selecti~n of ~ubjects within the projects and the
method of collecting data. Here again there were limitations and com-
promises:
From our preliminary survey and the theoretical analy$is,
it became clear that the focus of the study would be the tenants
themselves. Since there was, by and large, little evidence of much
organized or observable group activity, it was evident that the
main source of our information must come through interviewing.
Because 9f the limited number of interviews that our re-

[.OUIces made possible, it seemed advisable to concenb;ate on
some segment of the tenant population. W~ decided to ~ollect
AN ILLUSTRATION 17
our data primarily from white housewives. We made this choice
mainly on the ground that the horne is largely the domain of
the woman. She spends more time in it than anyone else; she is,
by and large, the initiator of activities and contacts that develop
directly out of the horne. Whether or not she "wears the pants
in the family," she is the key person in activities centered in the
place of residence. .
It was not financially feasible to interview both Negro and
white housewives in equal proportion. We decided to interview
more white housewives as a result of our conviction that preju-
diced interracial attitudes are more socially crucial among whites
than among Negroes. Segregation and discrimination are, after
all, enforced by the white, and not by the Negro segment of the
population.
A more serious problem was that of determining whether the
white housewives in the two types of project had had similar attitudes
toward Negroes before they moved into the projects. Since all the
projects had been occupied for several years, it was impossible to get
measures of attitudes the tenants had held before moving in. Instead,
the investigators drew on their knowledge of other studies, which had
shown that such cha)acteristics as religion, political attitudes, educa-
tion, and previous experiences with Negroes are frequently related to
/
attitudes toward Negroes. Accordingly, they included in the interview
schedule questions on these points. If it turned out that the women
in the integrated and in the segregated projects were similar in these
respects, there would be some reassurance-though by no means cer-
tainty-that they were similar in their initial attitudes toward Negroes.
Unfortunately, it developed that there were differences between the
two groups in religion, political attitudes, and education; thus these
factors had to be Jtaken into account throughout the analysis.
Decisions such as this, about what information was needed, led
I
naturally to consideration of the interview schedule, the instrument
by which th~ data were to be collected. The process of developing
questions was simplified by the fact that much research on attitudes
toward Negroes h~d already been carried out. The interview as finally
developed covered five major areas: the attitudes of the housewives
toward living in .the project, attitudes toward Negroes, the amount
and intimacy of contact with other women in the project, the social
supports for attitudes, and the characteristics of the housewives.
While the interview schedule was being developed, the samples
of women to be interviewed were being drawn from the lists of tenants
in the projects. During this same period, one of the staff members
began taking steps to recruit the number of skilled interviewers who
would be needed to collect the data within a reasonable time. The inter-
viewers were graduate students in social work and psychology. As soon
!1S the interview schedule was in a form that seemed reasonably satis-
factory, a "pilot test" was carried out; two or three of the most expe-
rienced interviewers, and the investigators themselves, carried. out in-
terviews with a small number of white housewives in other housing
projects similar to those that had been selected for the study. As was
expected, these pilot interviews pointed up questions that were not
clear, those that needed especially careful handling to avoid antagoniz-
ing the respondents, those that did not seem to elicit the inforniation
they were intended to get. Changes were made in the interview sched-
ule to overcome these difficulties. After another set of pilot interviews
had been checked, aU interviewers were trained in use of the sch~ule.
Each interviewer spent approximately twelve,hours in training sessiops
and conducted two practice interviews with residents of projects not
included in the study.
Finally, the actual interviewing got under way. Five hundred in-
terviews were conducted: four hundred with white housewives, one
hundred with Negro housewives. The interviews lasted, on the average,
from an hour to an hour and a half. There were nineteen intervieo/ers,
and all the interviews were completed within a month. During the
interviewing, the investigators spent time in each of the projects, super-
vising the assignments and inspecting the interviews as they were, cqm-'
pleted.
Once data collection had taken place according to the specified
research design, a number of irreversible decisions had been made
which largely determined the next step-analysis and interpretation. In
this study, the plan of data collection made it possible to compare the
white housewives in the area-segregated and the integrated projects
in terms of: (1) their extent of association with Negroes, (2) their
perception of the social norms concerning association with. Negroes,
AN ILLUSTRATION 19
(3) the relation between perceived social norms and extent of associa-
tion, (4) their attitudes toward Negroes living in the project and
Negroes in general, and (5) their attitudes toward living in an inter-
racial project_
The limitations and compromises that have already been pointed
out meant that It was not possible to be entirely sure that the differ-
ences found between the housewives in the two types of project repre-
sented the effect of occupancy pattern rather than general differences
in attitude in the two cities or differences in attitude between the two
groups of white housewives that existed before they ever moved into
the projects_ Nor was it possible to say whether the findings would hold
for projects with smaller proportions of Negroes and for projects where
the pattern of segregation was less marked-for example, in projects
segregated by buildings rather than total sections_
Finally, the pattern of area-segregation in the Newark proiects
meant that, even if it was found that the white housewives living in
integrated projects had more favorable attitudes toward Negroes than
those in the segregated projects, it would be difficult to draw inferences
about the processes that contributed to the difference in attitudes_ In
thinking about the ways in which different occupancy patterns might
be expected to lead to differences in attitudes, the investigators con-
sidered two major factors: (1) the extent to which Negro and white
tenants had occasion to see and to meet each other and thus had
natural opportunities to become acquainted, and (2) the implications
of the integrated and segregated arrangements in terms of social ap-
proval or disapproval at association between white and Negro families.
The pattern of segregation in the Newark projects, with white and
Negro families 1ivi~g in separate sections of the project rather than
in separate buildings scattered throughout the project, made it impos-
sible to disentangle the effects of perceived social standards from those
of simple physical :proximity, since in these area-segregated projects
not only did white and Negro families live relatively far from each
other but the segregated arrangement suggested social disapproval of
interracial associat\on. This was a limitation primarily on the potential
contribution of the study to theoretical knowledge of the dynamics
of attitude change; it did not interfere with the possibility of gathering
evidence on the practical question of the effects on attitudes of living

in an integrated or an area-segregated project.
In order to provide answers to those questions' which could be
answered by a study having this particular design, it was necessary, of
course, to analyze the replies to the interviews. The function of the
analysis was to make possible a comparison of the women in the inte-
grated projects with those in the segregated projects in terms of the
possible effects (or dependent variables) which had been specified in
the formulation of the research problem and in the construction of
the interview schedule.
The first step in the analysis was to derive from the answers to the
interview questions an indicator of each dependent variable with
which the study was concerned, or sometimes, several indicators. For
example, in the general area of social relations within the projects, the
main dependent variable with which the investigators were conct;rned
was intimacy of contact with Negroes on the part of white housewives.
The interview provided three indicators of this variable,: (1) whether
the housewife reported knowing any Negro residents "pretty well"; (2)'
w;~ether she included at least one Negro ~mong the five people in the
project whom she knew best; and' (3) the number of different types
of neighborly contact she reported with Negro women in the project.
In the general area of interracial attitudes, the investigators dis-
tinguished eI'even different variables, such as degree of esteem for
Negroes in the project, degree of esteem for Negroes in general, degree
of friendly feeling toward Negroes in general, etc. For each of these
variables, one or more indicators were derived from the answers to the
interview questions or from ratings by the interviewers.
The second step in the analysis ~s to prepare tables showing the
distribution, on each index, of respondents in each of the four projects.
These tables were then examined and suqjected to tests of statistical
significance to determine whether the differences between housewives
in the segregated projects and housewives in the integrated projects
on a particular variable were so large that they could not be reasonably
interpreted as resulting simply from accidents in the random selection
of subjects. -
Almost a,ll the tables showed large differences between housewives
in the two types of project, and the differences were consistently in the
AN ILLUSTRATION 21
direction of more social contacts with Negroes and more favorable at-
titudes toward Negroes in the integrated projects. Most of the differ-
ences were far too large to be reasonably attributed to sampling fluc-
tuations.
The next step in the analysis was to investigate the distribution of
respondents on each of the background variables, or characteristics of
respondents thought to be related to attitudes toward Negroes. As we
have previously noted, there were substantial differences between re-
spondents in the two types of project in religion, political attitudes,
and education. It was consequently necessary to prepare additional
sets of tables showing, for each index, the scores of Protestant women
in each of the four projects, of Catholic women, and of Jewish women.
From these tables it was possible to determine that the previously
found relationships between occupancy pattern and the dependent
variables remained approximately the same when Protestant, Catholic,
and Jewish women were examined separately-that is, when religion
was "held constant." Similar tables were prepared "holding constant"
educatIon and political attitudes, with similar results.
On the basis of this analysis it was concluded that the differences
in attitudes and social relations with Negroes between housewives in
the two types of project could not be accounted for by the differences
in religion, education, or political views. SimiJar analyses indicated
that these differences could not be accounted for by differences in
initial attitudes toward Negroes (as reported by the respondents them-
selves), or by differences in expectations regarding the occupancy
pattern at the time of moving in. It was not possible to rule out the
possible effect of differences in the "social climate" of New York City
and Nei,vark, Or of other uncontrolled variables that were not tapped
by the interview sphedule; but it was tentatively concluded that the
differences in the pependent variables were due to the differences in
occupancy pattern lin the two types of project.
Interest then turned to the question of how the difference in
l
occupancy pattern1might produce the observed differences in attitudes

toward Negroes. The investigatcrs had included in their interview
schedule a questiop on the ways in which someone living in the project
might get to know; the Negro women in the project. The most common
ways mentioned by housewives in the integrated projects were meeting
the Negro women as neighbors in the building, or on benches outside

the building. In the segregated projects only half as many women
mentioned any ways in which one would be likely to get to know the 0
Negro women, and the contacts that were most frequently mentioned,
such as meeting in stores or in streets around the project, did not
provide natural opportunities for extended conversation.
Another analysis was made comparing the housewives in the in-
tegrated projects who reported that their attitudes toward Negroes
had become more favorable since living in the project with those who
reported no change in their attitudes. This showed, among other
things, that the women whose attitudes had changed tended to be
those who had more intimate contact with the Negro women in the
project and who believed that their white friends in the project ap-
proved of their friendly association with the Negro women. As was
pointed out earlier, the study design did not make possible an evalua-
tion of the relative influence of association with Negroes and of per-
ceived social approval of such association on the process of attitude
change.
Because of this and other uncertainties of interpretation men-
tioned earlier, plans were made almost immediately for another study,
which would build on this one and carry its findings still further (Wil-
ner, Walkley, and Cook, 1955). This new study was carried out in
four cities (all of them outside the New York metropolitan area).
The projects all had a quite small proportion of Negroes, and the seg-
regated projects took the form of scattered Negro and white buildings'
rather than separate areas. These two latter characteristics made it
possible to examine separately tIle influence of physical proximity and
of implied official standards, since so~e white women in the obuilding-
segregated projects lived closer to Negroes than did some white women
in the integrated projects. The findings of this study indicated that
physical proximity was the more important influence.
This illustration can indicate only in barest outline the mi.ttire of
the research process. As would be the case with any illustration, it does
not cover all the possibilities of interrelation and interdependence of
research steps.1_'he pattern of interaction among the various procedures
that constitute a scientific inquiry will, of course, vary from study to
study. The point of this illustration is to show not only that early
AN ILLUSTRATION 23
steps influence subsequent ones-an obvious matter-but also that tho
interaction of each step with others is a major consideration in its
selection, and that subsequent steps often lead to a reconsideration of
preceding ones. Social research is not a deductive process, in which
everything follows from some clearly defined premises; it is a contin-
uous search for truth, in which tentative answers lead to a refinement
of the questions to which they apply and of the procedures by which
they were obtained.
Organization of the Text
This volume describes the major steps in the process of social re-
search. The demands of organization require that these steps be dis-
cussed separately and consecutively, but it should always be kept in
mind that the steps are not so clearly demarcated from one another
as an organized discussion makes them appear to be.
Chapter 2 discusses the problems and considerations arising in
the selection and formulation of a research question. Chapters 3 and
4 deal with research design ,and its function in a scientific inquiry.
Variations in design are discussed, from the relatively
/
unstructured ex-
ploration of a problem to the rigorous testing of hypotheses by means
of controlled experiments.
Chapter 5 presents some general problems of measurement in the
social sciences. It provides a background for the next five chapters.
Chapters 6, 7, and 8 discuss three broad groups of data-collection
methods-observational methods, questionnaires and interviews, and
projective techniq~es. Chapter 9 treats the ,use of data already avail-
able, such as statis~ical records and the content of communications and
personal documents. Chapter 10 discusses techniques for placing in-
dividuals on scalesion the basis of data collected by any of the methods
considered in the preceding chapters.
Chapter 11 deals with the analysis and interpretation of da~
Chapter 12 with !pe writing of a research report; and Chapter 13 with
the application of research. Finally, Chapter 14 discusses the contin-
uousand close interrelationship between empirical research and thooIY,
;lnd points to the significance of theoretical development for the prac-

tical application of social science.
There are three appendices. Appendix A discusses the need for
t.>areful planning of time and personnel requirements, and presents time
hudgets of three studies. Appendices Band C consider certain tech-
nical problems in greater detail than seems appropriate in the body
of the book. Appendix B discusses sampling procedures; Appendix C
discusses questionnaire construction and interview procedure.
SELECTION AND FORMULATIO~1
OF A RESEARCH PROBLEM
Selecting a Topic for Rese~rch

Formulating the Research Problem
Formulating Hypotheses
"
Defining Concepts
,
~stablishing Working Definitions
Relating the Findings to Other Knowledge

Summary
The formulation of a problem is often more essential than its
solution. A. EINSTEIN AND L. INFELD
Selecting aTopic tor Research
T HE RANGE OF POTENTIAL TOPICS for social research is as broad as

the range of social behavior itself. The range of topics that have
actually been selected for inquiry is, of course, much narrower. Yet'
even in one research area-for example, in the relations among groups
of different ethnic or religious origin-investigations have covered a
vast field, including studies '()f how such groups act toward one another
and feel and think about one another; how they differ in their tradi-
tions, beliefs, personalities, culture, and the way they treat their chil-
dren; how they grow to ,be what they are; how they respond to attempts
at ,changing their relation to another group; and so on. These in-
quiries have been conducted in many parts of the world; they have
involved members of many ethnic and religious groups in many walks
of life.
As we noted in Chapter 1, the general topic of a study may be
suggested by some practical concern or by some scientific or intelle~tual
interest. A wide variety of practical concerns may present topics for
research. For example, one may want information as a basis for dedd-
ing whether there is need for some new facility or service. Thus, the
director of a state or city department of health and welfare may want
a survey of all people over 65 to determine whether additional services
are needed for this age group, and, if so, what kind-more e?Ctensive
financial aid, case work services, social and recreational center~, special
housing, hospita)s, etc.
Or, one may want to evaluate the effects Df a settlement house
program on the mental health-of its participants, the effects of a leaf-
26
SELECTING A TOPIC
let campaign in influencing votes, the effects of a movie about Jackie

Robinson on attitudes toward Negroes.
Still another type of practical issue that may lead to research is
one in which information about the probable consequences of various
\,:ourses of action would be helpful in deciding among proposed alter-
natives. Thus, toward the end of World War II, the United States
Army carried out an extensive survey among its men to find out what
system of deciding on priorities for discharge would generally be viewed
as fair.
A slightly different practical concern requires the prediction of
some future course of events in order to plan appropriate action. For
example, in 1945 the United States Treasury requested a study that
would permit a prediotion of the course of war bond redemptions
during the following year, as a guide to policy and action related to
problems of inflation control. A builder who is considering the con-
struction of a development in whioh homes will be sold without regard
to the race of the purchaser may want an estimate of the probable
market and the extent to which it will be affected by the interracial
character of the development.
Scienti5c or inteIIectual interests may suggest an equally wide
range of topics for research. Perhaps the major difference between
topics suggested by practical concerns and thos6 dictated by scientific
interests is that the latter are less likely to involve the study of a specific
situation primarily for the sake of knowledge about that particular
situation. Scientific and intellectual interests are more likely to lead
to general questions, and to a concern with specific situations as ex-
amples of general .classes of phenomena rather than as objects of in-
terest in their own fight.
The investigator who is prompted by scientific curiosity may be
interested in explqring some general subject matter about which rela-
tively little is known. For example, when Piaget began his studies on
children's thinking, this was a relatively unexplored area.
'Or he may be interested in phenomena that have already been
studied to some e:lItent; in this case he is likely to be interested in such
matters as speCifying more exactly the conditions under which the
given phenomenon appears, and how it may be affected by other
28 FORMULATION OF A RESEARCH PROBLEM
factors. For example, let us say that he is interested in the extent to

which the judgments of individuals are influenced by the expres,sed
opinions of their associates. This is a question on which a good deal of
work ha6 been done; it has been quite clearly established that indi-
viduals' judgments are frequently influenced by the expressed judg-
ments of others. The investigator may want to establish differentiations
within this general finding. He may want to ask, for example: Are
judgments on questions to which there is no single factually "correct"
answer more likely or less likely to be influenced by the stated opinions
of others than judgments on questions to which there is an objectively
correct answer? As an example of the first type of question, he might
ask his subjects whether they consider Tennessee Williams or Arthur
Miller a 'better playwright; as an example of the second, he might ask
them to judge the length of lines. His research problem wquld be
answered by comparing the extent to which the subjects' judgmepts
were Influenced by those of other people in the two situations.
Or, if he happens to be working in a field in which there is a highly
developed theoretical system, he may want to test specific predictions
based on that theory. Thus Hunt (1941), interested in experimental
testing of psychoanalytic theory, chose to test the prediction that de-
privation in infancy leads to "miserly" behavior as an adult. He reared
three groups of rats: one was fed normally.during the developmental
period; the second was fed normally in infancy, scantily during the pre-
pubertal period, then normally again; the third was fed scantily during
infancy, but adequately later. In adulthood, all groups were subjected'
to feeding deprivation; the one wlJ.ich had been deprived in infancy
reacted by hoarding food.
Given such a variety of sources that may suggest topics for research,
how does the investigator select a problem to study? Among the ;im-
portant factors determining his choice are his own personal inclina-
tions and value judgments. In Whitehead's words:
Judgments of worth are no part of the texture of physical
science, but they are part of the motive of its production. . . .
There has been conscious selection of the parts of the scientific
fields to be cultivated, and this conscious selection involves
judgments of value. (Quoted from Cantril, Ames, Hastorf, and
Jttelson, 1949}. '
SELECTING A TOPIC 29
The role that values inevitably play in the selection of a subject>
for research is not, however, always recognized. In this connection,
Cook (1949) reports thefollowing incident:
I once described to another psychologist a study in which
our staff was trying to assess the effect of a certain experience
in changing the attitudes of non-Jews toward Jews. This psychol-
ogist, however, did not feel these attitudes should be changed.
My interest in measuring the success of efforts to bring about
ohange, it became apparent, implied to him endorsement of a
goal with which he strongly disagreed. Moreover, it brought up
the possibility that my endorsement would be interpreted by
others to mean endorsement by psychologists in general [italics
supplied].
If the fact that values are involved in the selection of every re-
search topic had been recognized in this instance, the fear that the
values leading to the selection of one study might be assumed to be
values common to psychologists in general might not have arisen. If
it is recognized that every investigator, in his selection of a topic for
research (whether it be racial or religious prejudice or the development
of children's spatial concepts), is expressing his judgment or feeling
about what is important to study, then the great variety of research
undertaken makes it clear that no one study fepresents a common
agreement by social scientists 'on appropriate research emphases.
Social scientists with different values choose different topics for
investigation. The social scientist who knows which of his personal
preferences have entered into the selection of his topic will be better
able to guard against biases they might introduce into his research
than the One who works under the illusion that he is guided by
scientific conside~ations only. Since personal values inevitably in-
fluence the ohoic~ of topic, the only means by which the rationality
of scientific proceflure can be maintained is the awareness of where
and how, they enter.
It would be erroneous, however, to assume that personal values
are or could be the only determinant in selecting a topic for inquiry.
Not only do social conditions under which science is pursued shape
th~,PJeferences o!f investigators in a subtle and often imperceptible
way, but there ar~ also a number of powerful and overt inducements
for pursuing research on one topic rather than another. Different

societies place premiums on work on different research topics. It may
bring more prestige to do research on cancer than to try to find a cure
for the common cold; more research funds may be-available for noncon-
troversial experiments with animals than for the investigation of topics
that may have political repercussions; better-paid positions may be
available for the market researcher than for the educational psychol-
ogist. There are few social scientists who can afford to ignore prestige,
research funds" and personal income. This state of affairs is likely to
result in reducing the effective freedom of choice of research topics
unless a variety of agencies with different interests, views, and values
support research, so that the individual research worker can choose
among them. In the United States at the present time, although tM
number of different sour,ces of research funds is fairly large, a,consider-
able proportion of the money available for research is concentrated
in a few government agencies and a few large foundations. As a result,
these organizations determine to a considerable extent the problems
on which res~rch shall be done. This is true for both the physical and
the social sciences. -
Formulating the Research Problem

The selection of a topic for research does not immediately put the
investigator in a position to start considering what data he will collect,
by what methods, and how he will analyze them. Before he takes these
steps, he needs to formulate a specific problem which can be inv,esti-
gated by scientific procedures) Unfortunately, it happens not. infre-
quently that an investigator attempts to jump immediately from the
selection of a topic to the collection of observations. At best, this
means that he will be faced with the task of formulatin'g a problem
after the data collection; at worst, that he will not produce a'scien-
tific inquiry at all.
\ ,
[For] it is an utterly superficial view, that the truth is ,to be
found by studying the facts. It is superficial because no inquiry
can ever get under way until and unless some difficulty is felt in
a practical or theoretical situation. It is the difficulty, or problem.
FORMULATING THE PROBLEM 31
which guides our search for some order among the facts in terms
of which the difficulty is to be removed (Cohen and Nagel,
1934) .
Scientific inquiry is an undertaking geared to the solution of
problems. tThe first step in the formulation of research is to make the
problem concrete and explicit.
Although the selection of a research topic may be determined by
other than scientific considerations, the formulation of the topic into
a research problem is the first step in a scientific inquiry and, as such,
should be influenced primarily by the requirements of scientific pro-
cedure. However, there is no foolproof rule which will guide the in-
vestigator in formulating significant questions about a given research
area. Here, the training and gifts of the individual are of major im-
portance. As Cohen and Nagel (1934) point out:
The ability to perceive in some brute experience the occa-
sion for a problem, and especially a problem whose solution has
a bearing on the solution of other problems, is not a common
talent among men. . . . It is a mark of scientific genius to be
sensitive- to difficulties where less gifted people pass by un-
troubled by doubt.
Yet it is possible to enumerate some conditions that experience
has shown to be conducive to the formulation of significant problems.
Among these conditions are systematic immersion in the subject mat-
ter through firsthand observation, the study of existing literature, and
discussion with persons who have accumulated much practical expe-
rience in the field of study.! Important as these conditions are in the
formulation of a research project, they contain a danger. In social
science, as elsewhere, habits of thought may interfere with the
discovery of the new and the unexpected unless preliminary firsthand
observation, readi~g, and discussion are conducted in a constantly
critical, curious, an~d
I
imaginative frame of mind.
The first step in formulation is the discovery of a problem in need
I
of solution. Without this step, research cannot proceed, as the follow-
ing episode demonstrates. Every summer an educational institution in
! See Chapter ~ 'for a detailed discussion of study designs that involve these
procedures. The prim"ary value of such exploratory inquiries is that they may lead
to the asking of more, significant questions than would otherwise have been possible.
this country brings together for six weeks about two hundred young
men and women from all walks of life. Every region of the cO"!-lntry, .
every creed, and every race is represented. Some of the young people
are workers or farmers; others are clerks or students. They are selected
for this summer school because of the promise of leadership they have
shown in their unions, social organizations, or colleges. The sponsor-
ing agency aims at giving these young people information about the
world they live in, an experience in living together, and skills to enable
each of them to meet the demands of a leadership role in his own
sphere of life even more effectively than before. The organizers of this
voluntary venture invited a team of social scientists to discuss the
possibility of doing research on the school. The reason for the invita-
tion was one that prompts many agencies to seek help from .social
scientists-the institution hoped to obtain reassurance about the value
of the program and was convinced that science could establish this-
value beyond doubt. The topic of the envisaged research was clearly,
then, the value of the educational venture. The' discussion between
sponsors and social scientists had' the purpose of transforming this
topic, step by step, into a research problem.
As is customary in such discussions the social scientists started
,I
with, the question: What would you like to find out about your enter-
priser'This was followed by the equally customary counter-question:
What can social science tell us about it? The remainder of the session
demonstrated to both parties the difficulties of research formulation.'
After the original impasse had been overcome, the representa-
tives of the institution explained in complete detail their long-term
objectives. They already had considerable knowledge of the. 9'rica1
problems and short-range effects of their ,pioneering effort, but their
own high goals and their educational outlook prevented them from
accepting as a trustworthy yardstick of success the obvious enjoyment
of the experience by the young people. What they sought was some-
thing that would show itself out~de the confines of the educational
setting and throughout the later life of each participant.} The im-
mediate problems of administering the school-recruitment, program,
organization, etc.-were apparently well in hand.
Although the discussion of long-term effects was of considerable
FORMULATING THE PROBLEM 35
feasible to start with this aspect, it provides considerable guidance for
the other steps.
Formulating Hypotheses
A hypothesis is "a proposition, condition, or principle which is
assumed, perhaps without belief, in order to draw out its logical con-
sequences and by this method to test its accord with facts which are
known or may be determined" (Webster's New International Dic-
tionary of the English Language, Second Edition, Unabridged,
1956). The role of hypotheses in scientific research is to suggest ex-
planations for certain facts and to guide in the investigation of others.
The importance of hypotheses in research has been emphasized by
Cohen and Nagel (1934), who state:
We cannot take a single step forward in any inquiry unless
we,begin with a suggested explanation or solution of the difficulty
which originated it. Such tentative explanations are suggested
to us by something in the subject matter and by our previous
knowledge. When they are formulated as propositions, they are
called hypotheses.
The function of a hypothesis is to direct our search for the
order among facts. The suggestions formulated in the hypothesis
may be solutions to the problem. Whether they are; is the task
of the inquiry. No one of the suggestions need necessarily _lead
to our goal. And frequently some of the ,suggestions are incom-
patible with one another, so that they cannot all be solutions to
the same problem.
It seems to us that this is an accurate statement of the nature and
value of hypotheses in scientific investigation, but we believe it to be
too sweeping in its assertion that research cannot begin until a hy-
pothesis has been 'formulated. As we shall argue later, a very important
type of research h~s as its goal the formulation of significant hypotheses
about a particular topic. .
A hypothesisI, m~y assert that something is the case in a given
instance, that a ,particular object, person, situation, or event has a
certain characteristic. For example, Freud's book Moses and Mon-
otheism begins with the hypothesis that Moses was actually an

Egyptian, not a Jew. Or a hypothesis may have to do with the
frequency of occurrences or of association among variables. It may
state that something occurs a certain proportion of the time, or that
something tends to be accompanied by something else, or that some-
thing is usually greater or less than something else. A great many social
science investigations are concerned with such hypotheses. For ex- ,
ample, psychologists have investigated the correlation between dif-
ferent kinds of ability; sociologists have studied the ecology of crime
and of mental illness; anthropologists have investigated th~ relation
between religious beliefs, marriage customs, and other practices in a
society.
Still other hypotheses assert that a particular characteristic or oc-
currence is one of the factors which determine another characteristic
or occurrence. For example, psychoanalytic theory involves t~e hy- ;
pothesis that experiences in infancy and early childhood are important
determinants of adult personality. Hypotheses of this type are referred
to as hypotheses of causal relationship or. simply as causal hypotheses.
They will be discussed in more detail in Chapter 4.
Hypotheses may be developed from various sources. A hypothesis
may be based simply on a hunch. It may rest on the findings of another
study or studies and the expectation that a similar relationship between
two or more variables will hold in the present study. Or it may stem
from a body of theory that, by a process of logical deduction, leads to.
the prediction that if certain conditions are present, certain results
will follow. Regardless of the squrces of a hypothesis, it performsi an
important function within a study: It serves as a guide to (1) the
kind of data that must be collected in order to answer the research
question and (2) the way in which they .can be organized most effi-
ciently in the analysis.3 I
But the sources of the hypotheses of a study have an important

bearing on the nature of the contribution the research will make to
the general body of knowledge. A hypothesis that arises siII?-ply from
intuition or a hunch may ultimately make an important contribution
to science. However, when it has been tested in only one study, there
3 The importance of hypotheses in planning the analysis of a study is dis<:ussed
in Chapter 11.
FORMULATING HYPOTHESES 37
are two limitations on its usefulness. First, there is no assurance that
the relationship between two variables found in the 'given study will
be found in other studies. To go back to the example of Morris and
Davidsen's study-suppose these investigators had just "guessed" that
foreign students who believe Americans think highly of the country
from which they come are likely to have more favorable attitudes
toward the United States than students who believe Americans look
down on their~home country. Even though this proved to be the case
in their study, there would be no way of judging whether the differ-
ence in attitudes between the two groups of students really stemmed
from the difference in their perception of the attitudes of Americans to-
ward their home country, or whether some other factor within the sit-
uation accounted for the difference. Secondly, a hypothesis based
simply on a hunch is likely to be unrelated to other knowledge or
theory. Thus the findings of a study based on such a hypothesis have
no dear connection with the larger body of social science knowledge.
They may raise interesting questions, they may stimulate further re-
search, they may later be integrated into an explanatory theory. But
unless these developments take place, they are likely to remain isolated
bits of information.
A ,hypothesis that arises from the findings of other studies is freed
to some extent from the first of these limitatioris. If the hypothesis is
based on findings of other studies, and if the present study supports
the hypothesis, then the result has helped to confirm this relationship
as a regularly recurring one. Continuing with our example-if two or
more studies find a relationship between perception of Americans' view
of one's' home country and one's attitudes toward the United States,
we have greater confidence in the proposition that a foreign student's
perception of the tegard in which Americans hold his home country
I
does in fact influence his attitudes toward the United States. The pos-
sibility that the apparent relationship is ,due to some special condition
in a given situation/is reduced, though of course not eliminated.
A hypothesis that stems not simply from the findings of an earlier
study but from a theory
I,
stated in more general terms is freed from the
second limitation...,--that of isolation from a larger body of knowledge.
Suppose, for example, that Morris and Davidsen based their predic-
tion on a theory of social stratification which holds that individuals

who are rising in status are likely to be favorably inclined toward in- .
dividuals and objects that are seen as. aiding their upward movement,
while those who are falling in status are likely to be unfavorably in-
clined t~ward individuals and objects that are seen as contributing to
their downward movement. In order to establish a relationship be-
tween their study and the general theory, they must be able to argue
convincingly that their measures could be considered as indicators
of change in status and that it could reasonably be inferred
that the students saw Americans as contributing to this change. 4
If the connection between the specific data of the study and
the general theory has been satisfactorily established in this way, and
if the results of the study confirm the prediction, the investigators
have made a double contribution. On the one hand, the study .findings
help to confirm the general theory by demonstrating that it holas in
this situation as well as in others where it has been applied. On the
other hand, the application of the general theory to the specific case
helps us to understand how it comes about that the foreign student's
perception of American views of his country influences his attitude
toward the United States. 5
In the case of Morris and Davidsen's study, the investigators were
able to draw both on previous research findings and on existing
sociological theory to formulate specific hypotheses, of which the
major one was a more complex version of the one stated in the prC?ced- '
ing paragraphs. The fact that there were in existence both relevant
research findings and a relevant theory simplified the task of formulat-
ing hypotheses; having clearly formulated hypotheses simplified tJ_1e
task of planning what kinds of data to 'gather and how to allalyze
them. The major hypothesis was supported by the findings of the
study. The relation to earlier research strengthened confidence in the
regular occurrence of the relation~hip between estimate of Americans'
views of one's country and one's attitude toward the United States.
The fact that the study was planned in terms of a sociologicaJ theory
4 How the investigators developed this argument is discussed in more detail
on pages 41-42.
5 For a more detailed discussion of the relation between empirical research and
theory, see Chapter 14.
FORMULATING HYPOTHESES 39
about the effects of upward and downward status mobility meant that
the findings contributed to confirmation of the general theory. At the
same time, the theory provided an explanation of the study's findings
as an instance of the more general psychological process stated in the
theory.
Whether or not the nature of the anticipated relationships can be
stated explicitly-i.e., whether or not they can be expressed as hy-
potheses in the formulation stage of an inquiry-depends largely on
the state of knowledge in the area under investigation. Scientific re-
search can begin with well-formulated hypotheses, or it can formulate
hypotheses as the end product of the research.
It goes without saying that the formulation and verification of
hypotheses is a goal of scientific inquiry. Yet there is no short cut to
this goal. In many areas of social relations, significant hypotheses do
not exist. Much exploratory research, therefore, must be done before
hypotheses can be formulated. Such exploratory work is an inevitable
step in scientific progress.
For example, not very much is yet known about the influence of
planned communities on the behavior, feelings, and outlook of the
people living in them. When Merton and his colleagues (forthcoming
report) undertook to investigate this topic in the case of public hous-
ing projects in the United States, there existed/no set of well-estab-
lished hypotheses with which they could start. There were, at the
time, few descriptive studies available that would permit the formula-
tion of hypotheses applicable to this new phenomenon on the Ameri-
can physical and social landscape. Merton's studies were conceived on
a broad basis; a whole series of interrelated research problems was
formulated and a variety of ways in which the impact 0. public housing
would manifest itself I
were anticipated.
One of these problems had to do with the-extent of political par-
I
ticipation. Tenants, in public housing have, as a rule, an opportunity
to share responsibi,lity in the administration of the projects and to
elect local representatives, much as in municipal government. At the
stage of problem formulation, little was known about this aspect of
I,
behavior and its implications. Two opposite effects of participation
in the administration of the projects seemed equally possible. On the
one hand, concern with local self-government might lead the tenants
to become so absorbed in the small matters of the housing project
...
that they were not interested in larger political issues. On the other
hand, participation in the operation of a small community, where
the workings of political forces could easily be observed, might serve
as a training ground for participants and thus lead to a heightened in-
terest in and understanding of political forces on a national plane.
However, since it was not even known whether and to what ex-
tent the tenants availed themselves of the opportunity to participate
in self-government, it would have been premature for the investigators
to base the plan of their work on any hypothesis concerning the
effects of this participation. In the course-of the investigation, it became
evident that participation was the rule. Moreover, it was discovered
that in two vastly different housing projects, concern with local politics
did not replace but rather initiated and reinforced concern with na-
tional affairs. In view of this evidence, it may now be possible to begin
an investigation in public housing-or in any other setting, for that
matter-,in order to verify this hypothesis in different circumstances.
- , On the other hand, in the area of prejudice, where a great amount
of resear~h has already been done, investigations are possible in which
hypotheses can be formulated in advance. This was demonstrated in
the study of public housing (Deutsch and Collins, 1951) which was
discussed in Chapter 1. It will be; remembered that in this' study
the problem was the impact of occupancy pattern on relations between '
Negro and white tenants. In the light of previous research, the in-
vestigators were able to formulate l in advance a number of interrelate~
hypotheses about the effects of the occupancy patterns on theattit~des
of the tenants.
These examples make it cleal; that it is pointless to regard a study
which sets Qut with hypotheses as more '!scientific" than one which
ends with hypotheses. The time for formulation of hypotheses varies
with the nature of the problem and the extent of prior knqwledge
about it. Formulation and reformulation of research questi9ns is a
continuing process. As the German sooiologist Max Weber said,
"Every scientific fulfillment raises new questions; it asks to be sur-
passed and outdated."
DEFINING CONCEPTS 41
Defining Concepts
Any investigator, in order to organize his data so that he may
perceive relationships among them, must make use of concepts. A
concept is an abstraction from observed events or, as McClellat<i
-(1951) puts it, "a shorthand representation of a variety of facts. Its
purpose is to sim lif thinkin b subsumin a number ~o!. ~nts
Un er one general ,heading." Some concepts are quite close to the
llbjects or tacts fhey represent. Thus, for example, the meaning of the
concept dog may be easily illustrated by pointing to specific dogs. The
concept is an abstraction of the characteristics all dogs have in com-
mon-characteristics that are either directly observable or easily
measured. Other 'COncepts, however, cannot be so easily related to the
phenomena they are intended to represent; attitudes, learning, role,
motivation are of this sort. They are inferences, at a higher level of
abstraction from concrete events, and their meaning cannot easily be
conveyed by pointing to specific objects, individuals, or events. Some-
times these higher-level abstractions are referred to as constructs, since
they are constructed from concepts at a lower level of abstraction.
The :greater the distance between one's concepts, or constructs,
and the empirical facts to which they are intended to refer, the greater
the possibility of their being misunderstood or cfrelessly used, and tJhe
greater the care that must be given to defining them. They must be
defined both in abstract terms, giving the general meaning they are
intended to convey, and in terms of the operations by which they will
be represented in the particular study. The former type of definition
is necessary in order to link the study with the body of knowledge using
similar concepts or :constructs. The latter is an essential step in carry-
ing out any research; since data must be collected in terms of observable
facts.
As an illustrati~n, let us go back to Morris and Davidsen's study
of foreign students.iTheir ,hypothesis had to do with the effects of the
student's estimate of the regard in which Americans hold his country
on his attitudes toward the United States. Actually, as we have noted,
I
the hypothesis was1more complex than this. It involved the constructs
of "national status' gain" and "national status loss" through coming
to the United States. National status was defined as "the relative

valuation placed upon one's own country in an international scale of ~
comparison." The hypothesis invol'{ed a distinction between two
estimates 'Of national status: "perceived national status"-the student's
own valuation of his country-and "perceived accorded national
status"-the student's estimate of how Americans would evaluate his
country.6 The difference between these two valuations gave a measure
of "national status discrepancy." This construct was linked to the
general theory of social stratification and status mobility by the as-
sumption that a student who believed Americans rated his home
country higher than he himself did had experienced "national status
gain," and that one who believed Americans rated his country lower
than he did had experienced "national status loss." These key COn-"
structs were given formal definition in terms of the meaning of
status gain and status loss in general sociological theory. A start"was
made on a working definition for this particular study by defining
national status gain and national status loss in terms of a comparison
be~een two ratings made by the student-'his own rating of his
co'untry's position, and :his estimate -of Americans' ratings. Let us now
, ~onsider how the working definitions were further spelled out.7
Establishing Working Definitions 8
No matter how simple or how elaborate an investigator's foqnal

definition of, his concepts, he must find some way of translating t~em
into 'Observable events if he is to carry outimy research. It is notP9ssible
to study "national status gain" or "loss" as such, since these constructs
have no direct counterparts in observable events. The investigator
6 "Actual accorded national status"-the valuation actually made by Americans
-was also taken into account, as a check on the student's perception.
7 A number of other constructs were involved in the hypotheses of this study,
among them identification and attitude. Since we are concerned with the study
for purposes of illustration only, we shall not consider how these terms were
defined.
8 The reader may notice a similarity between our concept of working definitions
and the more commonly used term operational d~finitions. We have avoided this
more usual term because it carries with it certain philosophical connotations that
we do not wish to go into here.
ESTABLISHING WORKING DEFINITIONS 43
must devise some operations that will produce data he is satisfied to
accept as an indicator of his concept. This stage of project formulation
may require considerable ingenuity, especially if the constructs are far
removed from everyday events and if little research using these con-
structs has been carried out.
In the case of the foreign student study, although there had been
considerable research on status gain and loss in other contexts, no
previous study had worked systematically with the concept of "national
status gain or loss," and thus no operations that would constitute work-
ing definitions of these terms had been devised. Morris and Davidsen
had already made a start on their working definition when they decided
that their basic measure would be the discrepancy between the stu-
dent's own rating of his country and his estimate of Americans' ratings.
But in what terms should these ratings be made? Could students be
given a list of countries and asked simply to arrange them in order of
the regard in which they held them? This seemed too ambiguous a
task; from their own reactions, the investigators concluded that coun-
tries would be arranged differently on the basis of different criteria ..
They finally settled on three criteria in terms of which countries were
to be rated: standard of living, cultural standards, and political stand-
ards. On the basis of these ratings, an index of "national status" would
be constructed. (It will be obvious from this eX51mple that the step of
establishing working definitions is very closely related to decisions
both about the data-collection instruments to be used and the pattern
of analysis. This is another instance of the way in which earlier steps
determine later ones, and of the necessity for anticipating later steps in
'earlier stages.)
Working definitions are adequate if the instruments or procedures
. based on them gather data that constitute satisfactory indicators of
the concepts they;are intended to represent. Whether this result has
been achieved is often a matter of judgment. An investigator may feel
that his data provide reasonably good indicators of his concepts; a critic
of the study may f~el that they do not. It frequently happens that the
investigator himself is aware that his data constitute only a very limited
reflection of the c;oncept he has in mind, but, especially in the early
stages of research :on a problem, he may not be able to devise a more
satisfactory One. In any case, although the investigator will usually
, r
report ,his findings in terms of his abstract concepts in order to relate

them more readily to other research and to theory, he and his reader~
must keep in mind that what he has actually found is a relationship
between two sets of data that are intended to represent his concepts.
Thus, what Morris and Davidsen actually found was that the dis-
crepancy betwe~n two ratings of the 'home country was related to the
answers to certain questions about the United States. Although they
may quite properly interpret this as indicating a relationship between
national status gain or loss on the part of a foreign student and his
attitude toward the host country, one should remember that this
interpretation depends on the assumption that the answers to the
interview questions are adequate working definitions of the concepts
national status gain, national status loss, and attitude toward the host
country.
Relating the Findings to Other Knowledge
Scientific research is a community enterprise, even though single

studies are frequently carried out by individual investigators working
alone. Each study rests on earlier ones arid provides a basis fo; future
ones. The more links that can be established between a given study
and other studies or a body of theory, the greater the probable contribu-
tion. .
There are two major ways of relatIng a given study to a larger body
of knowledge. One, obviously,! is to examine the research an,d the
thinking that has already been done on the given research pr~blem or
problems related to it, and to plan the study so that it ties in with this
existing work at as many points as possible. The second is to formulate
the research problem at a level sufficiently abstract so that findings
from the study may be related to findings from other studies concerned
with the same concepts.
The study of foreign students which we have been considering
was formulate<;l at a very high level of abstraction, and thus could
readily be related to other studies of stat,us mobility in different situa-
tions and with different populations. In studies that take their impetus
RELATING THE FINDINGS TO OTHER KNOWLEDGE 45
from some scientific question, it is usually not difficult to formulate
the research problem at a useful level of abstraction, since scientific
questions, by their very nature, are likely to be stated in general terms.
But studies that arise from the need to answer a practical question may
remain at such a specific level that they make no real contribution to
knowledge unless the investigator takes pains to transpose the question
to a higher level of abstraction.
Let us consider the example of an investigator who is asked by an
agency concerned with the improvement of intergroup relations to
carry out a survey for the purpose of evaluating the effectiveness of a
series of cartoons about prejudice. If he sets up the problem in such a
way that he cannot generalize beyond the specific cartoons in question,
he will fail in both his scientific and his practical assignment. The
cartoons he is about to investigate are unique products and to some
extent different from all other cartoons in content and form. If he
sets the research process in motion in order to determine the effective-
ness of these few cartoons, which may be topical today and forgotten
tomorrow, he is involved in a task that will have to begin all over again
as soon as it is finished. If he discovers, for example, that a specific
cartoon attracts and amuses a part of its audience but is misunderstood
by the majority, he has learned little that deserves to be classified as
scientific knowledge. Nor can suoh results provid~ much guidance for
the cartoon producer. In order to remove this limitation on his work,
the investigator, before he proceeds to data collection, must reformu-
late his concrete problem in a manner that will ultimately permit him
to draw conclusions about the more general aspects of both the
cartoon itself and the response of persons exposed to it. In other words,
his concern, in this; stage of problem formulation, must be with the
generalizability of his
I
results. It is not enough to ask whether one par-
ticular cartoon is understood. Rather, he must ask: What aspects of
I
the cartoon are understood-and by whom? To be able to answer the

first part of the qu~tion, he must analyze the general features of the
cartoon. He may emerge with categories such as "satire," "caption
required for understanding," etc. If he can then demonstrate through
i,
his inquiry that this type of cartoon is misunderstood because it is
taken literally instead of satirically, he is in a position to advise the
artist to experiment either with nonsatirical cartoons or with ways of

making the satire clearer.
Only when a problem is formulated in generally meaningful terms
can the social scientist hope to transfer to other problems the knowl-
edge gained from the study of a unique event. Generalizing from the
study of unique occurrences requires that the research problem be
formulated in more abstract terms than might be necessary if one were
concerned only with answering a question about the single event.
Every event in human life, when regarded in its full concreteness, is,
of course, unique. Yet, however rare and atypical, it becomes a legiti-
mate problem for scientific inquiry it underlying processes that may
occur in other unique configurations are specified. In this sense an
earthquake, the bombing of Hiroshima, or the execution of Mussolini
by enraged Italians-unique events in every sense-are subject matter
for research,9 provided the problem is formulated in terms thai refer
to processes potentiaIIy observable on other occasions.
To formulate a research problem in this manner permits the
repetition of studies under different unique conditions. This process
of repetition is known as the replication of research. It is essential to
the deve~opment of confidence in research findings. For example,
Morris and Davidsen's finding of a relationship between national status
gain or loss and attitude toward the host country is interesting and sug-
gestive, and its congruence with the general theory of status mobili.ty
lends credence to the finding. But it can be accepted as generally tru y
only after it has been repeated with other students, attending different
universities, studying in countr~es other than the United States. Such
replication of research will in the end show whether the underlYing
process that the social scientist had in mind accounted for the relation-
ship between the two sets of observed events, or whether it was ac-
counted for by some as yet undiscovered conditions that characterized
the specific study. Although it is essential to search for more general
processes, the investigator must maintain a careful balance 'between
his attention to the unique configuration and to the general aspects of
his observations. To neglect the unique configuration may lead to false
and premature generalizations; to neglect the general aspects may: lead
9 Of course, a major difficulty in investigating such events is thilt of managing
to be in a position where one can obtain trustworthy information about the event,
its initiating conditions, and its consequences.
RELATING THE FINDINGS TO OTHER KNOWLEDGE 47
to failure to develop principles that can be used in understanding situa-
tions other than the specific one studied.
Summary
Many considerations-and many difficulties-arise in selecting a

research ,topic and formulating it as a research project. Basically, these
prQcesses demand that the social scientist be aware of the choices
open to him and that he understand the rules governing problem
formulation.
The formulation of a research project proceeds in several steps:
First, a problem demanding solution must be perceived within
the area circumscribed by the selected topic.
Second, the research task must be reduced to manageable size
or divided into a number of subtasks, each of which can be handled in
a single study.
Next come a number of steps which are so closely interrelated that
their order cannot be specified. If relevant knowledge exists concerning
the research problem, the formulation will contain hypotheses, both
as a guide to the collection and analysis of data and as a means of relat-
ing the study to other studies .or to a body of th60ry. If there is not
enough knowledge to provide a basis for setting up hypobheses at this
stage, the formulation of the problem will indicate the areas in which
an exploratory study aims t.o establish hypotheses.
, Whether or not hypotheses are established at this stage, the study
will require the definition of the concepts that are to be used in organiz-
ing the data. These definitions will include formal ones, designed to
convey the general nature of the process or phenomenon in which
the investigator is interested and its relation to other studies and to
existing social science theory. They will also include working defini-
tions, which make ~ssible the collection of data the investigator is
satisfied to accept as i6dicators of his concepts.
Throughout these processes, there will be concern with the gen-
eralizability of the findings and their relation to other knowledge. This
involves careful study of work already done in the field, and formula-
tion of the research problem in terms general enough to make clear
its r:elation to other knowledge and to permit replication of the re-

search.
In formulating the research problem, subsequent steps in the
research process must be anticipated to ensure that the problem can be
tackled by available techniques. This anticipation should include both
scientific and practical steps.
3
RESEARCH DESIGN
1. Exploratory and Descriptive Studies
Formulative or Exploratory Studies

Descriptive Studies
/
Summary
O NCE THE RESEARCH PROBLEM has been formulated clearly enough
to specify the types of information needed, the investigator
must work out his research design. A research design is the arrangement
of conditions for collection and analysis of data in a manner that aims
to combine relevance to the research purpose with economy in pro-
cedure. It follows that research designs will differ depending on the
research purpose.
Each study, of course, has its own specific purpose. But we may
think of research purposes as falling into a number of broad groupings:'
(1) to gain familiarity with a phenomenon or to achieve new insights
into it, often in order to formulate a more, precise research problem
or to develop hypotheses; (2) to portray accurately the 'characteIistics
of a particular individual, situation, or group (with or without specific
initial hypotheses about the nature of these characteristics); (3) to
determine the frequency with which something occurs or with which
it is associated with something else (usually, but not always, with a
specific initial hypothesis); (4) to test a hypothesis of a causal relation-
ship between variables.
In studies that have the first purpose listed above-generally called
tormulative or exploratory studies-the major emphasis is on discovery .
\ of ideas and insights. Therefore, the research design must be flexible;
enough to permit t~e consideration' of many different a~pects of a
phenomenon.
In studies having the second and third purposes listed above; a
major consideration is accuracy. Therefore a design is needed that will
minimize bias and maximize the reliability of the,evidence collected.
(Bias results from the collection of eviqence in such a way that one
alternative answer to a research question is favored. Evidence is reli-
able to the extent that we can assert confidently that similar findings
would be obtained if the collection of evidence were repeated. For a
detailed discussion of bias and reliability in connection with measure-
ment procedures, see Ohapter 5.) Since studies with these second and
third purposes present similar requirements for research design, we
can treat them together; we shall call them descriptive studIes.
50
RESEARCH DESIGN: I 51
Studies testing causal hypotheses require procedures that will not
only reduce bias and increase reliability but will permit inferences
about causality. Experiments are especially suited to meeting this latter.
requirement. However, many studies concerned with testing hy-
potheses about causal relationships do not take the form of experi-
ments.
In practice, these different types of study are not always sharply
distinguishable. Any given research may have in it elements of two
or more of the functions we have described as characterizing different
types of study. In any single study, ,however, the primary emphasis is
usually on only one of these functions, and the study can be thought
of as falling into the category corresponding to its major function. In
short, although the distinctions among the different types of study
are not clearcut, by and large they can be made; and, for the purpose
of discussing appropriate research designs, it is useful to make them.
Formulative or Exploratory Studies
Many exploratory studies have the purpose of formulating a

problem for more precise investigation or of devel9ping hypotheses. An
exploratory study' may, however, have other functions: increasing the
investigator's fan!liarity with the phenomenon he wishes to investigate
in a subsequent, more highly structured, study, or with the setting in
which he plans to carry out such a study; clarifying concepts; establish-
ing priorities for further research; gathering information about practical
possibilities for carrying out research in real-life settings; providing. a
census of problems regarded as urgent by people working in a given
field of social relations.
Our discussion; will focus on studies that are directed primarily
toward the formulation of problems for more precise investigation or
toward the development of hypotheses. The points made and the
procedures described are, however, applicable to exploratory studies
having other goals. I,
The relative youth of social science and the scarcity of social
science research make it inevitable that much of this research, for a
52 RESEARCH DESIGN: I
time to come, will be of a pioneering character. Few well-trodden paths

exist for the investigator of social relations to- follow; theory is often
either too general or too specific to provide clear guidance for empirical
research. In these circumstances, exploratory research is necessary to
obtain the experience that will be helpful in formulating relevant hy-
potheses for more definitive investigation.
Suppose, for example, that one is interested in obtaining insight
into the process by which social environment influences mental health.
Although there has been much speculative writing on this topic and
some research that is incidentally related to it, the research worker
entering this area is not in a position to advance any precise hypothesis
fpr investigation. Indeed, it would be foolhardy for him to try to do
soC. Without some knowledge of the scope of the area, of the major
social variables influencing mental health, of the settings \n which
these variables occur, any hypothesis that is set forth is likely to be
trivial. In the case of problems about which little knowledge is ava!l-
able, an exploratory study is usually most appropriate.
Occasionally there is a tendency to underestimate the importance
of exploratory research and to regard only experimental work as "sci-
entific." However, if experimental work is to :have either theoretical or
social value, it must be relevant to broader issues than those posed tn
the experiment. Such relevance can result only from adequate explora-
tion of the dimensions of the problem with which the research is at-
tempting to deal.
Although, for the most part, we are discussing .the exploratory
study as an entity, it is appropriate to consider it also as an initial, st(lP
in a continuous research process. In practice, the most difficult portion
of an inquiry is its initiation. The most careful methods during the
later stages of an investigation are of little value if an incorrect or ir-
relevant start has been made. As Northrop (1947) has pointed out:
Again and again investigators h;Ive plunged into a subject

matter, sending out questionnaires, gathering a tremendous
amount of data, even performing experiments, only to/,come
out at the end wondering what it all proves. . .. Others, not-
ing the success of a given scientific method in one field, have car-
ried this method hastily and uncriticallv into their own, only to
FORMULATIVE OR EXPLORATORY STUDIES 53
end later in a similar disillusionment. All such experiences are a
sign that the initiation of inquiry has been glossed over too hast-
ily, without any appreciation of its importance or its difficulty.
For whatever purpose an exploratory study is undertaken, in-
genuity and good luck will inevitably playa part in determining its
productiveness. Nevertheless, it is possible to suggest certain methods
that are likely to be especially fruitful in the search for important vari-
ables and meaningful'hypotheses. These methods include: (1) a re-
view of the related social science and other pertinent literature; (2) a
survey of people who have had practical experience with the problem
to be studied; and (3) an analysis of "insight-stimulating" examples.
Most exploratory studies utilize one or more of these approaches.
Whatever method is 'chosen, it must be used flexibly. As the ini-
tially vaguely defined pmblem is transformed into one with more
precise meaning, frequent changes in theresearch procedure are neces-
sary in order to provide for the gathering of data relevant to the
emerging hypotheses.
THE SURVEY OF THE LITERATURE
One of the simplest ways of economizing effort in an inquiry is to

review and build upon the work already done by others. In a study of
the type we are discussing here, the focus of review is on hypotheses
that may serve as leads for further investigation. Hypotheses may have
been explicitly stated by previous workers; the task then is to gather
-
the various hypotheses that have been put forward, to evaluate their
usefulness as a basis for further research, and to consider whether they
suggest new hypotheses. More frequently, however, an exploratory
study is concerned with an area in which hypotheses nave not yet been
formulated; the tasf then is to review the available material with
sensitivity to the hyp'otheses that may be derived from it.
I
In many areas a bibliographical survey will undoubtedly be mOrC
time-consuming thah rewarding; often one will find that no research
of significance has, been done in one's area of interest. This is perhaps
less often true, ho\\[~ver, than is assumed by those who fail to build
upon the work of previous investigators. In any case, the conclusion
that there is no r~levant material would be unjustified without a
thorough search of journals which are likely to carry articles on the

given topic, and of such publications as the Psychological Abstracts,
the Child Development Abstracts and Bibliography, the Sociological
Abstracts, the bulletin of Current Sociological Research issued by
the American Sociological Society, the listing of Doctoral Dissertations
Accepted by American Universities compiled for the Association of
Research Libraries, and the Dissertation Abstracts available in micro-
film from University Microfilms, Ann Arbor, Michigan. In addition to
these general sources, some government agencies and yoluntary
organizations publish listings or summaries of research in their special
fields of interest. For example, the Children's Bureau of the U. S.
Department of Health, Education, and Welfare publishes a ~ulletin
of Research relating to Children; the Anti-Defamation League of
B'nai B'rith issues Research Reports, summaries of research relevant
to intergroup relations. Professional organizations, resea~ch groups,
and voluntary organizations are sources of information about un-
published research in their specific fields.
It would be a mistake to confine <?ne's bibliographical survey to
studies that are immediately relevant to one's area of interest. Perhaps
the most fruitful means of developing ,hypotheses is the attempt' to
apply to the area in which one is working concepts and theories de-
veloped in completely different research contexts. Thus, the theory
of adaptation level developed in work on psychophysical problems may.
provide stimulating analogies for work on factors influencing, for e~
ample, the perception of characteristics of members of an ethnic group
other than one's own; level-of-aspiration theory may provide a parallel
for studying changing comm~nity goals; learning theory may give
insights into the process of attitudinal change; the concepts 9 role,
social norm, psychological need, frustration, group structure, etc.,
may direct attention to important variables in any new situation to be
studied.
The sensitive descriptions to be found in !the works of creative
writers are also a fertile ground of hypotheses for study. Although the
social scientist does not aim at capturing the richness of th,e novelist's
descriptions, he may find in the world's literature many stimulating
suggestions about important variables in the situations he wishes to
study. For example, Paton's Cry the Beloved Country and Too Late
FORMULATIVE OR EXPLORATORY S'l'UDIES 55
the Phalarope, Sartre's Portrait of the Anti-Semite, Wright's Native
Son, and E. M. Forster's A Passage to India are laden with penetrating
analyses of the causes and consequences of prejudice.
THE EXPERIENCE SURVEY!
Probably only a small proportion of existing knowledge and ex-

perience is ever put into written form. Many people, in the course of
their everyday experiences, are in a position to observe the effects of
alternative decisions and actions with respect to problems of human
relations. The director of a settlement house, and the group workers
on his staff, are likely to develop insights into the characteristics of
young delinquents and the probable effectiveness of various approaohes
to them. The psychiatric social worker may acquire sensitivity to the
environmental conditions that impede the adjustment of patients
released from a mental institution and, on the other hand, to factors
that support adjustment. Such specialists acquire, in the routine of
their work, a reservoir of experience that could be of tremendous value
in helping the social scientist to become aware of the important in-
fluences operating in any situation he may be called upon to study. It
is the purpose of an experience survey to gather and synthesize such
experience. /
THE SELECTION OF RESPONDENTS. Research economy dictates that
the respondents in an experience survey be carefully selected. The aim
of the experience survey is to obtain insight into the relationships be-
tween variables rather than to get an accurate picture of current prac-
tices or a simple consensus as to best practices. One is looking for
provocative ideas and useful insigJ;W;, not for the statistics of the profes-
sion. Thus the respondents must be chosen because of the likelihood
that they will offer the contributions sought. In other words, a
selected sample of ~eople working in the area is called for.2
In an experience survey it is a waste of time and effort to interview
people who have little competence, or little relevant experience, or
! Much of the following is based on an unpublished article by C. Selltiz, S. W.
Cook, and R. Hogrefel~ntitled "The Experience Survey: A Step in Program Design
for Field Research on :Unexplored Problems."
2 In terms of the different kinds of sample discussed in Appendix B, this is at.
purposive sample. '
-,1"
who lack ability to communicate their experience. An individual may

have worked in a field for a number of years and not necessarily be a
good informant. Perhaps the most direct method of selecting inform-
ants is to ask strategically placed administrators working in the area
one desires to study to point out the most informative, experienced,
and analytical people. Although this does not guarantee insight-
stimulating respondents, one does obtain through this method people
with a reputation for good experience and good idea'S; The ~ike1ihood
of their being useful informants is, of course, increased if they are re-
commended by more than one source, particularly if the different
sources are known to have dissimilar points of view.
Although a random sample of practitioners may not be of value
in an experience survey, it is nevertheless important to select respond-
ents so as to ensure a representation of different types of experience.
Wherever there is reason to believe that different vantage points may
influence the content of observation, an effort must be made to include
variation in point of view and in type of experience. Thus, for example,
in an experience survey of factors affecting employee morale in indus-
try, it is advantageous to interview representatives of both management
and labor. It is also desirable to interview people at different levels in
each group-workers, foremen, personnel managers, shop stewards,
educational dir~tors of unions, etc., to obtain a varied perspective.
Apart from interviewing enough people to ensure adequate repre-
sentation of different types of experience, there is no simple rule for
determining the number of informants who should_be interviewed in
an experience survey. At a certain point, the investigator will find that
additional interviews do not provide new insights, that the answers
fall into a pattern with whioh he is already familiar. At thi~ point,
further interviewing becomes less and less rewarding.
THE QUESTIONING OF RESPONDENTS. Before any systematic ,attempt
is made to collect the insights of experienced practitioners, it is, of
course, necessary to have some preliminary ideas of the important
issues in the area. One source of such ideas, as we have Ipreviously
indicated, is a bibliographical survey. Before an interview schedule for
the systematic'questioning of informants is developed the information
from such a survey almost invariably must be supplemented by a
3
number of unstructured interviews with people who have had exten-
sive experience in the field to be investigated.
Even in the more systematic interviewing of the later informants,
it is essential to maintain a considerable degree of flexibility. The pur-
pose of providing the interview with structure is to ensure that all
people interviewed respond to the questions the researcher wishes to
have answered; however, the formulative and discovery functions of
the experience survey require that the interview always allow the re-
spondent to raise issues and questions the investigator has not pre-
viously considered.
In formulating questions for an interview schedule with social
practitioners, it is generally useful to orient inquiries to "what works."
That is, the questions should usually be of the following form: "If
(a given effect) is desired, what influences or what methods will, in
your experience, be most likely to produce it?" There are several reasons
for f'Ocusing primarily on change-producing influences. First, the social
practitioner, necessarily heeding the imperatives of his work, is oriented
toward producing change, toward "what works." He is more likely
to understand and to be able to respond to a practically phrased ques-
tion than to one that is worded abstractly. Second, the emphasis on
change allows the investigator to collect insights into processes 'Oper-
/
ating over a period of time, which the practitioner is in a uniquely
favorable position to observe. Third, if the investigator's concern is not
'Only with the theoretical relationships among variables but also with
their implications for social action, he needs to know how these vari-
ables tend to cluster in everyday life and how these commonly found
clusters of variables promote Dr hinder socially desirable objectives.
Where possiple, in order to stimulate the informant to compare
the major alternative methods for accomplishing a specific end, it is
desirable to prob~ beyond a m,9"e statement of-general principles about
prDducinga givenrchange. Concrete illustrations, from the respondent's
own experience, Of successful and unsuccessful attempts to achieve a
specific effect are. of particular value. They enable the investigator to
collate experien~s 'Of different people in diverse circumstances and
I
3 For a discussion of "structured" and "unstructured" interviews, see Chapter
7, pages 255-268,
thus to form tentative conclusions and generalizations which go beyond

the observations of anyone informant.
To illustrate one type of question that has been found particu-
larly useful in experience surveys, we present the following questions
from such a survey of intergroup relations in industry.4
Some people believe that a non-discriminatory employment
policy requires that minority group members with average quali-
fications for a particular job be brought into the job on a first-
come-first-served basis. Other people believe that in introducing
members of a minority groJlP it is important for the first member
of the group to have certain special qualifications. The assump-
tion is that this will make it easier to get acceptance for other
members of that minority group in this situation later.
Question: Should special qualifications be set for the first
members of a minority group to be introduced into a new sit-
uation?
If the first members of a minority group introduced into a

new situation are specially selected, there are several bases upon
which the selection may be made. ,
Alternative A: Some people believe that the first members of a
minority group introduced should be very high in ability
for the particular job they are to fill.
Alternative B: Other people believe it is more important that e
the minority group members should be very pleasant and
agreeable personally. .
Alternative C: Still other people believe the .main consideration
is that the first minority group members be as similar as
possible to the people they are to work with in physical
appearance, manners, speech, education, interests, and
previous experience.
Question: What are the relative advantages and disadvantages of
these three principles for selecting the first members of a
minority group to be introduced into a new situation?
In general, an experience-collecting interview is likely to be quite
long, frequently lasting several hours. In view of the miture of the in-
formation wanted, it is desirable to prepare the respondent a week or
4 This survey was conducted by J. Harding and R. Hogrefe of the Commission
,
on Community Interrelations of the American Jewish Congress.
FORMULATIVE OR'EXPLORATORY STUDIES 59
so before the interview is to take place by sending him a copy of the
questions to be discussed. This gives him an opportunity not only to
do some advance thinking, but to consult his colleagues and to add the
knowledge to be gained from their experiences.
SOME BY-PRODUCTS OF EXPERIENCE SURVEYS. An experience survey,
as well as being a good source of hypotheses, can provide information
about the practica1 possibi1ities for doing different types of research.
Where can the facilities for research be obtained? Which factors can
be controlled and which cannot in the situations one might wish to
study? What variables tend to be clustered together in community
settings? How ready are agencies, professional workers, and ordinary
citizens to cooperate in controlled research studies of the problem in
question? The answers to these and similar practical questions may
be one of the by-products of a carefully planned experience survey. In
addition, such a survey may provide a census of the problems consid-
ered urgent by the people working in a given area. This census may be
extremely useful in establishing priorities in a program of research.
The report of an experience survey also provides a summary of the
knowledge of skilled practitioners about the effectiveness of various
methods and procedures in achieving specified goals. In lieu of more
definitive knowledge, this information may be of enormous value as
a guide to "best" practices in a given field. Of course, in presenting
such a summary, it should be made clear that the survey was in no
sense based on a random sample of workers in the field. Its usefulness
comes from the presentation of insights and effective practices rather
than from the presentation of the "typical."
THE ANALYSIS OF "INSIGHT-STIMULATING" EXAMPLES 5
Scientists working in relatively unformulated areas, where there

is little experience to serve as a guide, have found the intensive study
of selected exal11ples to be a particularly fruitful method for stimulat-
ing insights and suggesting hypotheses for research. The remarkable
theoretical insights of Sigmund Freud were, of course, stimulated by
I,
5 Much of ~he following discussion is based on an unpublished paper by
J. P. Dean, "The: Method of Unstructured Pilot Inquiry." A cOlldensation of this
paper may be fOUlid in De<ln ( 1954) .
his intensive studies of patients. So, too, profound changes in our con-
ception of the relationship between man and society have been brought
about largely by anthropological studies of primitive cultures.
From these examples it should be clear that we are nO,t describing
what is sometimes called the "case-study" approach, in the narrow
sense of studying the records kept by social agencies or psychothera-
pists, but rather the intensive study of selected instances of the phe-
nomenon in which one is interested. The focus may be on individuals,
on situations, on groups, on communities. The method of study may
be the examination of existing records; it may also be unstructured in-
terviewing or participant observation or some other approach.
What features of this approach make it an appropriate procedure
for the evoking of insights? A major one is the attitude of the investi-
gator, which is one of alert receptivity, of seeking rather than of test-
ing. Instead of limiting himself to the testing of existing hypotheses,
he is guided by the features of the object being studied. His inquiry
.is constantly in the process of reformulation and redirection as new
information is obtained. Frequent changes are made in the types of
data collected or in the criteria for case selection as emerging hypo-
theses require Rew information. '
A second feature is the intensity of the study of the individual,
group, community, culture, incident, or situation selected for investi-
gation. One attempts to obtain sufficient information to characterize
and explain both the unique features of the case being studied and
those which it has in common with other cases. In the study of the
individual, this may entail an extensive examination of both his present
situation and his life history. In 'the study of a group, an incident, etc.,
individuals may be treated as informants about the object, rather
than being themselves the objects of intensive analysis.
A third characteristic of this approach is its reliance on the inte-
grative powers of the investigator, on his ability to draw together many
diverse bits of information into a unified interpretation. This last
characteristic has led many critics to view the analysis of insight-stim-
ulating instances as a sort of projective technique, in which conclusions
reflect primarily
I
the investigator's predisposition ratlier than the ob-
ject of study. Even if this reproach is appropriate to many case studies,
the characteristic is not necessarily undesirable when the purpose is to
FORMULATIVE (,)R EXPLORATORY STUDIES 61
evoke rather than to test hypotheses. For even if the case material is
merely the stimulus for the explicit statement of a previously unform-
ulated hypothesis, it may serve a worth-while function.
Social scientists who work with this approach have frequently
found that the study of a few instances may produce a wealth of new
insights, whereas a host of others will yield few new ideas. Although
here, as elsewhere, no simple rules can be established for the selection
of the instances to be studied, experience indicates that for particular
problems certain types are more appropriate than others. We list below
some of these types, together with the purposes for which they have
been found most useful. The list is not exhaustive, nor are the types
mutually exclusive.
1. The reactions of strangers or newcomers may point up charac-
teristics of a community that might otherwise lbe overlooked by an
investigator reared in the culture. A stranger is likely to be sensitive
to social customs and practices that are more 'or less taken for granted
by the members of a community. His ouriosity Dr surprise or bewil-
derment may call attention to features of community life to which
members of the community have become so accustomed that they no
longer notice them.
2. Marginal individuals, or groups, who are moving from one cul-
tural grouping to another and are on the periph~ /
of both groups, are
similar in some respects to strangers or outsiders. Because they.are "in
between," exposed to conflicting pressures of the groups from and to
which they are moving, they can often reveal dramatically the major
influences operating in each 'group. For example, in the field of inter-
group relations, the study of emigrants, of displaced persons, of Jews
who are trying to be assimilated into local cultural groups, of Negroes
who are trying to "pass" as whites, of people who are in the process of
conversion to or ~rom Catholicism, of people in areas of disputed na-
tional sovereignty, is likely to be highly rewarding.
3. Study of ihdividuals or groups who are in transition from one
stage of developIPent to the next has been fruitful, particularly in
anthropological investigations of the influence of culture upon per-
sonality. In hi~ I,investigation of any culture, the anthropologist is
necessarily limite,d by time to a cross-sectional study rather than one
that would tracer individuals from birth to death. The study of indi-
viduals who are at a point of transition helps overcome, to some extent,

the limitations of a cross-sectional investigation. For example, inten-
sive study of babies who are being weaned, or of adolescents, or of
women in the 'Period of menopause, is likely to give considerable in-
sight into the'process of ,change and into the socio-psychological charac-
teristics of contiguous stages of development (see Mead, 1946). Sim-
ilarly, the study of groups or societies in transition may be of value
in understanding the processes of social change.
4. Deviants, isolates, and pathological cases may, by indirection,
throw light on more common cases. The study of deviants (for ex-
ample, individuals who are inter~sted in world government although
most of their associates hold nationalistic and isolationist views) may
~erve to highlight thesocial norms and practices from which they ar~
deviating. It may indicate the types of pressure to conform and the
socio-psychological consequences of nonconformity; it may perhaps
even help to reveal the methods by which social changes may be pro-
duced. In a similar manner, the analysis of isolates may accentuate the
factors making for cohesiveness in a given group or community. It
may also reval much about the way in which attitudes and informa-
tion are transmitted in a social group (see, for example, Festinger,'
Schachter, and Back, 1950). The contributions of psychoanalysis to
the understanding of personality are a striking illustration of the in-
sights that may be uncovered by a study of pathological cases, which
frequently serve to underscore, through exaggeration, basic processes
in nonpathological cases. Thus, for example, Ackerman and Jahoda
(1950), in a study based on reports of psychoanalysts about cases
under treatment, found that dep~essed patients are rarely prejudi~ed.
This finding, with its implication that individuals who have turned
their aggression against themselves do not need prejudice as a channel
for aggression, provides an interesting hypothesis about the psycho-
dynamics of, prejudice. Extreme cases may also be enlightening when
the interest is in social dynamics. The study of the breakdown of
social controls and their reinstatement, as illustrated in natqral dis-
asters or a race riot-see, for example, Lee (1943)-may result in
worth-while insights into the processes of social control.
5. "Pure" cases are often productive. For example, Levy (1943),
in his study of maternal overprotection, was interested in three ques-
tions: \\!hat leads a mother to be overprotective, what effects does
maternal overprotection have on the child, and how can the difficul-
ties that may result from overprotection be prevented or remedied?
Reasoning that he could best find clues through the study of cases
of marked overprotection, he examined many records of cases treated
in a child guidance clinic. There were two major criteria for selecting
~ESes for intensive study. First, they must show extreme overprotection
by the mother, as evidenced by inseparability of mother and child,
the mother's treating the child as a baby, and the mother's prevention
of independent behavior on the part of the child. Second, they must
be "pure" cases, in the sense that the mother's behavior was consist-
ently overprotective and also in the sense that there was no evidence
of rejection of the child. (This latter criterion was introduced on the
ground that the combination of overprotection and rejection differs
from overprotection per se and may have different origins and con-
sequences.) Of the cases that met these criteria, only those were re-
tained that contained enough information about the mother to make
possible some inferences about the factors producing overprotective
behavior, enough information about the child to yield insight into the
kinds of problems produced by such behavior, and enough information
about ,treatment of the case to give clues as to the effects of therapy.
Of the more than five hundred cases examined, O1)ly twenty met all
these criteria; these formed the basis of Levy's study.
t 6. The characteristics of individuals who fit well in a given situa-
tion and those who do not fit well provide valuable clues about the
nature of the situation. Thus, the knowledge that the people who feel
at home in a given community, who seem to fit best, are either highly
dependent or authoritarian in personality provides some insight into
the characteristics of the community. Similarly, discovery that those
who feel thwarted by\a given situation are the young and ambitious
and those who have ~onsiderable personal initiative would provide a
clue to the nature of th'e situation .
.7. Selection of inqividuals who represent different positions in the
social structure helps to produce a rounded view of the situation they
are reflecting. In almostI,
all social groups, one finds variations in social
status and specialization of social roles or functions. Individuals oc-
cupying these different positions are likely to see any given situation
from different perspectives, and this diversity is productive of insights.

Thus, interviewing the porters in a housing project may be as impor-
tant fOI. understanding relations within the project as interviewing the
manag(;t. The discrepancies as well as the similarities in the social per-
ceptions of people holding different positions or fulfilling diverse
functions are frequently revealing.
8. A review of the investigator's own experience and a careful ex-
amination of his reactions as he attempts to "project" himself into
the situation of the ~llbjects he is studying may be a valuable source
of insights. After all, the "case" with which an investigator is likely
to have the greatest familiarity (though also the most bias) is himself.
As Jones's biography (1953) of Freud makes clear, many of Freud's (
most valuable insights came from his efforts to understand himself. To
be sure, there probably are few persons with the qualities of a Freud; we
cannot expect that analysis of one's Own experience will often have
such fruitful results. But, even so, here is a source of ideas that ought
not to be neglected. That the point needs to be made at all stems
from the fact that scientists are often- so preoccupied with the im-
portance of objectivity that they actively strive to maintain as great a
distance as they can between themselves and the objects of their study.
In the stages of research in which one is looking for ideas rather than
conclusions, such objectivity may be inappropriate.
Our listing of "insight-stimulating" cases is inevitably incompl~e.
The type of cases that will be of most value depends largely, of course,
On the problem with which one is concerned. Nevertheless, it is gen-
erally true that in an explanatory study, cases that provide sharp con-
trasts or .have striking feature~ are most useful, since in .exploratory
work the discernment of minute differences is likely to ~difficult.
It is important to remember that exploratory studies merely lead
to insights or hypotheses; they do noti test or demonstrate .them. In
selecting cases that have special characteristics, one has by definition
taken cases that are not typical. Although marginal, deviantl or "pure"
cases are likely to be fruitful sources of ideas about process~ that may
occur in more typical cases, one cannot assume that these ,processes do
in fact occur in cases other than those one has studied. 'Pressures on
marginal individuals may be quite different from those on individuals
who are well integrated in a group; deviant cases may be deviant not
only in their behavior but in the psychological processes underlying it.
An exploratory study must always be regarded as simply a first step;
more carefully controlled studies are needed to test whether the hypoth-
eses that emerge have general applicability.
Descriptive Studies
An enormous amount of social research has been concern~d with

describing the characteristics of comIl_!unities. One may study the
people of a co~munity: the~ age distribution, their national o;-ra~~al
background, the state of their physical or [email protected] heaJth, the amount
of educatIOnth~ have had-the list could be extended indefinitely ,
One may study community facilities and their use: the condition of
housing, the extent to which libraries are used, the amount of crime
in various neighborhoods-again, the list is endless. One may under-
take to describe the structure of social organization in the community,
or the major patterns of behavior.
Another vast body of research has been concerned with estimat-
ing the proportion of people in a specified population who hold
certain views or attitudes or who behave in certiln ways: How many
favor admitting Communist China to the United Nations? How many
'believe that Negroes and whites should live in separate neighborhoods?
How many think capital punishment should be abolished? How
many watched which television programs last week?
Still other studies are concerned with specific predictions: How
many people will vote for a certain candidate? How many will cash
their government bOnds during a given period? In what neighborhoods
is the growth of population likely to be great enough to require new
schools or new transportation facilities within the near future?
Others are con~erned with discovering, or testing, whether certain
variables are associated: Do more Catholics than Protestants vote
Democratic? Do people who spend a good deal of time reading go to
the movies often? ~Are people who are prejudiced against Jews also
likely to be prejudiced against Negroes? Do girls, on the whole, learn
to talk at an earlier age than boys? Note that none of these questions,;
as they have been present~ involves a hypothesis that on~ of the
variab1es leads to or produces the other; questions embodymg such
hypotheses pose different requirements for research procedures.
This is a considerable arra~ of research interests, which we have
grouped under the heading of Ja!escriptive studies. We have grouped
them together because, from the p~int of view of research proced~res,
they share certain important charactt~tics. The resear~1: qu~sbo;nS
presuppose much prior knowledge of th&,:~roblem to be mvestIgated,
as contrasted with the questions that fo~ IIp.e basis for exploratory
studies. The investigator must 'be able to define 1.S;learly what it is he
wants to measure and must find adequate methods fo-i)I, measuring it. I~
addition, he must be able to specify who is to be include~~ in the defim-

t lon 0
f" a gIven
. commumty ." or" a.gIven popuI " 1'{Ii
atlOn. j:
collecting
evidence for a study of this sort, what is needed is not so Titeuch fleXI-
bility as a clear formulation of what and who is to be measd: ted , and
techniques for valid and reliable measurements.6 sh
Descriptive
- studies are not
. _limited
- to anyone method ot ill\data
~ ted
collection. They may employ any or ~1I of the methodsJto be prese,t 'J. ,
in subsequent chapters. Thus Lundberg, Komarovsky, and MclI1yery
(1934), in their study of leisure, collected information through inln r-
views, questionnaires, systematic direct observation,_
I.,i ,
analysis
_o
of cp;
munity records, and parti,cipant observation: "".-/
, ! More than two hundred formal interviews lasting from one

_to three hours were secured with ministers, school officials, and
women in their homes. There were many more interviews inci-
dental to the circulation of schedules, and a very large number
through casual contact and association of living in ;:t community.
In addition to the customary analysis of secondary material such
as histories, annual reports of organizations, and newspapers, we
studied the mobility of the population through the analysis of
telephone and other directories of fifteen villages and cities, in-
volving a checking of more than two huIlqred thousand names
and addr'esses. Among the projects of formfl direct observation
was the recording of a schedule of the activities of some 6800 com-
muters on trains. Constant traveling throughout th<;)county,
6 Fo/a discussion of validity and reliability of measurement, see Chapter 5.
DESCRIPTIVE STUDIES 67
visits to homes, schools, playgrQunds, all major recreational re-
sorts and all the contacts incidental to daily living complete the
story of observations upon which this study of leisure of a people
is based.
Although descriptive studies may use a wide range of techniques,
this does not mean that they are characterized by the flexibilitt_ that
marks exploratory studies. The procedures to be used in a descriptive
_/
study must be carefully planned. Because the aim is to obtain complete
and accurate information, the research design must make much more
provision for protection against bias than is required in exploratory
studies. Because of the amount of work frequently involved in de-
scriptive studies, concern with economy of research effort is extremely
. important. These considerations of economy and protection against
bias enter at every stage: formulating the objectives of the study; de-
signing the methods of data collection; selecting the sample; collecting,
processing, and analyzing the data; and reporting the findings. The
following paragraphs point out some or the ways in which economy
and protection against bias are taken into account in the design of a
descriptive study.7 As an illustration, we shall consider a study of the
treatment of Negro patrons in New York City festaurants (Selltiz,
1955).
FORMULATING THE OBJECTIVES OF THE STUDY
--
~rst step in a descriptive study, as in any other, is to define
the question that is t? be answered. Unless the objectives are specified
"with sufficient precision to ensure that the data collected are relevant
to the question rais~the study may not provide the desired informa-
tion.
In our example, the research question was: Do restaurants in New
York City discriminate against Negro patrons? But before data could
be gathered to answer this question, 'it was necessary to specify what
was meant by discrim.i.nation. It was defined as any inequality between
the treatment accorded white and Negro diners, unless there seemed
~ 7 For a more detailed discussion of descriptive studies, and especially those that
_ake the form of surveys of opinion, attitudes, etc., see Hyman (1955, Part II),
and Parten (1950).
reason to believe that the difference in treatment was due to some

factor other than the difference in race. This general concept was
translated into a working definition by specifying the kinds of behavior
on the part of restaurant personnel that might be taken as indicative of
discrimination: outright refusal to serve prospective Negro diners; in-
direct refusal to serve Negroes, by keeping them waiting indefinitely
or by stating that reservations were necessary; evidence of confusion at
the appearance of Negroes in the restaurant, such as a hasty conference
between headwaiter and waiter; direction of Negroes to an undesirable
table or to one that placed them out of view of other diners; poor
service; inferior food; overcharges. Each item in the working definition
included the basic concept of inequality of treatment; thus, poor
~ervice in itself was not to be considered evidence of discrimination.
The focal point of the study was the question: Do restaurants in
New York City discriminate against Negro patrons? Discussions with
various people who might be expected to have a good basis for esti-
mating the probable extent of discrimination showed such wide diver-
gence of opinion that there seemed little basis for advancing a predic-
tion about the existence or the extent of discrimination. A number of
subsidiary questions were also formulated: Would discrimination occur
more frequently in relatively high-priced restaurants? Would discrim-
ination be more likely in restaurants with headwaiters? In "American"
rather than "foreign" restaurants? These subSidiary questions point~d
to the need for collecting certain kinds of information about the res-
taurants studied.
Considerations of economy also entered into the specification of
the research question. The study was undertaken by a voluntary group
with limited resOurces. Looking forward to later stages of the study-
selection of a sample of restaurants, collection of data, reporting of
findings:.....the planners considered whether they were likely to be able
to test a large enough sample of all types of eating places throughout
New York City to provide reliable findings. The obvious negative
answer led to a restriction both of the geographical area to be covered
(about 150 square blocks in east midtown Manhattan) and of the price
range of restaurants to be included (eliminating the' most and the
least expensive) .
DESIGNING THE METHODS OF DATA COLLECTION
After the problem has been formulated specifically enough to

indicate what data are required, the methods bv which the data are to
be obtained must be selected. Techniques for collecting the informa-
tion must be devised if, as is likely, no suitable ones already exist. Each
of the various methods-observation, interviewing, questionnaires,
projective techniques, the examination of records, and so forth-has
its advantages and limitations, which are <discussed in detail in later
chapters.
In the study we are using for illustration, the specific method of
collecting data was somewhat unusual: teams of Negro and white
diners went to restaurants, ate, and reported their experiences. This was
an expensive procedure, but it was considered the one most likely to
give trustworthy evidence about restaurant practices. Basically it was
an observational technique; as such, it was especially appropriate for
the study of behavior in its natural setting. Had the study been con-
cerned with the feelings of restaurant owners about serving Negroes,
or with the opinions of the white population about whether Negroes
are or should be served in all restaurants, other techniques, such as
interviewing, questionnaires, or projective methods~ would have been
more appropriate. Had it been concerned with a question such as the
volume of qpsiness done by restaurants, it might have been possible to
obtain the information by examining records.
The stage of developing data-collection procedures is one of the
major points at which safeguards against bias and unreliability are
introduced. In the study of rstaurant discrimination, they were
directed against two major possible sources of bias: differences in
treatment of white ~nd Negro diners due to some factor 'other than
I .
the difference in race, and distortions in the reporting of experienceS.
Elaborate precahtions were taken to rule out other possible rea
sons for differences [in treatment; such as the possibility that those,
arriving first would be given the most desirable table, that better-diesSed
people would be givrn better service, that men Or older people would
be given better service than women or younger ones, etc. First, t~H: pairs
of diners going to a' given restaurant were matched in sex and age. It
the Negro team consisted of two men, the white team going to the
same restaurant consisted of two men; with few exceptions, both
teams were of about the same age. Since the entire group was relatively
I
homogenous in socioeconomic level, in dress, and in general social

behavior, no special matching of teams in these respects was needed.
Next, the testers were given instructions to ensure that their behavior
would be alike in certain respects, such as the number of courses.
ordered and the approximate price of the meal. Above all, they were
instructed to accept without protest any behavior on the part of res-
taurant personnel, and ~ot to give any indication that they had any
purpose for going to the restaurant other than eating.
Efforts to guard against bias in reporting entered into b,oth the
construction of the report form and the training of the testers. Most of
the questions in the report form called for short factual an~ers:
''When did you enter the restaurant?" "When were you seated?"';'Did
you pick your own ta?le, or were you a~igned by a restaurant em-
ployee?" It seemed unlikely that replies to such questions would be'
greatly distorted by the testers' feelings. Moreover, they made it pos-
sible to identify cases in which the Negro team was given better treat-
ment than the white team as well as those in which it was treated less
well. Thus the possibility that the over-all findings would be biased
by undue attention to instances of discriminatory treatment of the
Negro tea~ was reduced. In an intensive training session, the testers
were given detailed instructions about procedures and an opportunity'
to practice in imaginary situations.
Analogous safeguards must 'be introduced in any study, whatever
the method of data collection. Questions must be carefullY' examined
for the possibility that their wording may suggest one answer rather
than another; interviewers must be instructed not to ask leading ques-
tions or express their own opinion; ob~ervers must be trained so that
they all record a given item of behavior in the same way.
Once the data-collection instruments have been constructed, they
mustbe pretested before they are used in the study proper. 'Questions
that:seem clear ,and straightforward to the research staff may, in a trial
testing; prove difficult to comprehend, or'ambiguous, or simply not
productive' of useful information. Observational categories, statistical
DESCRIPTIVE STIJDIES 71
forms, etc., may be awkwa~d or inappropriate to the material being
studied.
In the survey of restaurant discrimination, there were two pre-
tests: one by members of the committee planning the study, covering
only a few restaurants; the second, by volunteers similar to the testers
in the final study, covering a sample of cafeterias and luncheonettes.
In this case, the pretests indicated need for only minor revisions in
the testing instructions and report form, but they revealed serious
organizational and administrative problems-in recruiting testers,
making assignments, filling out and returning report forms, supervis-
ing progress, etc.-and led to much more careful planning of these
aspects in the survey proper. Much difficulty can be avoided by care-
fully pretesting the techniques to be used, to ensure that they will
collect the information needed.
SELECTING THE S~PLE
In many-though by no means all-descriptive studies, the in-

vestigator wishes to be able to make statements about some defined
group of people or objects (in our example, restaurants). It is rarely
necessary to study all the people in the group in order to provide an
accurate and reliable description of the attitudef and behavior of its
members. More often than not, a sample of the population to be
studied is sufficient.
Much work has been done on the problem of designing samples
in such a way that they yield accurate information with a minimum
amount of research effort. At this point, it may be profitable to illus-
trate how an aware~ess of statistical considerations may result in con-
siderable research ~nomy. Rowntree (1941), in his classic study of
poverty in York, England, investigated every working-cl:Jss household.
To check the accur~cy of sampling methods, he selected, according to
a systematic procedure, every tenth interview schedule and compared
the results obtained thereby with those obtained from all the cases.
Similar calculation~ were based on samples of 1 in 20, 1 in 30, 1 in 40,
and 1 in 50. The t~ble below shows his results for one type of informa-
tion-the proporti9n of income spent on rent by families in five differ-
ent income groupings. It is apparent that the various samples, regard-

less of their size, gave results fairly close to those found for all the
households in each income class. Thus, comparing the extreme right-
hand column (figures based on a sample of 1 in 50 families) with the
left-hand column (figures based on the complete survey), we see that
PERCENTAGE OF INCOME SPENT ON RENT

Income
Class Complete Survey Sample Surveys
1 in 10 1 in 20 1 in 30 1 in 40 1 in 50
uA" 26.5 26.6 25.9 27.0 2S.3 27.1
"B" 22.7 22.9 23.5 23~3 22.3 22.6
...,c" 19.8 18.1 17.2 IS.3 17.2 18.0
"0" 15.S 16.0 14.4 IS.S 17.1 16.9
"E" 11.3 11.0 10.1 10.7 11.2 11.5
the sample shows families in income class "A" spending 27.l per cent
of their income on rent, while the total survey shows that such families
spent 26.5 per cent of their income on rent; in income class "B," the
sample shows 22.6 per cent of income spent on rent, while the total
survey shows 22.7 per cent; and so on. For no income group does the
figure shown by the sample based on 1 family in 50 differ by more
than 2 percentage points from that shown by th~ complete survey. In
other words, by ;taking a sample of 1 in 50 instead of every working-
class hou.sehold 'iJ the city, essentially the same results would have
been obtained. That is, a substantial saving in time and effort Could
I
have been effected without significantly impairing the results. ,

Even very slight differences between figures-for example, the
difference between 26.5 per cent and 27.l per cent-may be statistically
significant. In deciding whether a difference is worthy of attention,
I
two kinds of considerations are relevant: statistical ones and ,practical
ones. Whether a difference between two figures is statistically signifi-
cant involves precisely the question being discussed here-the probable
deviation of figures based on samples from those for the total popula-
tion from which the samples are drawn. 8 If a particular ~ifference is
8 The concept of statistical significance is discussed in Chapter 11, pages 414-
422. Fl)r more detailed discussion. consult any standard statistics textbook.
statistically significant, the decision whether it has practical signifi-
cance is, of course, a matter for judgment in terms of the consequences
of the different figures. If some major decision-as, for example, a
general increase in wag~s-is to be made on the ~asis of the findings,
then a difference of as lIttle as one percentage pomt between the sam-
ple findings and the true state of affairs for the total population may be
serious. However, in most surveys it seems unlikely that a difference
as small as those shown in the table above would lead to any major
change in interpretation or in recommendations for action, if such
recommendations were among the goals of the study.
It is, of course, important that the study findings based on a
sample (that is, on only part of the group about which statements are
to be made) should be. a reasonably accurate representation of the
state of affairs in the total group (referred to, in sampling terminology,
as the "population"). This means that the sample should be selected
in such a way that findings based on it are likely to correspond
closely to those that would be obtained if the population were studied.
To bring this about, considerable attention has been paid to problems
and methods of sampling. These are discussed in Appendix B.
In the restaurant survey, a systematic stratified sample was used.
(See Appendix B for definition of these terms). The units in the
sample were, of course, restaurants. A complete list of restaurants in
the area had been compiled by volunteers who walked through every
. block, recording each eating place-its name, address, price range, and
other relevant information. Since one of the questions to be investi-
gated was whether the occurrence of discrimination was related to the
price level of restaurants, the sample was stratified on that basis. The
cards on which the data about each restaurant had been entered were
arranged in order of the estimated price of an average meal. It was not
I
known in advance e~actly how many testers would be available, and
therefore how many!restaurants could be included in the sample. In
view of the decision Ito concentrate on restaurants in. a middle price
range, the median card was selected as the first case in the sample; the
other cases were sdeeted by taking, alternately, every fourth card above
and every fourth ca~d below the median. This meant that, whatever
the final size of the sample, it would constitute 25 per cent of the res-
taurants within a specifiable price range in the given geographical

area. The final sample consisted of 62 restaurants, constituting.25 per
cent of the 248 restaurants with average prices from $1.30 to $3.75.9
The research purpose determines the appropriate sampling unit.
In an election study, the sampling units would be eligible voters; in a
study of family budgets, families; in a study of infant behavior, they
might be time periods. Whatever the sampling unit, it is important to
have a basis for identifying the total population of such units and a
specified method of selecting units from that population (see Ap-
pendixB).
COLLECTING AND CHECKING THE DATA
To obtain consistent data free from the errors introduced by in-

dividual interviewers, observers, and others, it is necessary to supervise
the staff of field workers closely as they collect and record informa,tion.
Checks must be set up, for example, to ensure that interviewers are
honest and that the data they collecr are unbiased.lO As data are
collected, they should be examined for completeness, comprehensibil-
ity, consistency, and reliability.
In the restaurant survey, all testers, immediately after leaving the
restaurants, returned to the study headquarters to fill out their report
forms. The two members of each team filled out the form together, but
without any discussion with the other team that had gone to the same
restaurant. After the two ,reports for a given restaurant had been com-
pleted, a member ofthe supervisory committee checked both, to make
sure that they were complete and to see whether the two teams agreed.
If there was any discrepancy, it was discussed with the two teams
9 As pointed out in Appendix 13, systematic sampling procedures such as this
have certain limitations that are not present when samples are randomly selected.
However, given the decision that the sample should constitute 25 per cent of the
restaurants within the price range covered by the study, plus the uncertainty about
how many restaurants could be tested, systematic sampling was more appropriate
because the boundaries of the price range to be covered could be expanded as
arrangements were made for additional tests. A random sampling procedure, on
the other hand, would have necessitated an advance decision as to the, "population"
of restaurants to be sampled-i.e., the price range to be covered.
10 For a discussion of "cheating" by interviewers and methods of detecting it,
see Blankenship et al. (1947).
jointly. In most cases it became apparent that the discrepancy did not
represent genuine disagreement, but rather an oversight by one team
in filling out the report, or a physical inability on the part of one team
to observe behavior visible to the other. In the two or three in-
stances of genuine disagreement, the supervisor wrote a detailed report
of the versions given by the two teams. Such checking at the time
reports are turned in prevents much difficulty at later stages and en-
sures the usability of data that might otherwise have to be discarded.
ANALYZING THE RESULTS
The process of analysis includes: coding the interview replies,

observations, etc:. (p1acing each item in the appropriate category);
tabulating the data (counting the number of items in each category);
and performing statistical computations. These procedures are dis-
cussed in considerab1e detail in Chapter 11. Here we may simply note
that both considerations of economy and the need for safeguards
against error enter into each of these steps. In general, considerations
of economy require that the analysis be planned in detail before actual
work on it is started. In this way, the investigator may avoid unneces-
sary labor, such as working out tables for whichI he later finds he has
no use or, on the other hand, re-doing some tables because he failed
to include relevant data. To be sure, complete planning of the analysis
in advance is not always possible or even desirable; new ideas occur
to the investigator as he examines his preliminary findings. But, except
in exploratory studies, it is always possible and desirable to work out
in advance the basic outlines of the analysis.
Safeguards against error in coding ordinarily take the form of
checking the reli;lbility of the coders-that is, determining the
extent to which they agree in assigning a given item to a given cate-
gory.1f the code requires complex judgments, the usual procedure is to
have two or more coders independently code a sample of the material,
continuing-wit~ additional training or, if necessary, modifications
in the code-until I,they have achieved a satisfactory degree of reliability.
In the case of simple codes, one coder may process the entire group of
cases without advance determination of reliability; a second person
may then code, say, every twentieth case, in order to provide a check
on accuracy.
If the material is to be tabulated by machine, it must be entered
on appropriate cards; this is usually done by punching holes corre-
sponding to a given code. It is advisable to check the accuracy of punch-
ing; again, it is usual to check only a sample of the cards.
The accuracy of tabulation may also be checked by having a
sample of the tables re-done. However, at this stage it is possible to
make a rough check by comparing figures from different tables. For
example, the figures in each table should add up to the total number
of cases, unless there is reason to omit some from a given table. More-
over, certain classifications are likely to be used in more than one table,
and these figures provide a partial check on accuracy. For example, in
the restaurant survey, in addition to the basic table showing the num~
ber of restaurants in which discrimination was found and the'number
in which it was not, there were tables showing the number of res-
taurants in which a given kind of discrimination was encountered, the
Occurrence of discrimination in restaurants at different price levels, in
American and "foreign" restaurants, etc. If any of these mOre detailed
tables had shown a different number of restaurants as discriminatory
than the basic table, this would have been evidence of errOr.
Finally, statistical computations are needed in a study of any com-
plexity; averages, percentages, correlations must be computed. Again,
these operations may be checked by having a second person re-do a
sample of them.
Statistical operations of ~nother sort a re introduced for ,the pur-
4
pose of safeguarding against drawing unjustified conclusions from the

findings. These involve such procedures as estimating from the sample
findings the probable occurrence of some characteristic in the popula-
tion the sample is intended to represent, and estimating the probability
that differences found between subgroups in the sample represent dif-
ferences between the corresponding subgroups in the population rather
than simple chance differences due to sampling. (The logic underlying
such procedl,ues is discussed briefly in Chapter 11; the procedures are
discussed in detail in standard statistics texts) .
The process of analysis in the restaurant survey was relatively
simple. The number of cases was not large; the only complex coding
judgment required was the decision whether or not a given restaurant
was to be classified as having shown discrimination against the Negro
diners; and tl~e number of characteristics of restaurants that were to
be examined in relation to the occurrence of discrimination was small.
In view of the small number of cases, the importance of the decision
whether or not discrimination had occurred, and the structure of the
organization that carried out the survey, an unusual and uneconomical
procedure was followed in coding restaurants as having shown or not
shown discrimination. Eight persons served as coders. Preliminary
classification of each restaurant as discriminatory or nondiscriminatory
was made by a pair of coders working together. Finally the whole group,
working as a committee, reviewed all the tests and made the final de-
cisions as to whether or not there had been clear inequality of treat-
ment that could not reasonably be considered accidental.
The small number of cases and the few variables to be examined
in relation to the occurrence of discrimination made tabulation a simple
process.
The appropriate statistical operations to safeguard the drawing
of conclusions were carried out. Thus, from the finding that 42 per
cent of the restaurants tested gave the Negro diners clearly inferior
treatment, it was estimated, by means of the for~llia for the standard
error of a percentage, that probably between 36 per cent and 48 per cent
of all restaurants in the given area and price range discriminated against
Negroes.l l Similarly, appropriate tests of significance were used to
determine whether differences in frequency of discrimination in res-
taurants of different price levels, different nationality, with and with-
out headwaiters, etc., probably represented true differences or merely
chance variations. Of the variables examined, only price proved to be
significantly related I to discrimination. When the restaurants were
divided into three piice ranges, it was found that a far smaller propor-
I
tion of those in the lowest range were discriminatory in their treatment
of Negroes. I
. 11 Another survey qf thp. same area two years later, in 1952, after a campaign to
mduce nondiscriminatiqp, fnund discrimination in only 16 per cent of the res-
taurants tested. Computing the standard error of this percentage led to the esti
mate that at this time probably between 12 per cent and 20 per cent of all res-
taurants in the given area and price range discriminated against Negroes.
78 RESEARCH DESICN: I
Summary
In this chapter we have pointed out that the function of research

design is to provide for the collection of relevant evidence with
minimal expenditure of effort, time, and money. These considerations
are important in any study, whatever its purpose. But how they can best
be achieved depends to a considerable extent on the research purpose.
When the purpose of a study is exploration, a flexible research design,
which provides opportunity for considering many different aspects of
a problem, is appropriate. When the purpose of a study is accurate
description of a situation or of an association between variables, ac-
curacy becomes a major consideration; a design is needed that mini-
mizes bias and maximizes the reliability of the evidence collected.
Designs appropriate to exploratory and descriptive studies have been
discus.sed in this chapter.
When the purpose of a study is to test a hypothesis of a cause-and-
effect relation between variables, other requirements are introduced.
Research designs appropriate for such studies are discussed in the
following chapter.
4
RESEARCH DESIGN
II. Studies Testing Causal Hypotheses
The Logic of Testing Hypotheses about

Causal Relationships
Causal Inference from Experiments
Causal Inference from Other Study Designs
Summary
Experimental observations are only experience carefuJ1y
planned in advance, and designed to form a secure basis of new
knowledge. R. A. FISHER
The Logic of Testing Hypotheses about Causal Relationshipsl
A HYPOTHESIS of causal relationship asserts that a particular char-

acteristic or occurrence (X) is one of the factors that deter-
m!ne another characteristic or occurrence (Y). Studies designed to
test such hypotheses must provide data from which one can legitimately
infer that X does or does not enter into the determination of Y. Be-
fore we consider the kinds of research procedures tha,t can provide
grounds for inferences of this sort, some .discussion of the concept of
"causality" is needed.
THE CONCEPT OF "CAUSALITY"
The concept of causality is complex, and a thorough analysis of it

would b.e far beyond the scope of this book. We shall limit our
discussion to those points that seem essential to understanding the
.requirements for research procedures in studies designed to. test causal
hypotheses.
"Common-sense" thinking about causality tends to be. along the
line that a single event (the I'cause") always leads to another single
event (the "effect"). In modern science, the emphasis is rather on a
multiplicity of "determining conditions," which together make the
Occurrence of a given event probable. Both common-sens~ and sci-
entific thinking are concerned with discovering necessary and suffi-
cient conditions for an event. (We shall define these terms in the
1 For a more detailed discussion of this topic, see Churchman (1948~, and
Cohen and Nagel (1934. Chapter 13).
SO
LOGIC OF TESTING CAUSAL HYPOTHESES 81
folluwing paragraphs.) But while common sense leads one to expect
that one factor may provide a complete explanation, the scientist rarely
if ever expects to find a single factor or condition that is both necessary
and sufficient to bring about an event. Rather, he is interested in
contributory conditions, contingent conditions, alternative conditions
-all of which he will expect to find operating to make the occurrence
of the event probable, but not certain. (Further explanation of these
terms, too, follows. )
A necessary condition, as the term implies, is one that must occur
if the phenomenon of which it is a "cause" is to occur. If X is a neces-
sary condition of Y, then Y win never occur unless condition X occurs.
Example: Prior experimentation with drugs is a necessary condition of
drug addiction, since it would be impossible for addiction to occur if
the individual had never used drugs.
A sufficient condition is one that is always followed by the phe-
nomenon of which it is a "cause." If X is a sufficient condition of Y,
then whenever X occurs, Y will always occur. Example: Destruction
of the optic nerve is a sufficieI).t condition of blindness, since no one
whose optic nerve has been destroyed can see.
A condition may be both necessary and sufficient for the occur-
rence of a phenomenon. In such a case, Y would never occur unless
X occurred, and whenever X occurred, Y would also occur. In other
words, there would be no instance in which either X or Y appeared
alone. Clearly, neith~r of our examples fits this pattern. Although
drug addiction (Y) can never occur unless the person has experimented
with the use of drugs (X), it is nevertheless true that an individual can
experiment with drugs without becoming addicted; thus experimenta-
tion with drugs is a .necessary but not sufficient condition of addiction.
In order to underst~nd the occurrence of drug addiction, we must find
other contributory conditions.
I
On the other hand, although destruction of the optic nerve (X)
will always lead to blindness (Y), blindness may come about in other
ways even though the optic nerve is undamaged; destruction of the
nerve is a sufficien;t but not a necessary condition of blindness. To
understand all the ~'causes" of blindness, we must seek alternative con
ditions that may produce it.
We shall use our drug-addiction example to illustrate the seard.
82 RESEARCH DESIGN: II
for contributory, contingent, and alternative conditions. A contributory

condition is one that increases the likelihood that a given phenomenon
will occur, but does not make it certain; this is because it is only one of
a number of factors that together determine the occurrence of the
phenomenon. Thus, research on drug addiction is not satisfied with
recognizing that experimentation with drugs is a necessary condition
of addiction, but goes on to consider what personal, family, and neigh-
borhood factors are conducive to such experimentation and what
factors make it mQre likely that an individual who has experimented
will become addicted (Chein, 1956). Studies comparing adolescent
male addicts and non-addicts may find, for example, that a considerably
higher proportion of those who have become addicts grew up in homes
from which the father was missing; in other words, these studies suggest
that the absence of a father figure from the home during childhood is
a contributory influence in the development of drug addiction in
adolescent males.
But the behavior with which social science deals is extremely com-
plex; the interaction of factors must be taken into account. A factor
that operates as a contributory condition of a phenomenon under one
set of circumstances may not do so; under another. The conditions
,under which ,a giyen variable is a contributory cause of a given
phenomenon are ~ailed contingezit conditions. Much research in social
science is concerned with identifying such conditions. In our drug-
addiction example, further studies may be directed toward I
discovering
whether the relationsh,ip between father's absence and drug addiction
holds under a variety of conditions. They may finC! that in neighbor-
hoods where drug use by adolescents is rare or nonexistent, boys
brought up without a father in the home do not turn to drug~. The
hypothesis may then be refined as follows: In neighborhoods where
the use of drugs is common (~ontingent condition), absence .of a boy's
father contributes to the probability that the boy will become an
addict.
Awareness of the mUltiplicity of contributory causes leads also
to an intere~t in alternative conditions that may make the occurrence
of a phenomenon more likely. Thus, it may be found that, in neighbor-
hoods with a high rate of drug use, the rate of addiction is especially
high not only among boys who have been brought up without !1 fathe:
but among boys whose fathers have been present in the home during
their childhood but have treated them with hostility or indifference.
The hypothesis would then be reformulated to take account of these
alternative contributory conditions: Either the absence of a father
figure or the occurrence of hostile or indifferent treatment by the father
contributes to the probability of addiction in neighborhoods where
the use of drugs is common. The social scientist may then seek a
hypothesis that includes a factor common to both the alternative con-
tributory causes-for example, that in neighborhoods where drugs
are easily available, the lack of opportunity for identification with a
father figure during childhood makes drug addiction more likely in
adolescence.
BASES FOR INFERRING THE EXISTENCE OF A CAUSAL

RELATIONSHIP BETWEEN Two VARIABLES
Most hypotheses in social science are concerned with contributory

or alternative conditions and the contingencies under which they
operate. However, it is impossible to demonstrate directly that a given
characteristic or occurrence (X) determines another characteristic or
occurrence (Y), either by itself or in combination with other char-
acteristics or occu~rences (A, B, C, etc). Rather, /we are always in the
position of interring from observed data that the hypothesis that X
is a condition for the Occurrence of Y is or is not tenable with some
specified degree of confidence. What evidence is necessary to justify
such inferences?
CONCOMITANT VARIATION. One type of relevant evidence concerns
concomitant variation-that is, the extent to which X and Y occur
together or vary togfther in the way predicted by the hypothesis. Let
us say we wish to tes,t the hypothesis that X is a contributory condition
of Y. Unless we find that the proportion of cases having the char-
acteristic Y is greathr among cases having the characteristic X than
among those not h4ving the characteristic X, we will ordinarily con-
clude that the hypothesis is not tenable. Moreover, if our hypothesis
also specifies that the
I,
amount of Y is determined by the amount of X,
we should also find that, on the whole, those cases that show a higher
amount of X also sh,ow a higher amount of Y.
T
!
84 RESEARCH DESIGN: 11
An example may help to make this clear. Suppose that physicians

in a northern town notice a sudden increase in the number of patients
who come to them with severe digestive upsets. Suppose it also happens
that a tropical fruit-let us say, the mango-has recently become avail-
able in the markets of this town. One of the doctors thinks these two
sets of events may be related; he advances the hypothesis that eating
mangoes (X) may lead to severe digestive upsets (Y) in people who
are not accustomed to the fruit (contingent condition). For a period
of time, several doctors ask all patients with digestive upsets whether
they have recently eaten any mangoes. If none of them has, the
hypothesis that eating mangoes is a cause2 of the digestive upsets
would be rejected.
But suppose some of them have eaten mangoes. We would then
ask a number of people without digestive disturbance whether they
had eaten any mangoes. If it turned out that the proportion having
eaten the fruit in this group was as great.as the proportion in the group
with digestive upsets-say, 30'per cent in each group-we would infer
that eating mangoes did not have much to do with developing diges-
tive disturbances; in other words, we would reject the hypothesis that
it was a cause of the upsets. It is important to note, however, that we
may be wrong in rejecting the hypothesis. It may be that those who
ate the mangoes and became ill had a slight malfunctioning of the
liver, which was aggravated by eating the mangoes and thus led to the
digestive upsets; while those who ate the mangoes and did not become
ill did not have this liver difficulty. In other words, eating mangoes
may be a contributory condition of digestive upset under. the con-
tingent condition of malfunctioning of the liver. By negleqting the
possibility of a contingent condition under which eating mangoes leads
to digestive upsets, we may have maqe an incorrect inference from
the fact that equal proportions of people with and without digestive
upsets had eaten the fruit.
Suppose, on the other hand, that 90 per cent of p~tients with
digestive upsets had eaten mangoes, and that only 30 per cent of the
2 Since, as has been pointed out, modem science approaches causality in terms
of multiple determining conditions rather than with the expectation of finding
a single factor that always leads to a given event, the word "cause" must be under
stood, throughout this book, as meaning "one of a number of determining con
ditions which together make the occurrence of a given event probable."
people without digestive upsets had done so. We would conclude that
the hypothesis that eating mangoes leads to digestive disturbances
waS tenable.
Evidence of whether X and Y vary together in the predicted way
may be sought from the other direction. We might locate a group of
people who had eaten mangoes and :another group who had not, and
compare the incidence of digestive upsets in the two groups. If similar
proportions of people in the two groups showed symptoms of the
disease, we would again conclude that eating the fruit did not lead to
development of the disease. Suppose, however, that 70 per cent of
those who had eaten mangoes had digestive upsets, while only 10
per cent of those who had not eaten them showed such symptoms; we
would conclude that the hypothesis of a causal relationship between
mangoes and digestive upsets was tenable.
But it would be simply tenable, not proved. Three other possible
explanations of the relationship would, at this point, be equally ten-
able: (1) having digestive disorders in some way led people to crave
mangoes, so that the direction of the relationship was the opposite of
that hypothesized; (2) some other condition led to both eating man-
goes and having digestive disturbances; (3) some other condition,
which merely happened to be associated with eating mangoes, was
responsible for the upsets. /
TIME ORDER OF OCCURRENCE OF VARIABLES. The alternative hy-
pothesis that digestive disturbances lead to a craving for mangoes and
thus to eating them brings us to a second type of evidence'relevant to
inferences about causality-the time order of the two events. One
event cannot be considered the "cause" of another if it occurS after
the other event. The occurrence of a causal factor may precede or may
be simultaneous with the occurrence of an effect; by definition, an
effect cannot be produced by an event that occurs only after the effect
has taken place. Ho"wever, it is possible for each term in the relation-
ship to be both a "Cause" and an "effect" of the other term. Many of
the basic relationships of physical science are of this sort. For example,
Boyle's law asserts that, under conditions of constant temperature, the
pressure of a gas is,_inversely related to its volume. This means that
any change we are able to produce in the volume of the gas will
simultaneously result in a change in its pressure, and also that any
change we are able to produce in the pressure will simultaneously

result in a change in its volume. The two variables in the relationship
function interchangeably as causal factor and as effect, so the relation-
ship is said to be symmetrical.
An example of a symmetrical causal relationship in the social
sciences is Homans' hypothesis that "the higher the rank of a person
within a group, the more nearly his activities conform to the norms
of the group . . . . The relationship is strictly mutual: the closer the
person's activities come to the norm, the higher his rank will tend to
be, but it is also true that, rank being taken as the independent variable,
the higher the person's rank, the closer his activities will come to the
norm." (Homans, 1950.)
Although symmetrical causal relationships are frequently found
in social phenomena, it is often useful and convenient to focus upon
the influence of one factor on the other. Thus, in testing Ijomans'
hypothesis, one might wish to see whether one could cause a deviant
member of a group to conform more closely to the group norms by
increasing his rank within the group. _One would not conclude that
increasing an individual's rank in the group was a cause of greater
conformity unless the increase in rank was followed by an increase in
conforming behavior. In distinguishing between cause and effect, it
is useful to establish which of the two e~ents came first, assuming they
did not occur simultaneously. Knowing that an increase in rank, in a
specific instance, preceded an increase in conformity, we know that
the increase in conformity was not the causal factor. (The finding that
in this instance an increase in l rank has led to greater conformity does
not, of course, rule out the possibility that the reverse would also
hold true.) However, knowledge of temporal priority is -not in itself
sufficient ground for inferring causalitr; other factors than the pre-
sumed cause may have produced the effect.
In our digestive-upset example,- ifi people ate mangoes after they
had developed digestive ups~ts, rather than before, then eating man-
goes would not be considered a cause of digestive upsets. Let us say,
however, that questioning showed that every patient suffering from
digestive upset who had eaten mangoes had eaten the fruit before the
symptoms developed. The hypothesis that eating mangoes is a COn-
tributory condition of the disease would remain tenable, and we would
have ruled out one of the alternative hypotheses-that having digestive
disorders led people to crave mangoes. But we must still consider the
remaining alternative hypotheses: that some other condition led to
both eating mangoes and digestive disturbances; and that some other
condition, which merely happened to be associated with eating
mangoes, was responsible for the upsets.
ELIMINATION OF OTHER POSSIBLE CAUSAL FACTORS. \Ve might
consider the possibility that some chemical imbalance led both to a
craving for mangoes and to digestive disturbances. Such a hypothesii
might be tested by examining a sample of people, identifying thos(,
with the specified type of chemical imbalance, and tracing both sub-
groupS over a period of time to discover whether a larger proportion of
those with the chemical imbalance ate mangoes and developed diges.
tive upsets.
Or we might search for factors accidentally associated with eating
mangoes which might be responsible for the digestive upsets. Let us
say that mangoes were sold in only four stores in the town. Suppose that
in three of these stores, the mangoes were sprayed with a certain prep-
aration to improve their appearance, while in the fourth they were
not. Suppose further that it was found that all those who had eaten
mangoes and had developed digestive upsets had bought the fruit in
one of the stores where it was sprayed, while all those who had eaten
the fruit ahd had not developed such symptoms had bought it ip. the
store where it was not sprayed. The hypothesis that eating mangoes
was in itself a cause of digestiv~ upset would be discarded, and atten-
tion would be turned to the effects of the spray; for example, did the
spray lead to digestive upsets when it was used on other fruit, or was
there some interaction between the spray and the chemical composi-
tion of mangoes thatiproduced the toxic effects?
SUMMARY OF TYPES OF EVIDENCE RELEVANT TO CAUSAL INFERENCE.
Let us briefly sumu{arize the three major types_of evidence that are
relevant to testing h~potheses about causal relationships:
1. Evidence ofi concomitant variation-that is, that X (the
assumed causal, or independent, variable) and Y (the assumed eHect,
or dependent, or criterion, variable) are associated in the way prooicted
by the hypothesis. Ih
I
the case of a hypothesis that X is a contributory
condition of Y, this would mean that Y should appear in more cases
where X is present than in cases where X is absent. Other types of

causal hypotheses (that X is a necessary or sufficient cause of Y, or a
contingent cause in association with A, or an alternate cause with B)
would call for other specific patterns for association between X and Y.
2. Evidence that Y did not occur before X.
3. Evidence ruling out other factors as possible determining con-
ditions of Y.
It must be stressed that such evidence merely provides a reason-
able basis for inferring that X is or is not a cause of Y; it does not
provide absolute certainty. Ori the one hand, if the evidence does not
support the hypothesis of a causal relationship, it may be that we have
neglected some condition under which X is a determinant of Y, and
thus that we have been wrong in completely rejecting the possibility
that X is a determining condition of Y. On the other hand, if the
evidence supports the hypothesis, it may still be that we have neg-
lected some other factn[\ .associated with X which is, in' fact, the
determining condition of Y. (In short, we may conclude that it is
reasonable to believe that X is or is not a cause of Y, but we can never
be certain that the relationship has-been conclusively demonstrated.
The cumulation of studies that point to one or the other conclusion
helps to increase our confid.ence in its probable correctness"but still
does not constitute absolute proof.
:ADEQUACY' OF DIFFERENT' RESEARCH DESIG"NS AS

Si?URCES 0; EVIDt~cE< il[ "
Some types of study design provide more convincing grounds .for

dr;nying qlUsal intereI1ces than do others. In our example, all the pro
ced~res suggested so far:fo~.testing the hypothesis that eating mangoes
is a cause of digestive upsets have made use of situations th<lt occurred
in the <;>rdinary course of events: cO!J1pariso~ of individuals who had
become ill with thQse who had not; comparison of those who had
eaten mangoes, for whatever reasons, with those who had not; etc.
~ucp. an approach called fOLa number of different studies, anyone of
which provided only;,tenuQus grounds for testing the hypothesis be-
cause it left a~temativ~ hypotheses untested. For eXilmple, an investi-
gation focu~ed,.on wh~~th<1r .. ~tirig qf mangoes had preceded or fol
lowed the digestive upset would provide no evidence as to the possible
operation of other factors.
An experimental design provides both greater certainty and greater
efficiency by making possible the simultaneous gathering of various
lines of evidence. In an experimental test of the hypothesis about the
relation between mangoes and digestive upset, the investigator would
arrange for a number of subjects to eat mangoes and for a number of
comparable subjects not to eat any of the fruit during the period of
the experiment. In other words, he would select the subjects to be
assigned to different "treatments," and would in one treatment expose
the subjects to the presumed causal variable and in the other treatment
not expose them to it (technically referred to as "manipulating the
independent variable"). He would assign the subjects to the different
treatments in such a way as to ensure that the two groups did not differ
except by chance before eating the mangoes. Comparison of the in-
cidence of digestive upsets in the two groups after one group had
eaten mangoes and the other had not would provide evidence of
whether eating mangoes and having digestive upsets varied together.
By keeping careful records of the time of eating mangoes and the time
of onset of digestive disturbances, he would gain evidence as to which
of the variables came first. Equating the two groups before they ate
the mangoes would lend assurance not only that they did not differ in
incidence of digestive upsets before eating the mangoes (thus con-
tributing further evidence about the time relation between the two
variables)' but that they did not differ in some other way or ways that
might lead to a subsequent difference in incidence of digestive upsets.
In addition, the experimenter might introduce "controls" to protect
against the possibility that different experiences during the experiment,
other than the eatingior not eating of mangoes, might be responsible for
a difference in occurrence of digestive upsets; for example, he might see
to it that all the other food eaten by all the subjects was the same.
Moreover, he might, build into his experiment provision for testing
hypotheses abou~ specific alternative causal factors. For example, he
might simultaneously test the hypothesis about the effects of the spray
used on the mangoe~ by having some subjects eat mangoes that had
not been sprayed, some eat mangoes that had been sprayed, others eat
other fruits that had been sprayed, etc. Thus. in a single study, he
could gather evidence of the three kinds needed to provide a basis for
inferring a causal relationship: concomitant variation, time order, and
the possible influence of other factors.
When an experiment is possible, it is the most effective method
of testing a hypothesis that one variable, X, causally influences another
variable, Y. Many questions of causal relationship lend themselves
easily to experimental study. For example, the investigator who is in-
terested in the relative effectiveness of group discussion and decision
versus reading a pamphlet or listening to a lecture as a method of
changing behavior can set up a situation in which one or more groups
of individuals discuss a certain issue and come to a decision about it,
wbile compa~ble individuals read a pamphlet or listen to a lecture on
the subject. Simil~rly, the influence of subliminal stimuli on the per-
ception of subsequent supraliminal stimuli can be investigated by ex-
posing subjects to such stimuli. Or the effects of "packaging" on the
evaluation of a product can be tested by giving one sample" of people
a product in a container of a given style, a comparable sample the'
same product in a different container. In such cases, the investigator
himself manipulates the independent' variable.
In other pr~lems, however, manipulation of the independent
variable by the experimenter, or assignment of subjects to different
treatments, is not feasible. Suppose one wishes to study the effects of
different methods of child-rearing on the personality structure of chilo
dren. He is not likely to be able to assign certain children to be brought
up in one way, others in another. (He might be able to do something
of this sort in the case of children in institutions, -but his findings could
I
notweII be generalized to children in family settings.). He mu;;t'proceed
by locating children who have been brought up in different ways and
assessing their personalities. If he fin,ds a correlation, he has secured
evidence of concomitant variation. In order to provide.a hasis for in-
ferring that the child-training practIces (X) are a cause, of the per-
sonality structure (Y), he must gather evidence that Y did not precede
X and that other possible factors are not the determining ones.
Ordinarily, the evidence on these points will be less convincing than
that provided by an experiment.
Hypotheses about the effects of attributes of individuals (rather
than of the situations in which they are placed) often are not amenable
to experimental investigation in the sense of manipulation of the "in-
dependent" variable by the investigator. To be sure, a hypothesis that
hungry subjects will be more likely to interpret ambiguous pictures as
representing food than will subjects who are not hungry can be
tested experimentally; the degree of hunger can be controlled reason-
ably well by specifying the length of time subjects must go without
eating before viewing the pictures. Many attributes of individuals,
however, cannot be manipulated in this way. Non-manipulatable at-
tributes are involved, for example, in such hypotheses as: Brain damage
impairs the ability to think abstractly; or, People will tend to remem
ber those parts of a message that are consistent with their own views
and to forget those that are contrary. The investigator working with
human subjects will not manipulate the variable of brain damage
willfully by destroying portions of the brain; he must seek existing
cases of brain damage. And he cannot assign certain views to certain
individuals; they bring their views with them. The investigator achieves
the variation, he wants, not by direct manipulation of the variable itself,
but by selection of individuals in whom the variable is present' or
absent, strong or weak, etc. He presents brain-damaged and non-
damaged subjects with the same task; he asks individuals with different
views to read the same passage; etc.
The logic of testing hypotheses about the presumed effects of an
attribute of a person, such as brain damage, which is not created ex-
perimentally, is essential1y the same as that of testing hypotheses in any
other nonexperimental study. The nonexperimental study, in its de-
sign, does not allow One to rule out in advance, with any confidence,
the possibility that the effect was created by some other factor that is
correlated with the presumed causal factor. Hence, one is faced with
the necessity of ruli~g out on an ex post facto basis (i.e., after the
presumed causal variable has already occurred) the possibility that
other factors correlated with the presumed causal factor may have
produced the observ~d effect. For example, if we exposed patients
with brain damage and patients without brain damage to a test, dif-
ferences in their test performance might reflect the effects of brain
damage or they ~i~ht reflect such other factors as differences in
anxiety that are asso~iated with different types of illness, differencet
associated with socipeconomic variables (e.g., brain damage occurs
more frequently in certain types of occupations), etc. Hypothetically,

if we felt free to create brain damage (as is done with lower animals),
we might assign people to the treatment of "being brain-damaged"
or to the treatment of "not being brain-damaged" in such a way as to
rule out in advance the possibility that test differences would be ex-
plainable by other factors. One could then rule out known possible
alternative explana"tions such as anxiety and socioeconomic variables
by assigning subjests to the "brain-damaged" and "not brain-damaged"
conditions in, such a way that the two groups would be similar in the
types of illness represented, the degree of anxiety, the variety of oc-
cupations, etc. Moreover, in the ideal experiment, one is able to
eliminate alternative explanations that may not have been thought of
in advance. For example, one might not consider, in advance, the pos-
sibility that the presumed differences .due to "brain damage" might
reflect sex or age differences. Even so, if one were experiment?lly creat-
ing the "brain-damaged" and "not brain-damaged" conditions and as-
signing subjects to the treatments randomly, the subjects in the two
treatments should not differ in their _sex or age distribution except by
chance. Hence, the random assignment of subjects to experimentally
created conditions would automatically eliminate these alternative ex-
planations whether or not they had been (wnsidered beforehand. Ob-
viously, however, investigators are not going to inflict brain damage o~
humans in the interest of getting more certain knowledge about its
effects. Instead, they will concentrate on finding substitute methods for
ruling out the possible effects of other variables. _
Occasionally, natural si~uations may provide-both the desired con-
trasting conditions and the opportunity for sufficiently rigorous pro-
cedures to make possible a reasonably firm basis for inferences. For
example, suppose that a social scientist has an inside track on the
deliberations of a labor union concerned with some important service
to consumers. He learns that the 'union leaders consider a drawn-out
strike inevitable in the near future. He is interested in several hy-
potheses concerning the effects of suffering personal discomfort on
attitudes ,toward the person or group seen as responsible l for the dis-
comfort. He quickly takes advantage of the pending situation by
organizing a study in two comparable cities, one where the strike will
occur and one where it will not occur. As a result of his alertness, he
gets measures of attitudes toward labor unions, procedures for settling
industrial disputes, etc., in two comparable groups (i.e., the residents of
the two cities), one of which undergoes the critical experience and one
of which does not. Does it matter that he himself has not instigated
the strike? Clearly it does not, provided he can assume that the two
cities are really comparable before the strike and that the decision to
strike in the one city and not the other has not been influenced by
characteristics of the cities that might be related to their potential
reactions to the strike.
Ordinarily, however, natural situations are complicated and do
not permit the investigator to assume, with an'y confidence, that the
groups to be compared differ only by chance. As Greenwood (1945)
has stated:
... the created situation gives us better control power over
our p~omenon. We can determil1e at our own discretion the
circumstances which shall be present, and ,thus arrive at more
conclusive evidence of causality . . . . The ability to produce
the necessary changes permits the test of hypotheses otherwise
not amenable to verification.
No matter how carefully controlled the study, however, there is
no way to be completely certain of the validity of iIJferences that may
be drawn. There is always the possibility, for example, that later
research will reveal the influence of some factor not ta"i\en into account
in the study; or that the evidence collected was not critical to the
hypothesis under scrutiny (in other words, that the working defini-
tions did not provide adequate indicators of the ',concepts in the
hypothesis); or it may be that the research procedures did not meet
the logical requirements for the making of inferences. Certainly in the
social sciences, where there is little knowledge of what fac~ors s?~u~,d
be controlled, and where I
many of the relevant factors (such as in-
,I
dividual characteristics, or life events outside the. experimental situa-

tion) are difficult or impossible to control, these,. possibilities are ever
present; but the principle ap,plies to the physical sciences as well,
especially at the fronti(j~s of knowledge.
This possibility of fallacious inference makes it necessary to eval-
uate experimental findings in the context of other knowledge. Con-
fidence in a research result requires not only statistical evidence of its

reliability (for example, that the effects of X and X' are significantly
different statistically) 3 but, in addition, evidence that the interpreta-
tion is in keeping with the interpretations of other "facts" about which
one has considerable confidence. This is why the establishment of
confidence in the imputation of any causal relationship between events
requires repetition of research and the relating of the findings to
other research.
Causal Inference from Experiments
The basic outline of an experiment is simple: an "experimental"

group is exposed to the assumed causal (or independent) variable while
a "control" group is not; the two groups are then compared in _terms
of the assumed effect (or dependent variable). This pattern makes
possible the collection of the three major types of evidence relevant to
testing hypotheses about causal relation.ships: (I) evidence of COn
comitant variation-that is, that the causal variable and the dependent
variable are associated; (2) evidence that the dependent variable did
not occur before the causal variable; and (3) evidence ruling out other
factors as possible determining conditions of the dependent variable
Evidence of the first type-Tconcomitant variation-is provided
very simply in an experiment. The investigator knows which subjects
have been exposed to the assumed causal variable (X); h~ measures
all subjects in terms of the assumed dependent variable (Y). He
then determines whether Y occurs more frequently among the'
subjects who have been exposed to X than among those who have pot,
or whether those who have been exposed to X show a greater amount
of Y than those who have not, or:whateve~ specific relationship between
X and Y is predicted by his hypothesis. For example, suppo~e an in
vestigator is testing the hypothesis that the use of televised instructional
programs in eighth-grade general-science classes will lead to greater
gains in information than more conventional teaching methods. He
will arrange for the use of television in some classes (the experimental
3 For a discussion of the concept of statistical significance, see Chapter II, pages
H4-422.
CAUSAL INFERENCE FROM EXPERIMENTS 95
group, exposed to X, televised programs) but not in others (the control
group, not exposed to X). At the end of the semester he will compare
the scores of the experimental and control groups on a test of general-
science information. If children in the classes where television was
used score higher, on the average, than those in the regular classes, he
will have evidence that televised instructional programs and informa-
tion about the subject matter are related.
Evidence of the second type-that the assumed effect did not
occur before the assumed cause-is secured in one or both of two
ways: by setting up the experimental and control groups in such a
way that it is reasonable to assume that they did not differ in terms of
the dependent variable before exposure to the independent variable,
or by measuring their position on the dependent variable before ex-
posure to the independent variable. In our television example, the in-
vestigator might assign classes to the television and non-television
"treatments" in such a way that it was reasonable to assume that they
were comparable in general-science information before taking the
course; or he might administer a test of information at the beginning
of the course; or he might do both. (Various procedures that might be
llsed, and some of their advantages and limitations, will be discussed
later in this chapter.) Note that the requireme~t that X cannot be
considered a cause of Y if Y occurred before X does not mean, in the
case of a hypothesis such as the television one, that the students must
have had no prior general-science knowledge. Since the hypothesis
.ass~that the use of television will lead to acquisition of more in-
formation than conventional teaching methods, what is required is
that the experimental group should not have had more information
than the control groqp before the start of the course.
Evidence of thd third type-ruling out other factors as possible
determining conditions-may be secured in several ways. A variety of
J
factors need to be considered as possible determining conditions.
Among the major odes are: (1) factors that have occurred in the past
or are relatively enduring characteristics of the subjects, (2) con-
temporaneous events, other than exposure to the experimental variable,
(3) maturational or: developmental changes, and (4) the influence of
the measurement Wocess itself. Somewhat different procedures are
appropriate for ruling out each type of factor as a possible determining

condition.
1. In order to rule out factors that have occurred in the past or
are relatively enduring characteristics of the subjects, the investigator
may set up his experimental and control groups in such a way that it is
reasonable to assume that they do not differ systematically in terms of
characteristics or past experiences that might be relevant; or he may
measure them before the experiment in terms of such factors; or he
may do both. For example, in the case of the television hypothesis,.
another factor that immediately suggests itself as a possible determin-
ing condition of acquiring general-science information is intelligence.
In order to rule out difference in intelligence as the cause of a difference
in information between the experimental and the control group after
the experiment, the investigator will set up the two groups so that it
is reasonable to assume that they are comparable in IQ, or he will
administer inteIIigence tests before the experiment, or both. (Again,.
procedures will be discussed later.)
2. In attempting to rule out contemporaneous events other than.
exposure to the experimental variable as causes of differences in posi-
tion on the dependent variable, the investigator may arrange for certain
conditions to be the same in the experimental and the control groups,
or he may deliberately vary them within his experiment. He may,.
for example, make sure that the same subject matter is covered in the
classes with and without television instruction, that the teachers using:
the two methods are approximately equal in teaching ability, that the
courses involve the same amou.nt of time, etc. On the other hand,
he may deliberately select teachers of different degrees of competence ..
In this case, he will assign both some of the more skilIfuf and SOJ;Ile
of the less skillful to the television classes as well as to the classes il1l
which television is not used. Thus he is in a position to assess the
effects not only of television instruction but also of the teacher's
ability, and to determine whether televised programs are more helpful:
to relatively skillful or relatively unskillful teachers.
Some contemporaneous events that may affect the outcome of
experiments in' the social sciences cannot be controlled, either in the
sense of holding them constant or of deliberately manipulating them.
For example, a headline-making scientific event occurring during the
experiment may lead to an increase in information regardless of the
general-science course. If such an event affects the experimental and
the control group in the same way, no problem is created, since an
effect common to the two groups could not be a cause of differences.
between them in terms of the dependent variable. However, one ma)
not be justified in assuming that the event affects the two groups in
the same way; its impact may be different in the group exposed to the
experimental variable than in the control group. Complex designs-
discussed later in this chapter-have been devised to take account of
this possibility.
3. Normal changes associated with maturation and development
may also become confounded with the results of the experimental treat-
ment. For example, whether or not they are taking science courses, it
is possible that children growing up in a science-minded society such
as ours may show an increase in general-science knowledge during the
period of a year. Whenever an experiment extends over a long period
of time, processes of growth and development must be considered
as possible determining conditions of changes. Again, if maturational
processes can be assumed to be the same in the experimental and the
control groups, and if it can be assumed that the effect of the experi-
mental variable is not specific to a given stage of development, the
effects of maturation can be ruled out by comparing the two groups.
More complex designs may be used if there is reason to believe that
thes,e assumptions are not justified.
4A"ne measurement process used in the experiment may itself
affect the outcome. If people feel that they are "guinea pigs" being
ex'perimented with, or if they feel that they are being "tested" and must
make a good impression, or if the method of data collection suggests
responses or stimulafes an interest the subject did not previously feel,
the measuring process may distort the experimental results. Variations
in experimental design have been worked out to take account of the
effects of the measuring process; they will be discussed later in this
chapter.
The entire design of an experiment has the function of providing
for the collection of ~vidence in such a way that inferences of a causal
relationship between the independent and the dependent variables
can legitimately be drawn. However, certain aspects are especially im-
portant in this respect: the method of selecting experimental and

control groups; the points in time when the dependent variable is
measured, and the pattern of control groups used; and the number of
possible causal variables systematically included in the study. Each of
these is discussed in the following sections.4
SELECTING EXPERIMENTAL AND CONTROL GROUPS
In any design that involves comparing two or more groups of

subjects who have been exposed to different experimental treatments,
there is an underlying assumption that the groups being compared
are equivalent before the introduction of the treatments. However,
the investigator cannot simply make this assumption; he must take
steps to see that it is met.
Clearly, the task of creating or of unearthing groups that are
equivalent in all respects is an impossible one. Before considering how
one gets around this problem, it is necessary to distinguish two dif-
ferent reasons for wanting equivalent e~perimental and control groups.
The first is to provide a basis for inferring that differences which may
be found on the dependent variable do not result from initial dif-
ferences between the two groups, in terms either of position on thl;
dependent variable or of other factors. The goal here is to ensure, as
far as possible, the validity of the inferences made on the basis of the
experiment. But there is a second goal, that of increasing the sensitiv-
ity of the experiment-i.e., increasing its ability to register small
effects of the experimental treatment that might be obscured by the
effects of other factors. -
These two goals call for somewhat different procedures in estab-
lishing the equivalence of groups. The goal of protecting the validity of
the experiment by ensuring that experimental and control groups differ
initially only by chance is achieved by procedures termed rf}ndomiza-
tion. The goal of increasing the sensitivity of the experiment, so that
the effects of the causal variable will be apparent even if they are
relatively small or if there are relatively few subjects, is achieved by
4 For a fuller discussion of factors that may affect the outcome of an experi-
ment, and of the ways in which different experimental designs deal with them,
see Campbell (1957).
matching procedures. These two types of procedure are discussed more
fully below. It is important to realize that all well-conducted experi-
ments involve randomization, since experimental inference presup-
poses randomization; hence, whenever matching procedures are used,
randomization procedures should, ideally, also be employed. It is not
necessary, however, that randomization be supplemented by matching,
although this is often desirable.
RANDOMIZATION. Randomization provides the basic safeguard
against differences between experimental and control groups that
might lessen the validity of inferences about the effects of the ex-
perimental treatment. Random assignment of members of a pool
of subjects to experimental and control groups entails the same
principles and. procedures as those involved in selecting a simple
random sample for a descriptive study (see Appendix B). The assign-
ment procedures must give each subject the same chance as that of
any other subject of being assigned to any given condition. The pro-
cedures are such that any selection based, either consciously or un-
consciously, on the investigator's judgment is ruled out. For example,
one may flip a coin for each subject, assigning him to the experimental
group if it comes up "heads," to the control group if it comes up
"tails." Or one may number each person and then, by using a table
. /
of random numbers (see AppendIx B), select as many cases as are
wanted for the experimental group and assign the remaining ones to
the~nt~ol groUp.5 The chance assignment of individuals to the dif-
ferent conditions precludes the possibility of systematic or non-
random differences between the groups selected. This does not m~n
that the experimental and control groups will be exactly alike, but
rather that whatever differences exist before the introduction of the
t
experimental variable are the result of chance alone. The rules of

probability make it possible to specify the extent of differences that
might be expected by chance in the long run (that is, if the selection
were repeated a large number of times). If, after one group has been
exposed to the experimental treatment, the two groups are found to
fi [n most experlments, groups of equal size are used. However, when the cost
of introducing the eXperimental variable is very high, it may be more economical
to llse more cases in the control than in the experimental group. See McNemar
( 1940) for a more d~tailed discussion.
100 RESEARCH DESIGN: r1
differ more than would be expected by chance, one may infer that the
experimental variable led to the difference. This inference, of course,
must always be made tentatively, subject to the possibility that some
other factor may have led to the difference.
R. A. Fisher (1951), one of the outstanding figures in the develop-
ment of experimental design, has pointed out that: "
... the uncontrolled causes which may influence the result
[of an experiment] are always strictly innumerable. When any
such cause is named, it is usually perceived that, by increased
labour and expense, it could be largely eliminated. Too fre-
quently it is assumed that such refinements constitute improve-
ments to the experiment. ... whatever degree of care and ex-
perimental skill is expended in equalising the conditions, other
than the one under test, which are liable to affect the result,
this equalisation must always be to a greater or less extent incom-
plete, and in many important practical cases will certainly be
grossly defective . . . . the simple precaution of randomisation
will suffice to guarantee the validity-of the test of significance, by
which the result of the experiment is to be judged.
To go back to our television-teaching illustration: Let us say that
our subjects are to be all the eighth-grade children in a given schooi,
half of whom (the experimental group) will be assigned to a class in
which television will be used, half( the control group) to a class using
the conventional methods. But the children will certainly not all have
the same IQ; some of them m,ay already have more"science information
than others; some will be more interested than others in the subject
matter; some will have better eyesight than others; etc. Frqm the
point of view of the validity of the inferences to be drawn, it is neces-
sary that the experimental and the control groups shall ,not differ
on any of these variables to such an extent that it leads to a difference
in science information, as meas~red at the conclusion of the experi-
ment, which will be incorrectly interpreted as resulting from the dif-
ference in teaching methods. Since all tests of statistical significance
are based on the assumption that cases have been randomly assigned
to the groups being compared, they are specifically d~signed to take
account of chance differences in the initial characteristics of the two
groups. Therefore the statistical test of significance6 offers protection
against the possibility that differences on the dependent variable that
result from chance initial differences between the experimental and
the control group will be incorrectly interpreted as effects of the ex-
perimental treatment.
The social scientist, however, is not always in a position to assign
cases randomly to different conditions. Compromises with the ideal
of random selection are often necessitated by practical circumstances.
In our television-teaching example, for instance, it may not be feasible
to select randomly from among all eighth-grade children those who are
to be~ssigned to the experimental class. In order not to disrupt school
routines, it may 'be necessary to assign existing classes to one or the
other treatment. In this case, the classes may be randomly assigned
to one or the other treatment, but this does not afford as much protec-
tion as the random assignment of individuals. Sometimes such com-
promises may be made without invalidating the bases for inference
within the study (though, in terms of our definition, a study in which
cases are not randomly selected does not constitute an experiment).
One extreme form of nonrandom assignment, however, does seriously
impair the .grounds for inference. This is assignment on the basis of
self-selection. For example, if an investigator wishes
/
to test the hy-
pothesis that social case work with the families of delinquent children
red?s the delinquent behavior, he would be ill advised to draw his
experimental sample from families who have voluntarily come to
social-agencies and his control sample from families with similarly
delinquent children who have not sought such help. The reaSOn is
dbvious: Families who seek help of this sort may have certain char-
acteristics that either directly affect the probability that the delinquent
behavior would bel roouced even without the case work service or that
make the service effective with them although it would not be with
other families. We might suppose, for instance, that a mother who
J
seeks the help of a social agency in dealing with her child's delinquency
is both more concerned about the delinquent behavior and more aware
6 For a brief dis~ussion of the meaning of statistical tests of significance, see
Chapter 11, pages 414-422. For a fuller discussion, consult any standard statistical
text.
of community facilities for dealing with it than a mother who does not.
Either of these characteristics might mean that she would be likely
to take steps intended to change her child's behavior even if she did
not have the help of a social case worker. And the fact that she applied
for case work help mjght mean that she would be more receptive to
it, and thus that it would be more likely to have an effect on the
delinquent behavior, than if she had been assigned involuntarily to
receive such help. The same principle applies whenever subjects place
themselves in the "experimental" or the "control" group.7
MATCHING. Although random assignment, where it is feasible, is
generally considered to provide adequate protection agaInst interpret-
ing differences on the dependent variable as resulting from the in-
dependent variable when in fact they stem from prior differences be-
tween the two groups, it is not the most effective procedure from the
point of view of increasing the sensitivity of the experiment., In the
interest of research efficiency, it is desirable that the experiment reveal
true differences brought about by the experimental treatment, even if
they are small in relation to differenc(ls produced by other variables.
In our t~levision example, teaching method may have less influence
than IQ on science information. Random assignment of children to
groups being taught by one or the other method would not be likely
to lead to exact matching of the two groups in terms of IQ. 'Bhis dif-
ference in IQ might lead to a difference in inforination at the end
of the study. As already noted, statistical tests of significance based
on the assumption of random sampling would provide protection
against attributing this difference to the difference in t,eaching
methods. However, there might be a small difference in the effective-
ness of the two teaching methods which would be obscured by the
difference in information related to IQ. The more such "extraneous"
differences are reduced, the more ch~nce there is for the effects of the
experimental treatment to show up.
An oversimplified hypothetical example may help to make this
point dear. Suppose, in our study of the effects of teaching ~cience with
7 This is a frequent problem in studies designed to test causal !hypotheses that
do not follow the pattern of controlled experiments. Methods of dealing with it are
considered in the section of this chapter which discusses such studies.
the help of television, we took existing classes rather than individual
students as our sampling units. Suppose further that eight classes were
to be used in the experiment-four to receive the televised instruction,
four to serve as controls. Let us say these classes differed in average
IQ; four had a mean IQ of over 100 (these will be called the "highs"
in the table below), and four of under 100. If the classes were ran-
domly assigned to the television or no-television treatments, we might
have a pattern such as that shown below, with grades on the final test
shown in the right-hand column.
Class Television Mean IQ Mean Grade on Test

1 Yes High 90
2 No High 80
3 Yes Low 70
4 No High 80
5 Yes Low 70
6 No Low 60
7 Yes Low 70
8 No High 80
In this example, the mean score of all classes on the final test is
75. The mean score of the high-IQ classes is 82.5; that of the low-IQ
classes, 67.5. But both those with television and those without have
mean sco~ of 75, even though it is apparent from inspection of the
table tliarhigh-IQ classes with television instruction score higher than
high-IQ classes without, and that low-IQ classes with television instruc-
tion Score higher than low-IQ classes without it. But the fact that
random assignment has led to an arrangement whereby three of the
four classes receiving tel~vision instruction are of low IQ, while three
of the four without te1 vision are of high IQ, obscures the effect of
1
television when the average score of aU classes receiving the experi-
mental treatment is compared with that of all the control classes.
To illustrate the eff~ct of matching, let us suppose that the tele-
vision and no-television treatments had had equal numbers of high-
and low-IQ classes. Making the same assumptions as in the preceding
\
table about the relative contribution of intelligence and teaching
method to scores on the ;final test, the results would be as follows:
Television Mean IQ Mean Grade on Test

Yes Low 70
Yes High 90
Yes Low 70
Yes High 90
No Low 60
No High 80
No Low 60
No High 80
Now the mean score of the classes with television is 80; that of
those without television is 70. By equating the groups in terms of in-
telligence, the effects of teaching method have been permitted to
appear.
It should be noted that in the matching procedure it is impmtant
not to sacrifice randomization. In our example, randomization might
have been incorporated in the procedure in a number of ways; for ex-
ample, by tossing coins to determine which two of the four high-IQ
classes, and which two of the low-IQ classes, should receive the tele-
vision treatment. Or, if there were a large number of classes from which
to select, they might be divided into two groups-high- and low-IQ's;
then, by means of a table of random numbers, two classes from each
group might be selected for the television treatment and two from
each group to serve as controls.
The fact that matching may make an experime~t mor<;: sensitive
by controlling the effects of oth~r variables which might obscure that
of the variable in which the investigator is interested often leads ex-
perimenters to supplement randomization by matching procedures.
Two methods are commonly used: precision control and trequency
distribution controJ.8 Both, when combined-as they should be-with
randomization procedures, are methods of stratified random sa,mpling
(see Appendix B) .
The equating of groups by precision control involves matching the
individuals in the groups, case by case. To take a complicated' problem
-suppose we wish to determine the effect of psychoanalytjc therapy
of a certain sort upon the attitudes of prejudiced people. We would try
8 For a more detailed discussion of these methods of matching, as well as of the
method of random assignment, see Greenwood (1945).
to set up two groups of persons who are matched, individual for in-
dividual, in attitudes and in factors that might be relevant to their
predisposition to attitude change. That is, for person A who is highly
prejudiced, who is exposed to pressures from his social group to be
prejudiced, who is intelligent, who has no, strong unconscious needs
that motivate his prejudices, etc., we would try to find an exact coun-
terpart, N. A would be assigned to one group and N to the other.
For B, who is moderately prejudiced, who is exposed to social pressures
not to be prejudiced, who is of average' intelligence, and who has
underlying insecurities that find an outlet in his prejudices, etc., we
would try to find a B'. And so on until for every individual in the ex-
perimental group we had a matched individual in the control group.
The matching of individuals is obviously a very difficult task for
several reasons. First, jf matching is to be precise and if individuals are
to be matched on several factors, there must be..a large number of cases
to select from ,in order to achieve an adequate pairing. AlI of these
cases' have to be measured in the relevant fadors, but only a few will
be us~d. The more precise the matching, and the.greater the number
of factors on which matching is to take place! the greater the number
of cases for which no match is available. Secondly, it is frequently dif-
ficult to know which factors, of the many possible relevant ones,
are the most ii'nportant to use in obtaining precision control. Matching
on more than two or three factors with any degree of precision is rarely
possible. Fortunately, however, relevant factors are ,often so interrelated
that matching on one factor brings with it partial matching on other
factors; there is a "diminishing return" as additional factors are con-
trolled. Third, it is often difficult to obtain adequate measures of the
Ifactors
I
on which it may be important
,
to match; consider, for instance,
lour suggested experiment on the effects of psychoanalysis. 1 no
adequate measures of the ass~med relevant factors are available, then
obviously matching is not like~y to be very accurate.
Successful matching .can greatly increase the efficiency of an
experiment by decreasing the I.size of the differences on the dependent
variable that would occur between the experimental and the control
groups by chance alone. When the chance differences are small, it is
easier to demonstrate a difference that is due to the effect of the ex-
perimental variable. If one is able to match the members of each

pair in terms of initial position on the dependent variable or on some
attribute that is highly correlated with it, one would need considerably
fewer cases to be confident that an obtained difference was not due to
chance than would be required with unmatched groups. However, if
the matching is in terms of some other characteristic that in fact has
no relation to the dependent variable, then the matching procedure,
no matter how elaborate, provides no experimental or statistical
advantage. Suppose, in our effect-of-therapy-on-attitudes example, in-
stead of being able to match cases on all the characteristics we sug-
gested, we were able to match them only on intelligence. If intelli-
gence were unrelated to prejudice, we would not have succeeded in
matching cases for their initial degree of prejudice, and thus we would
not have eliminated such initial differences as one of the causes of the
differences observed after therapy. In such a situation, not only dQes
matching provide no advantage; for reasons~ that are too technical to
discuss here, unsuccessful matching may even lead to a statistical dis-
advantage.9
As has already been emphasized, matching is not a substitute for
random assignment, but a supplement to it. Matching procedures
can take account of only a few variables; those that are unaccounted for
should be randomly distributed between the experimental and the
control groups. Thus, if our matching procedure in the therapy ex-
periment has given us a number of pairs of matched individuals (A and
N, Band B', C and C', etc.), we must use some random procedure for
determining whether A or A' is to go into the experimental group,
whether B or B' is to go into that group, etc. '
One further point should be emphasized: If one is experimenting
with social groups that have some psychological unity, if the interest
is in a functioning collective (a dub, a factory, a class) rather than in
the separate indivKluals, it is appropriate to match group with !group
rather than individual with individual. The interactions of ,the in-
dividuals, their attitudes toward one another, their complementary
I skills and interests-all of these require that, if groups are the object of
study, groups rather than individuals be matched (see Deutsch, 1949).
9 For a detailed discussion, see Edwards (1950, Chapter 14.)
Frequency distribution control is an attempt to reap some of the
advantages of matching without paying the price of losing as many
cases as one usually does in precision control. As the name implies,
frequency distribution control attempts to match an experimental and
a control group in terms of the over-all distribution of a given factor
or factors within the two groups rather than individual by individual.
For example, if age were considered relevant to the effects being
studied, one would attempt to see that the average ages in the two
groups were alike and that the distributions of ages in the two groups
were similar. For each of the other factors considered relevant, one
would try to equate the distributions in the two groups. It may be
necessary to eliminate cases to equate the distributions, but compared
to the method of matching individual by individual, relatively few
cases are lost.
Matching by distribution rather than by precision control has a
potential disadvantage of some importance. Although distributions on
single factors are equated, the groups may actually be badly mis-
matched on combinations of these factors. Suppose, for example, that
we were matching two groups in terms of income and age. Despite
the fact that the distributions of income and of age separately were
equated in the two groups, it would be possible to have in one group
young people with high income and older people with low income
and in t~ other group just the opposite combination:
Two~ther points must be noted in considering whether matching
by frequency distribution has in fact yielded equated groups. First,
one' should not assume that the distributions in two groups are
si~ilar simply because their averages are similar. This may not be
the case; one group may range widely about the average, the other
may have a narrow range of variation. These differences will, of course,
affect research result~. In addition to the average, other statistical
measures (such as ~he standard deviation, skewness, kurtosis-see
Chapter 11, pages 412-413) which provide an indication of the nature
of the distribution of leases are relevant to the equating of distributions.
Second, even though a statistical test indicates that the two distribu-
tions do not differ significantly (i.e., that the two samples could have
been drawn from the same population,). one is not ne.::essarily iustified
in concluding that they are equivalent. It is a logical fallacy to assume

that if you have not conclusively demonstrated things to be unequal,
you have demonstrated them to be equal. The differences between the
two groups should be as small as it is possible to make th~m with the
available resources. Again, it should be remembered that, ideally,
randomization should always be used to equate the unmatched factors
whenever precision or frequency distribution control is employed.
TYPES OF EXPERIMENTAL DESIGN
If one wishes to test the hypothesis that X is a cause of Y, by

comparing a group that has been exposed to X with one that has not,
it is obviously essential to measure the two groups with respect to Y
either during or after exposure to X. Sometimes it is desirable or even
essential to have, in addition, measures of their position with respect to
Y before they have been exposed to X. The time at which the depend-
ent variable is measured provides a basis for classifying experiments in
two main groupings: "after-only" and "before-after." The pattern of
control groups used provides a basis for further subdivision.
"AFTER-ONLY" EXPERIMENTS. In studies of this type, as the name
implies, the experimental and the control groups are observed or meas-
ured with respect to the dependent variable (Y) only during or after
the exposure of the experimental group to the assumed causal variable
(X) .10 As in all experimental designs involving control groups, the
experimental and the control groups are selected before the introduc-
10 Usually the "control group" is exposed only to the "absence of X" rather
than to some contrasting variable, X', whose influence is to be compared with X.
However, it is possible to expose the "control group" to X' and thus 'to have it
constitute a second experimental group. In such a case, it is sometimes feasible to
use only one group of subjects, who "serve as their own controls." For example,
in testing the hypothesis that people will tend to remember unfinished tasks better
than finished tasks, one group of subjects might be allowed to finish all their tasks
and another group might be prevented from finishing all their tasks. Here, one
would be using two groups of subjects--one of which is exposed to X ("finished
tasks"), another to X' ("unfinished tasks"). It is possible to test the' same hypoth
esis using one, group of subjects, all of whom would be - exposed to both X
and X', that is, they would be allowed to finish some tasks and not others. In doing
so, it is necessary to consider carefully the sequence of finished and unfinished
tasks, in order to avoid complications which might arise from such factors as a
tendency to remember the first task, or the last task, etc.
tion of the experimental variable (X), and the variable is introduced
either specifically for the purpose of the experiment or at a specified
time and in a specified manner known in advance to the experimenter.
As discussed in the preceding section, subjects are randomly assigned
to the two groups, with or without supplementary matching pro-
cedures.u If one of the matching techniques is used, the matching is
in terms of characteristics other than initial position on the assumed
dependent variable (Y), since this is not measured before the experi-
mental treatment. The two groups are observed or measured with
respect to Y only during or after the exposure of the experimental
group to the assumed causal variable, X. Column 1 of the table on
page 110 shows a schematic representation of this type of study.
In this design, one concludes that the hypothesis is or is not
tenable simply by comparing the occurrence of Y (or its extent, or its
nature, etc.) in the experimental group after exposure to variable X,
with the occurrence (etc.) of Y in the control group, which has not
been exposed to X. Thus, in Column 1, we would compare Y2 with
Y'2. Clearly, this provides evidence as to whether X and Y vary
together.
What about evidence that Y did not precede X in time? This is
inferred from the method used in setting up the two groups. If they
were randomly selected, prob~bility theory telll us to what extent
they might have been expected to differ by chance in terms of their
position on Y before the introduction of the experimental treatment,
and the test of significance takes account of such chance differences. If,
in' addition, they have been matched in terms of some variable or
variables that are related to position on Y, the probable initial dif-
ferences on Y will be even less than might otherwise have been ex-
pected by chance. 1pe randomization procedures used in setting up
the groups (whether: or not they are supplemented by matching) also
provide a basis for cbnduding that differences between the groups in
other past factors or! relatively enduring characteristics that might in-
fluence their position on Y after the experimental treatment are no
I
11 Studies in whic]:l' groups that have not been randomly selected are exposed
to different conditions and then compared do not, strictly speaking, constitute
experiments. Such studies are discussed later in this chapter.
_-... -
0-
...... 0...
t:: ::l
00
CJ ...0.0
o
Z
,'"<
s...
.-. I ~
...C'l ~ ~
...
s... s...
\3 \3 II ~
><><~~
s: ~
I ~
,01 ,'" I
e.;:... "
o (l) ~ II
'"
~z
0
.-< Z >< >< ~
s...
I
.,'" '".,
>< ><
.-<
s...
~ +<'1
~I
~
~~
",~II II
., ., ~ 0 .-i
><><'"0 >< Z s... ~
8~
..... ::l
ot:: 0...
CJo.o o
'" 0 o Z
"Z
><# Z
.,
'"
><
CA USAI. INFERENCE FROM EXPERIMENTS 111
greater than the chance differences taken account of by the test of
significance.
Although this design shares a problem of all social research, that
the measurement procedures used may alter the characteristic they are
intended to measure, the problem is less serious here than in "before-
after" studies. (The difficulties introduced by measurements made
before exposure to the experimental variable will be discussed later,
in connection with "before-after" designs.)
What about the effects of other contemporaneous events or ma-
turation? The assumption is made that both groups are exposed to the
same external events and undergo similar maturational processes be-
tween the time of selection and the time at which Y is measured. If
this assumption is justified, the position of the control group on the
dependent variable (Y'2) at the close of the experiment includes the
influence of the external events and maturational processes that have
affected both groups. Thus the difference (d) between Y2 and Y' 2 may
be taken as an indication of the effect of the experimental treatment,
provided that neither external events nor maturational processes inter-
act with tbe experimental variable to cbange its effects. (The possibil-
ity of interaction between the experimental variable and other factors
will be discussed later. )
a
The "after-only" design may be illustrated by study of the effects
of a film, Tbe Battle of Britain, carried out by the Experimental Sec-
tion of the Research Branch in the War Department's Information
and Education Division (Hovland, Lumsdaine, and Sheffield, 1949).
In this study, the experimental group was shown the film, the control
group was not. In assigning men to the two groups, random selection
of individuals did not seem feasible, since this would have required
pulling specified meri out of their regular units to see the film. Such
an unusual procedur~ would not only have ma<:1e for administrative
difficultie::; but would presumably have raised questions in the men's
minds about the purpose of the operation. Therefore selection of cases
for the sample was on the basis of company units rather than individ
uals. In view of evidence that companies differed in certain ways, two
groups of companies }vere set up which were comparable in character
istics such as average:score on the Army General Classification Test
education, age, region of birth, stage of training, etc. Randomization
took the form of tossing a coin to decide which of the matched groups
should see the film, which should not. 12
The experimental group was shown the film during their weekly
orientation hour, as part of the regular training procedure. The control
group did not see the film. Approximately a week later, the men in both
groups were asked to fill out a questionnaire as part of a War Depart-
ment survey "to find out how a cross-section of soldiers felt about
various subjects connected with the war." Mixed in with "camouflage"
items were a number of factual and opinion items that might have
been expected to be influenced by the film but were not so specifical~y
related to it as to suggest a connection between the "survey" and the
film. The measure of the effects of the film was the difference (in
excess of chance difference) between the proportion of the experi-
mental and the control groups who responded to each of the relevant
items in a given way.
"BEFORE-AFTER" EXPERIMENTS. In addition to the measures of Y
after exposure to the experimental variable, an investigator may wish
to have measures of Y before such exposure, for a variety of reasons:
1. As discussed in the section on selecting experimental and con-
trol groups, he may wish to increase the sensitivity of his experiment
by matching cases in terms of their initial position on the dependent
variable. As pointed out in that section, such matching should be
accompanied by procedures of random assignment.
2. He may want to check whether there is "room" for the experi-
mental variable to have an effect. For example, suppose one were
studying the effectiveness of an advertising campa~gn to induce women
to use perfume. If, by chance, one selected an experimental and a
control group of women such that 100 per cent of them were already
using perfume before the onset of the advertising campaign,_ it would
I
12This procedure of esta1:>lishing two matched groups and making. use of

randomization only to determine which one should be exposed to the experimental
treatment is, of course, a rather marked departure from the recommended procedure
of deciding on a random basis which individual or which unit of each matched pair
should be assigned to the experimental group. Strictly speaking, an estimate of
the probable chance differences between the two groups which is arrived at on the
assumption that individuals have been randomly assigned does not proyiae an ap-
propriate estimate of the probable chance differences between two groups set up in
this way. This fact was recognized by the investigators. The study illustrates the
point that realistic requirements sometimes necessitate compromls~ with ideal
procedures.
be impossible to determine the effectiveness of the advertising, since
a "ceiling effect" would be operating. A "before" measure makes it
possible to determine the possibility for change in the dependent
variable and to take it into consideration in evaluating the effects of
the experimental variableP
3. The hypothesis with which the study is concerned may specify
initial position on the dependent variable as one of the determining
conditions, or as one of the conditions on which the effect of some
other variable is contingent. For example, the hypothesis may state
that a film carrying a message of interracial friendliness will have a
greater effect on persons who are already neutral or favorable toward
Negroes than on those who are prejudiced. In such a case, an initial
measure of prejudice, as well as another measure after the film, is
obviously required by the hypothesis.
4. Even though the major hypothesis of the study does not refer
to initial position on the dependent variable, the investigator may wish
to analyze the data in such a way as to see whether the experimental
treatment has different effects on persons who were initially at differ-
ent positions on the dependent variable. If he is to do this, he must
of course have a measure of position on the dependent variable before
subjects are exposed to the experimental treatment.
5. In the event that he has not been able to ilssign subjects on a
completely random basis to the different conditions, he may wish to
check their initial comparability. As pointed out earlier, failure to
assign subjects to experimental and control conditions on a completely
random basis is a departure from the requirements of an ideal experi-
ment, but compromises in this respect are sometimes necessary, espe-
cially in studies carried out in real-life settings. In such cases, evidence
from a "before" measure that the experimental and the control groups
were initially equated on the dependent variable helps to increase
confidence that a difference on the "after" measure is due to the effects
of the experimentaljvariable. However, in the absence of randomiza-
tion, the possibility remains that the two groups may have differed in
other characteristics or past experiences that may interact with the
experimental variable to lead to final differences on the dependent
13 See Hovland, Lumsdaine, and Sheffield (1949) for a discussion of the prob-
lem of "ceiling effects" and proposed solutions.
variable. Thus evidence that the two groups were initially comparable
in their position on the dependent variable is only a partial substitute
for random assignment.
Studies using "before" as well as "after" measures of position on
the dependent variable may follow various arrangements with respect
to control groups. (1) Only one group may be used in the study, with
the "before" measure serving as a "control" in the sense that it is
assumed to represent the level of the dependent variable in the absence
of the experimental treatment. (2) The "before" measure may be taken
on one group and the "after"'measure on a different but presumably
equivalent group. (3) "Before" and "after" measures may be taken
both on the experimental group and on one control group. (4) There
may be two or more control groups. These four patterns will be dis-
cussed below.
Whatever the pattern of control groups, the "before-after" e~peri
I1lent, like the "after-only;'! 'provides evidence of concomitant varia-
tiQn between the independent and'the dependent variables by com-
paring the occurrence (or the extent, or th.e increase) of Y in the group
that has been exposed to X with the occurrence (etc.) of Y in the
group that has not been exposed to X. That Y did not occur before X
is inferred from the assurance provided by randomization that the
groups are not likely to have differed initially in their position on Y
by more than the specified chance amount taken account of in the
test of significance. This initial equivalence with respect to Y may be
checked by comparison of the "before" measures of the hyo groups.
In the everit that random assignment has not been possible, the "be-
fore" measures still provide evidence of whether there were differences
in Y that preceded differences in X. But if random assignment has not
been possible, there is no basis for ruling out the possibility that there
were greater-than-chance differences on ()ther factors that might, ac-
count for a difference in position on Y after exposure to the experi-
mental treatment.
The variations in control group arrangements are concerned with
attempts to take account of contemporaneous events, maturational
processes, and the effects of the initial measurement. Although the
measuring process itself may affect the characteristic being measured
in any type of social research, the' "before-after" desi~n is especially
subject to this difficulty. For example, the attempt to measure the
subjects' attitudes before the experiment begins may crystallize the
attitudes; it may exhaust the good will of the subjects; etc. The second,
or "after," measurement may introduce other problems: the subject
may be bored and therefore unwilling to respond; he may try to give
responses that are consistent with his previous responses (thus mini-
mizing the apparent change); or he may try to make his responses
"interesting" by varying them from one interview to the next (thus
increasing the apparent change). The process of repeated measurement
may also affect the "measuring instrument"; for example, in the course
of repeated measurements, an observer may become bored, fatigued,
more sensitive or less sensitive to the phenomena he is recording.
The different control group arrangements differ in the extent of
protection they offer against mistakenly attributing to the independent
variable differences on the dependent variable that may really be due
to other contemporaneous events, to maturation, or to the effects of
the initial measurement
The "before-after" study with a single group. Barker, Dembo, and
Lewin (1941), in their study of the effects of frustration on young
children's play, used a "before-after" design without a control group.
Each child was taken into a room where there were simple toys with
which he was allowed to play for half an houri" during this time his
play was rated by an observer on a scale of "constructiveness." Next
a partition was raised; in the part of the room now exposed was an
elaborate and attractive set of toys. When the child had become
thoroughly involved in playing with these, the experimenter took him
by the hand, led him back to the part of the room in which he had
been playing earlier, and locked the new toys behind a wire-net parti-
tion through whichl the child could still see them. The child's play with
the original toys 'Vas again rated for constructiveness during a half-
hour period. The qifference in ratings of constructiveness of play dur-
ing the "pre-frustration" and "post-frustration" periods was taken as
evidence of the amount of regression 'induced by the frustrating ex-
perience. (At the end of the experiment, the child was allowed to play
as long as he wished with the more attractive toys, in order to undo
I, .
the frustrating effects of the expenment).
In this design~ each subject "serves as his own control." The differ-
ence between his position on the dependent variable before and aftel
exposure to the independent variable is taken as a measure of thE
effect of the independent variable. (See Column 2 of table on page
110). But other influences may have operated between the "be-
fore" and "after" measures. External events unrelated to the experi-
mental treatment may lead to a change in position on the dependent
variable; so may processes of growth and development. The initial
measurement itself (in this case, the period of play on which the initial
rating was made) may lead to changes. This design does not make it
possible to separate such effects from those of the experimental treat-
ment. Thus its use is justified only when one has good reason to believe
(as did Barker, Dembo, and Lewin): (1) that the "before" measure
itself will not in some way affect either the response to the experimental
treatment or the "after" measure; and (2) that there are not likely
to be any other influences, besides the experimental treatment, dpring
the course of the study that might affect the subjects' response at the
time of the second measurement. In order to be reasonably sure that
such assumptions are justified, one must !lave considerable knowledge
of the probable effects of his measurements and of the conditions
other than the experimental treatment that are likely to influence the
dependent variable. This may be true of many problems in such fields
as learning and sensory perception, where much experimental work
has been done; it is much less likely to be so, at the present time, in
social psychology and sociology.
The "before-after" study with interchangeable groupS.14 One ap-
proach to ruling out the effects pf the initial measurement is to meas-
ure one group before the introduction of the experimental fa~tor, a
different group after exposure to the experimental factor~ The two
groups are selected in advance from the population that is to be ex-
posed to the experimental variable; as in other designs, random selec-
tion provides assurance that the groups-probably did not differ by more
than a specifiable amount before introduction of the experimental vari-
able, and thus that they may be treated as interchangeable. Again,
matching may be used to supplement randomization. The .difference
14 In the earlier edition of this book, this was called the "simulated before-after"
design. D. T. Campbell, in a forthcoming paper ("Quasi Experil1\ental Designs
for Use in Social Science Settings"), refers to it as the "offset before and after"
design.
between the "before" measure taken on the first group (Y'l) and the
"after" measure taken on the second group (Y2 ) is assumed to be a
measure of the effect of the experimental factor. (See Column 3 of
the table on page llO).
This design was used in a study of a publicity campaign about the
United Nations in the city of Cincinnati (Star and Hughes, 1950).
Two equivalent samples, of a thousand persons each, were drawn from
the city's population. One was interviewed before the start of the
publicity campaign, the other two months later. To determine the
effectiveness of the campaign, the responses of the two groups were
compared. As it turned out, there was very little difference between
them.
The "before-after" study with interchangeable groups eliminates
the possibility of -confounding an effect of the initial measurement
with that of the experimental variable. Suppose the same group of
respondents had been interviewed before and after the campaign. The
initial interview might have aroused their interest in the United Na-
tions and thus made them especially sensitive to the publicity cam-
paign. If this were true, a simple "before-after" study, in which the
difference between the "before" and "after" responses of a single re-
interviewed group was taken as the measure of the effect of the experi-
/
mental variable (the publicity campaign), would have been mislead-

ing.
On the other hand, the "before-after" study with interchangeable
groups provides no protection 'against the effect of other events that
may occur between the two measurements. Suppose a difference had
been found between the "before" response of the one group and the
"after" response of th~ other; suppose also that world events in which
the U.N. played a p'rominent role had occurred between the two
measures. This design Iwould give no way of determining whether the
change in response was due to the publicity campaign, to the world
events, or to a combination of the two.
The "betore-atter'~' study with one control group. In an attempt to
take account of the effects both of the initial measurement and of con-
temporaneous factors,La control group is frequently included in the
"before-after" design. This design is shown in Column 4 of the table on
page llO. In such a study, both the experimental and the control group
are measured at the beginning and at the end of the experimental

period. The experimental variable, of course, is introduced in the ex
perimental group only. Since the control group as well as the experi-
mental group is subjected both to the initial measurement and to the
contemporaneous influences, the difference between the scores of the
two groups should constitute a measure of the effectiveness of the
experimental variable alone. For this purpose, either the final scores
of the two groups (Y 2 and Y' 2)' or their change scores (d and d'),
may be compared.
Another study of the effects of the Battle of Britain film, parallel
to the "after-only" study described on pages Ill-Il2, provides an
example of a "before-after" study with one control group. In a different
camp, the effects of the film were tested by this design. The procedure
was the same as that already described, except that about two weeks
before the film was shown to the experimental group, members of,.both
groups were asked to fill out the questionnaire as part of " a survey."
Then the film was shown to the experimental group in the course"of
its regular training procedures; it was not,.of course, shown to the con-
trol group. About a week after the showing of the film, the men in both
groups were again asked to fill out a questionnaire, which repeated the
essential items of the "before" measure, bllt which was presented as a
"revised" version of the survey instrument that had been changed on
the basis of the earlier administration. To support this explanation,
the format had been changed, and some "camouflage" items omitted
while others were added. The differences between the "aft~r" replies
of the experimental and the control groups were taken as the measure
of the effects of the film.
If the "before" measure and the events other than the experi-
mental treatment affect experimental and, control groups in the same
way, a "before-after" design witli one control group provides aqequate
safeguards against attributing to the experimental variable a difference
on the "after" measure that is really due either to the effects of initial
measurement or to contemporaneous events. However, it may happen
that either the initial measurement or external events interact with
the experimental variable in such a way that its effect is changed. For
example, comparison of the two Battle of Britain studies. showed an
interaction between the initial measurement and the experimental
variable; there was less difference between the "after" replies of the
experimental and the control groups in the "before-after" than in the'
"after-only" study. The investigators suggested that an effort to be
consistent in replies to the two versions of the survey questionnaire
might have lessened the changes in opinion that might otherwise have
been produced by the film, or that the explicit taking of one position or
another on the first questionnaire might have tended to crystallize
views that would otherwise have been more fluid.
When there is such an interaction between the initial measure OJ
contemporaneous events and the experimental variable, a "before,
after" study with one control group does not provide a firm basis fOl
inferences about the effect of the experimental variable, since this
design does not provide any way of discovering the interaction and
thus separating out the effect of the experimental variable alone. Solo-
mon (1949) devised more elaborate designs that make it possible
to take account of such interactions. They involve the use of additional
control groups.
The "before-after" study with' two control groups. This design, .
shown in Column 5 of the table, makes it possible to separate the in-
fluence of the experimental variable from that of the initial measure-
ment, even if there is interaction between them. It involves adding a
second control group, which is not pre-measurec,l, but which is exposed
to the experill.1ental variable and to an "after" measurement. As in
the "before-after" design with one control group, the first control
group is given the "before" measure and the "after" measure, but is
not exposed to the experimental treatment. The three groups (the
experimental and the two control) should, of course, be selected in
such a way that they differ only by chance at the beginning of the
experiment. As in; other experimental designs, this means that they
should have beenl constituted by random assignment or by some
matching procedure supplemented by randomization. If this has been
I
done, we may assu~e that the pre-measure for control group II would
have been similar Ito the other pretest scores; i.e., the average initial
score for the experimental group and control group I. This inferred
initial score for control group II is indicated by Y" 1 in Column 5.
I
This produces a gr,oup with both a "before" (assumed) and an "after"
measure. which has received the experimental treatment but in which
there is no possibility of interaction between experimental treatment

and effects of the "before" measure.
It may be noted that the treatment of the second control group in
thi~ design (no pre-measure, but exposure to the experimental variable,
and a post-measure) is the same as the treatment of the experimental
group in the "after-only" design and of the group that is measured only
after exposure to the experimental variable in the "before-after" design
with interchangeable groups. The two-control-group design, however,
not only provides a group in which there is no possibility of interaction
between the pre-measure and the experimental treatment, but also
gives a measure of the extent of that interaction, through comparison
of the experimental group and the two control groups.
If one has reason to believe that contemporaneous events and
developmental processes are not likely to be important influences in
a given study, the "before-after" design with two--control groups may
be interpreted as follows: (1) the change of control group II (d") is
due to the experimental variable alone; -( 2) the change of control group
I (d') is due to the effects of the pre-measurement alone; (3) ifthe
change score of the experimental group (d) is different from the sum
of the change scores of the two control groups (d' + ef'), this is a
reflection of interaction between the pre-measure and the experimental
variable. Such interaction may either enhance or reduce the.effects of
the experimental variable.
Although it is generally recognized that Solomon's solution is a
sound one, his design has not often been used. However, a few studies
in which it has been employed have suggested that interaction effects
are frequently large enough to have marked effects on experimental
results. For example, Canter (1951) used this three-gro,up,design in a:
study of the effectiveness of a human relations training course for
supervisors. The members of the experimental group were given various
tests of information and opinion before the training course, took the
course, and were then again given the questionnaires. The first con-
trol group was given the tests at the same times as the experimental
group (that is, before and after the training program), but was not
given the training program. The second control group was not pre-
measured, but was given the training program and the "after" measure-
CAUSAL INFERENCE FROM EXPERIIVlENTS 121
J1}ents. Comparing the -changes in scores of the three groups, Canter
found that each of the tests in the pre-measure interacted in some way
with the experimental variable: some of the initial tests seemed to
J1}ake the supervisors more receptive to relevant a5pects of the training
course; others seemed to make them less responsive.
The "before-after" study with three control groups. Solomon has
suggested a still further elaboration of this design, to provide safe-
guards when -contemporaneous events or developmental changes may
be expected to influence experimental results. lbis involves the ad-
dition of a third control group, as shown in Column 6 of the table.
Here again, all the groups should -be selected in :luch a way that they
differ only by-chance. In this design, the experimental group and con-
trol group I are pre-measured. Control groups II and III are not pre-
measured; it is assumed that their pre-measure scores would be the
same as the average of the combined scores of the experimental group
and control group 1. The experimental variable is introduced in the
experimental group and in control group II; it is absent in control
groups bnd III. All four groups are assumed to De equally exposed to
contemporaneous events, and all four are meaS1)red after the experi-
ment.
In such a design, the change in control group III (that is, the
difference between the post-measure and the' inferred pre-measure)
represents the effect of contemporaneous events or of developmental
processes, since neither the pre-measure nor the experimental variable
was present in this group. The change in control group II represents
the effects of the experimental variable and of contemporaneous
events or developmental processes; change in control group I, the
effects of the pr~-measurement and of contemporaneous events or
developmental prpcesses. The effect of the experimental factor alone
can be determined by subtracting the change in control group III
(the result of cohtemporaneous events and mflturational processes)
from the change jn control group II (the result of contemporaneous
events, maturational processes, and the experimental variable). The
extent to whkh I the change in the experimental group reflects the
effects of the pre-~easure, of the experimental variable, of uncontrolled
events, or of interaction between the factors can be determined by
comparison of the change in this group with those in the other groups.
It may be observed that this four-group design amounts to doing
the experiment twice, once with a "before-after" design with one
control group (experimental group and control group I), and once
with an "after-only" design (control groups II and III). If the results
of these two experiments are consistent, we have greater assurance that
the outcome is not an artifact than we would with either version
alone, since we have replicated the finding within the study.
TESTING THE EFFECTS OF Two OR MORE EXPERIMENTAL TREATMENTS
All the examples we have considered so far involve the comparison

of an experimental group subjected to a given treatment with a control
group, not subjected to that. treat_ment. Some of them, as has been
P?infea, out, pro;Vide. ~~~ disentangling the effects of. that treatment
from the eff_e.ots ~ of lllltIal measurement, of maturatIon, or of can
tempor~ne6u~.dents.'_However; the 'designs as described so far are
not adequate for testing hypotheses about- the joint contribution of
two or more independent variables in influencing a dependent variable,
or for comparing the effects of two or more experimental treatments
with no experimental treatment. For example, they can provide an
'answer to such a question as: Does a publicity campaign about the
United Nations change attitudes toward 'that organization? But in
their simple form, they do not efficiently provide answers to such
questions as: Which are most effective in changing attitudes-lec-
tures, movies, television program's, or group discussions? or, Is a
publicity campaign most likely to change the attitudes of persons ~ho
were initially unfavorable, initially neutral, or initially favorable? How-
ever, each of the basic designs can be adapted to answer such questIons
efficiently if the experiment is planned iiI such a way as to permIt an
analysis at variance.
Let us suppose that we are interested in finding out which o,f three
techniques in answering bigoted remarks is most effective in il).fluenc-
I ing bystanders: (l) an appeal to intelligence, calling prejudice a dis-
play of ignorance and stupidity; (2) an appeal to "fair play" and
democratic tradition; (3) an appeal to the notion of indiVidual dif-
ferences-i.e., that there are "good and bad" individuals in all groups.
We may decide to include also a situation in which no answer is made:
this is equivalent to adding a control group. Although it is possible to
dispense with control groups in some comparative experiments, since
the groups being contrasted can be considered as "controls" for each
other, this procedure is not recommended. The use of control groups
in comparative experiments establishes a base line against which it is
possible to assess the various methods being studied. For example, a
comparison of two advertising techniques may reveal that one produces
more sales than the other. This "fact" might take on entirely new
significance if it were found that a control group not subjected to
advertising buys more than either of the two experimental groups.
Going back to the answers to bigoted remarks, let us further
suppose that we suspect that each of the techniques of answering may
have different results under different circumstances; for example, one
answer may be best when the answerer reacts in a highly indignant and
emotional manner, another when his manner is calm and dignified.
Eight groups of people are needed for the experiment. The
members of the pool of subjects are randomly assigned to the eight
groups. Each group is shown a scene in which a bigoted remark is
made, but the answer presented before each group involves a different
/
Group Content of Answer Manner of Answer

I Intelligence Indignant
2 Intelligence Calm, dignified
3 Fair 'play Indignant
4 Fair play Calm, dignified
5 Individual differences Indignant
6 Individual differences Calm, dignified
7 No answer Angry turning
away with an ex-
pression of
disgust
No answer Acting as if
nothing has been
said
combination of "content" and "manner," as shown in the preceding

table (p. 123). Suppose that we use an attitude scale to measure the
response of the audience. If the study is set up as an "after-only" ex-
periment, the differences among the groups in scores on this one
administration of the scale will be used as a measure of the effects of
the experimental variables. If it is set up as a "before-after" experiment,
the effect of the experimental variables may be evaluated by comparing
either the final scores of the different groups or the differences in
extent of change between "before" and "after" measures.
With such a design, one can answer a number of research questions
in a single study by comparing different groupings of subjects. In order
to measure the relative effectiveness of appealing to intelligence, to
fair play, or to the concept of individual differences, or of saying
nothing at all, one computes the average score of subjects in groups 1
and 2 (intelligence), in groups 3 and 4 (fair play), in groups 5 and 6
(illdividual differences), etc., and compares them. Similarly, to measure
the relative effectiveness of answers made with different manners, the
ayerage score of individuals in groups 1, 3, 5, and, 7 (indignant manner)
!~ compared with that of individuals in groups 2, 4, 6, and 8 (calm
manner) . In order to determine whether the effectiveness of the various
contents is influenced by the manner in which they are delivered, one
compares the scores of the eight separate groups. If the differences
among them are greater than can be accounted for by the differences
in content and in manner considered separately, one concludes that
the effectiveness of the various contents is influenced by the manner
of delivery. For example, it may be found that the "fair play" argu-
ment is especially effective when made in a calm manner, the 'iin~
telligence" argument in an indignant manner. The statistical tech-
nique which inakes possible the investigation of these three questions
in a single study is known as an analysis of variance. 15
It is apparent that a design which permits an analysis of variance
makes possible a study of complex interrelationships. It also results
in an efficient design; that is, it permits more reliable conclusions about
,
15 Designs of this type and the statistical techniques for their analysis were
developed by R. A. Fisher and are discussed in his book, The Design of Experiments
(1951). Simpler discussions may be found in Cochran and Cox (1957), Edwards
( 1950) , Lindq uist (1953), and other statistical texts.
___- - -
more hypotheses with fewer cases than if the hypotheses were tested
in separate studies.
REPRESENTATIVE DESIGN
A radical criticism of the types of experimental design that wr

have described in this chapter has been advanced by Brunswik (1956).
He expresses the view that the classical experimental designs have
the unwitting consequence of artificially "untying" or "tying together"
variables in a manner that is not representative of the way they exist
in reality. Thus, he points out (Brunswik, 1955), classical experimental
designs in medicine have sometimes led to harmful practices. For ex-
ample, the research which demonstrated that boiling of liquids would
destroy bacteria led to the practice of boiling milk to make it germ-
free. Boiling of milk, however, not only destroys bacteria but also
results in a vitamin loss. Brunswik's thesis here is that the usual sys-
tematic experimental design, which attempts to relate an independent
variable with a dependent variable, throws the picture of the interplay
of factors out of balance: the connection between the boiling of milk
and its devitaminization is ignored if the study focuses only on the
connection between boiling and antisepsis. /
Similarly, in the social sciences one could cite investigations in
which different kinds of leadership style, or different types of visual
display, etc., have been studied in highly controlled laboratory experi-
ments with considerable precaution to keep other variables constant.
Brunswik's criticism of these experiments would be that leadership
style,-Qr a given type of visual display, or any other variable one might
wish to study, occ~rs in a variety of contexts; and that to understand,
for example, the relationship between "permissive leadership" and
"membership participation in group affairs," one must study a re-
presentative sam pie of situations in which leadership occurs. There
may be many diff6rent ways in which the effect of a high degree of
membership participation may be achieved. Permissiveness may pro-
duce it in one coritext (e.g., with a highly motivated and intelligent
membership) and! fail to produce it in another context; authoritarian
leadership may be' effective or ineffective in differing circumstances.
Brunswik, in effect, cans for a survey type of research design in

which a representative sampling of situations is made. This type of
design means that the variables being studied can be investigated in
the context of the naturally occurring concomitant variation of other
factors. The use of this design entails the abandonment of customary
notions of experimental control; we must take the interaction among
variables as we find it in the situations included in the sample. The
essential tools of analysis are partial and multiple correlational pro
cedures. The fruitfulness of Brunswik's proposed methodology has
been demonstrated in perceptual experiments (Brunswik, 1956) and
in some investigations in clinical psychology (l-!ammond, 1955).
Brunswik's criticisms of systematic experimental design have the
salutary effect of highlighting some of the misinterpretations and faulty
overgeneraI,i:zations that are commonly made from traditional e~peri.
ments. Fqrther,: fie has stressed the need for an adequate sampling of
experimental conditions and situations as well as an adequate sampling
of people, if appropriate interpretations and generalizations are to be'
made. However, there are limitations in his methodology, which be-
come apparent when we attempt to answer some of the questions
necessary to the fulfillment of a "representative design." For example,
if one wanted to study the conditions that lead to "active membership
participation in group affairs," one would have to compile a complete
list of group situations in which membership participation may occur
and study a representative sample of this universe, taking measurements
of all relevant variables in each situation that appeared in the sample.
The theoretical difficulties of compiling such a list and of identifying
the variables relevant to membership participation, plus the practical
difficulties of studying a representative sample of such group sItuations,
are, of course, insuperable in anyone investigation.
Practically speaking, the research investigator must always settle
for less than an investigation of all of the relevant conditions of the
phenomen~ he wishes to understand. Part of the art of ,experimental
investigation lies in having a "feel" for one's subject matter, in being
sensitive to the nature of the phenomena which one is investigating
so that the significant dimensions are taken into consideration. Sci-
entific research, as well as any other meaningful human activity, requires
udgment. There are, of course, no guarantees that even well-tutored
ludgment will be correct or fruitful.
Causal Inference from Other Study Designs16
The investigator may not be in a position to test a hypothesis by

assigning subjects to different conditions in which he directly controls
the presumed causal variable. Consider, for example, the research
that has been done on the relationship between cigarette smoking and
lung cancer in humans. The extent of smoking has not been controlled,
as it would be in an experimental study, by assigning different individ-
uals to smoke different numbers of cigarettes; for many reasons, such
an experiment is not likely to be performed on humans. Rather, a
record (or a retrospective estimate) is secured of how much an in-
dividual has smoked, and of whether he has lung cancer; the relation
between extent of smoking and occurrence of lung cancer is then
computed.
But, as has been pointed out, the existence of a correlation be-
~tween two variables does not necessarily demonstrate that one is a
"cause" of the other. \Ve must, therefore, at least consider the pos-
sibility that the kinds of people who become smokers are also, for
some as yet unknown reason, the kinds of people who develop lung
cancer. We might remember that, before smoking became common
among college students, it was found that smokers made poorer grades
than nonsmokers. Eventually it became clear that, at least in those
days, th,e students who smoked were also the students who were
especially likely to I1eglect their studies.
Whatever the :form ora study, if it is to provide a test of a causal
hypothesis, it mu~t provide grounds for making inferences about
causality and safeguards against unwarranted inferences. Nonexperi-
mental studies cannot provide safeguards as adequate as those given by
random assignment of subjects to experimental and control groups,
direct manipulation of the experimental variable, and control over
some of the extran~ous
, variables that might operate during the course
16 For a detailed and enlightening discussion of the testing of causal hypotheses
by means of nonexperimental studies, see Hyman (1955, Part III). Our discussion
of this subject is very much indebted to Hyman's presentation.
of the experiment. What substitute safeguards are available? For direct

manipulation of the experimental variable, the investigator may sub-
stitute One or more of several lines of evidence: comparison of people
who have been exposed to contrasting experiences, attempts to deter-
mine the time order of variables that are ~ssociated, examination of
the relationship between variables in terms of the pattern of relation-
ships that might be anticipated if one or the other were the causal
factor. For assignment of subjects to experimental and control groups,
the investigator may substitute evidence which provides a basis for
inferring that groups of people who have undergone contrasting ex-
periences were or were not similar before those experiences; or he may
select from his total group subs am pIes matched in terms of certain
characteristics but with contrasting experiences; or he may restrict his
sample to persons with certain characteristics. For direct control over
extraneous variables, either past or contemporaneous, he substitutes
the gathering of data on other characteristics or experiences of his
subjects which he believes may be relevant to position on the depend-
ent variable, and makes use of these data in .his analysis. This latter
operation will be discussed in Chapter 11. Various procedures intended
as substitutes for direct manipulation of the assumed causal variable
and for random assignment of subjects to experimental and control
groups will be discussed in the remainder of this chapter.
SUBSTITUTES FOR DIRECT MANIPULATION OF THE

ASSUMED CAUSAL VARIABLE
I
COMPARISON OF GROUPS EXPOSED TO CONTRASTING EXPERIENCES. If

an investigator is not in a position to assign subjects to different groups,
one of which will be exposed to a given treatment and one of which
will not, an obvious substitute solution is to locate groups of people
who are about to be, or have been, exposed to experiences that differ
with respect to the assumed causal variable in which he is interested.
For example, Cook, Havel, and Christ (1957), in planning a stu9Y of
the effects of a summer orientation program for foreign students in
the United States, knew in advance what students were to attend the
orientation centers, although they had no part in selecting them. They
compiled from all available sources a list of foreign students who were
CAUSAL INFERENCE FROM OTHER STUDY DESIGNS 129
to begin their formal studies in the United States at the same time as
these "orientation" students, but without receiving the prior orienta-
tion. Students in both groups were similar in that they were coming to
the United States under some form of grant or scholarship made avail-
able by the United States government. Starting with the list of students
who were to attend the orientation centers, these investigators selected
students from the comparable non-orientation list on the basis of
frequency distribution control, roughly matching the two samples
in terms of such chaJ;rlcteristics as nationality, age, and field of study.
Since the students were identified in advance, it was possible to secure
initial measures early in their stay in the United States, and thus to
check whether the two groups were initially similar in their position
on the dependent variables. This study approximated a "before-after"
experiment with one control group: the students attending the orienta-
tion centers represented the "experimental" group; those not attend-
ing the centers, the "control" group. However, there was an important
limitation: the "experimental" group and the pool of subjects available
for the "control" group were constituted on the basis of administrative
considerations having to do with the type of scholarships they received,
and they may therefore have differed from each other in some sys-
tematic way. In a "pure" experiment, the total group of students
would have formed a pool from which individuals were randomly as-
signed (with or ;.vlthout supplementary matching procedures) to
attend or not'to attend an orientation center.
The investigator often is not in a position to know in advance,
as Coqk, Havel, and Christ did, which individuals will be exposed to
a given experience and which will not. He may instead locate a group
of people who have been: exposed to the type of experience in which
he is interested and a gr6up, similar in other respects, who have not
been exposed to such an: experience; he compares the groups on an
ex post facto basis. The Qeutsch-Collins (1951) study of the attitudes
of white tenants toward Negroes in public housing projects (described
in Chapter 1) followed tHis procedure. The investigators selected hous-
ing projects that differed in the independent variable in which they
were interested-occupa.J.?cy pattern. Within these projects, they
selected white housewives on a random basis. Thus they approximated
an "experimental" group, which had been exposed to the experience
of living with Negroes as neighbors in an unsegregated occupancy

pattern, and a "control" group, which had not had this experience. By
selecting as their "control" group residents of bi-racial housing projects
within which whites and Negroes were segregated, rather than other
residents of the city, they controlled certain extraneous variables, such
as the socioeconomic level of the subjects and the experience of living
in a housing project. Studies using this ex post facto pattern have an
even more serious limitation than those resembling the foreign student
study just described; not only are subjects not randomly assigned to the
different conditions, but there is no possibil~tyof priormeasu~ment to
check whether the two groups w.ere initially sifuiIC!r:'in their; position on
the assumed dependent variable or in certain characteristics believed to
be relevant to it. For this reason, such studies must provide for the
collection of data from which it can be inferred whether the two gr~ups
were or were not initially similar. Kinds of evidence that may provide
a basis for such inference are discussed later in the chapter.
EVIDENCE OF THE TIME ORDER OF VARIJ\.BLES. As we have said, one
kind of information that may help to provide grounds for an inference
that a specified variable (X) is the cause of another (Y) is evidence
that Y did not occur before X. In some cases the temporal relationship
of two variables is so clear that no supplementary evidence is needed.
For example, if one finds that individuals born in different countries
differ in their views on some political issue, there is no question as to
which variable, nationality or political views, is prior in tilpe. Often,
however, the time relationship between two variables is not so clear.
Even though one appears to be prior to the other, it may not actually
be so. For example, in a study of the effect of . childhood experiences
on adult adjustment, an investigator may rely on his adult subjects'
accounts of their childhood. Obviously, childhood experiences are
prior, but the selection and interpretation of events may be colored by
present mood, with the result that the apparently earlier variable, In
the form in which it is measured, is really a reflection of the later
variable. In other cases, there may be no basis at all-not even a decep-
tive one-for making a judgment of time relationships; this is likely
to be the case, for example, when one is dealing with two/attitudes that
are related.
The investigator may introduce various procedures in an attempt
to secure evidence of the time relationship between the occurrence of
variables. Two such procedures-asking people about time relation-
ships and gathering evidence over a period of time-are discussed in the
following paragraphs.
Asking respondents about time relationships. The investigator
may include in his questionnaire or interview questions about when
certain things happened, or how the respondent felt about something
before a certain event took place, or whether there have been any
changes in his feelings. Thus, Deutsch and Collins asked such ques-
tions as: "Before you moved into [the project], how did you feel about
coming to live here? ... Since you've moved into [the project], have
you come to like the idea of living in a project where there are colored .
and white families more, about the same, or less? . . . Can you remem-
ber what you thought colored people were like before you moved into
the project? . . . How much have your ideas about colored people
changed since you have lived in the project? . . . (If some change)
In what ways have they changed?"
There is, of course, always the danger that replies to such ques-
tions may be inaccurate. People may not remember; their present
attitudes may distort their recollections; etc. While it is not possible
to check conclusively the accuracy of such retrospective reports, the
investigator can sometimes devise indirect checks on the probability ,
of distortion. For example, Deutsch and Collins compared the replies
of respondents in the four projects they studied to the question about
prior attitudes toward Negroes. Other questions ill their interview
showed striking differences between tenants of integrated and segre-
gated projects in attitudes at the time of the study. If current attitudes
were distorting recall, the investigators reasoned, this should lead to
consistent differences ~tween the tenants in the two types of project in
their reports of the attitudes they had before moving in. As it tUTllP-d
out, although the two; segregated projects were very similar in the
proportion of tenants I reporting favorable, neutral, and unfavorable
attitudes toward Negro~s before moving in, there was no such similarity
between the two integrated projects. In one, more tenants than in the
segregated projects rep~rted initially favorable attitudes; but the other
had the highest prop~rtion reporting initially unfavorable attitudes.
Fr~inthe fact that the reports of tenants in the integrated and the
segregated projects about their initial attitudes did not differ con-
sistently, whereas their current expressions of attitude did, the in-
vestigators concluded that recall was not. being systematically distorted
by attitudes at the time of the study. Further, they found that women
who were similar in education, religion, and political attitudes gave
similar reports of their initial attitude toward Negroes, regardless of
which project they lived in; this consistency was taken as further
evidence against distortion of recall by present attitudes.
Gathering evidence through studies extended over time. In studies
which are limited to a single interview or observation or other measure-
ment of each respondent, and in which the investigator does not have
supplementary information about individuals' experiences, there is
little possibility of getting evidence about time sequences except by
asking the respondent to recall when things happened. But in studies
that focus on the same people over a period of time, the investigator
may secure direct evidence of time relationships among variables. Such
longitudinal studies may take the form of repeated observations of the
same subjects, or 'fepeated interviews with them,17 OF of different meas-
urement procedures at different times.
Stouffer et al. (1949a) provide an example of a study using differ:
ent kinds of data about the same subjects at different times. The in-
vestigators were interested in the relation between acceptance of the
official value-system of the Army and promotion. Had they simply
intef\;iewed a cross-section of Army personnel and found ~at those
of higHer rank expressed attitudes and opinions more in line with
official Army values, they would have h~d no grounds for inferring
whether acceptance of the official value-system was conducive to
promotion or whether being promoted increased acceptance of the
! system. To avoid this dilemma, they interviewed ,a group of newlY
inducted soldiers, using questions fromwhich an index of "acceptance
of Army value-system" could be constructed. Four months later, they
examined the. Army records of these same men... and found that a
higher proportion of those who had expressed views in keeping with
the Army's values had become privates first-class than-of those who
had not. Thus it was clear that conformity with the Army's value-
J 7 For a detailed discussion of studies using repeated interviews with the same
respondents ("panel studies"), see Rosenberg, Thielens, and Lazarsfeld (1951).
system was conducive to promotion. (It is, of course, entirely likely that
the relationship between these two variables is a mutually reinforcing
one; further research might well have shown that after promotion,
views were even more in line with official Army position.)
SEARCH FOR PATTERNS OF RELATIONSHIP INFERRED FROM COMPETING
CAUSAL ASSUMPTIONS. Sometimes one can infer which of two factors
that vary together is the "causal" one on the ground that the two
variables would show a certain pattern of association if X were the
"cause," a different pattern if Y were the "cause." For example, it is
sometimes reasonable to expect that if X were the cause, it would
affect Y cumulatively-that is, that individuals who had been exposed
to X for a longer time would show a higher degree of Y-but that this
would not be so if Y were the causal factor.
Such an inference was central to the plan of a study by Newcomb
(1943, 1947), which focused on the question of what kinds of people
accept certain kinds of social change. One of the hypotheses of the
study was that "values come to be values largely through the mediation
of the groups with which an individual has direct contact." Studying
stu~ents at Bennington College, Newcomb considered the college
community as a group with which the students had direct contact, and
attitude toward public affairs as a relevant value. This attitude was
/
selected because ,the college was characterized by a high degree of
concern with public affairs and a "liberal" attitude on controv~rsia)
issues. The investigator reasoned that if group membership werefudeed
the causal variable, then those who had been exposed to the group
atmosphere for longer periods should show attitudes more in keeping
with those characteristic of the group (in this case, more liberal atti
tudes). If, on the other hand" it was the possession of liberal attitudes
that led to attending the col~ege, there would be less reason to expect
an increase in liberalism with increased years of attendance. 1s Using a
variety of measures, Newcomb found that length of exposure to the
Bennington community, as iqdicated by college class, was accompanied
by increased information about public issues and increased liberalism
18 Again, of course, there is Ithe possibility that attending the college migh~,
further have strengthened initial liberal attitudes, but presumably this would not1
have led to such marked differences between longer.exposed and shorter-exposed'
students as would be expected if college membership were the major causal factor.
of views. On all his measures, freshmen were most conservative, seniors I

least so.
Similarly, in an earlier study, Klineberg (1935a) reasoned that
the comparatively low IQ's of Negroes in the South and in rural areas
might be attributable to poor environments. This led him to expect
that the IQ's of Negro children should increase with length of resi-.
dence in a city such as New York. Investigations of twelve-year-old
Negro children in New York City supported this expectation.
However, the fact that scores on Y differ with different lengths
of exposure to X does not provide a clear-cut basis for an inference of
causality. Not only may X and Y be mutually reinforcing, as we have
noted. Other factors may be associated with differences in length of
exposure to the independent variable, and it may be these other factors
which account for differences in the dependent variable. For example,
selective processes may be different at different times. In the Bennillg-
ton illustration, it might have happened either that increased prestige
of the college (which had been founded only a few years before the
study) or a change in admissions policy leg to a change in the type
of students entering in different years, and this change, rather than
the influence of the college atmosphere, might have accounted for the
differences between freshmen and seniors. Similarly, it is conceiyable
that more intelligent Negroes might have migrated to New York City
earlier, and that the higher IQ's of children who had been in New York
longer reflected this characteristic of their parents. .
A variety of checks for such possibilities ha~been used. One is
repetition of the study at another .time. Newcomb repeated his meas-
ures of attitudes of Bennington students for four consecutive years
(1935-1939). Each year, he found that juniors and seniors were more
liberal than those in lower classes; thus the inference was strengthened
that it was the college experience rather than some other factor which
accounted for the change in attitudes and information. Klineberg
also used the repetition procedure, conducting parallel investigations
in two successive years, 1931 and 1932. He reasoned:
If the findings [of the 1931 study-that Negro children who
have been in 'the North longer have higher IQ's] are due to a
progressive deterioration in the quality of the migrants. rather
than to an environmental effect, the results obtained . . . in
1932 should be consistently below those obtained ... in 1931.
A specific example will make this reasoning clearer. The twelve-
year-old boys in the 1931 study who have been in New York
four years, for example, arrived in 1927; those in the 1932 study
who have been in New York for a similar period arrived in 1928.
If the migrants are becoming inferior as time goes on, the four-
year group in the later study ought to be inferior to the corres-
ponding group in the earlier one.
Comparison of the 1931 and 1932 results, however, showed, for
each length-of-residence group, a slight but consistent superiority in
the 1932 study. For example, the average IQ of students who had been
in New York for three or four years in 1931 (that is, who had migrated
in 1927 or 1928) was 66.86, while the average IQ of those who had
been in New York for three or four years in 1932 (that is, who had
migrated in 1928 or 1929) was 79.06. In other words, the differences,
although slight, were consistently in favor of the more recent arrivals.
.
Klineberg commented:
This difference may be due to improvement in the schooling in
the South; in any case there is no evidence that the more recent
arrivals are inferior. The conclusion is therefore justified that the
superior showing of those subjects who have had a longer period
of residence is due to this longer residence, and n~t to any regular
change in the quality of the migrants.
There are, of course, methods other than repetition of the study
to determine whether other factors may be responsible for the dif-
ferences found, on the dependent variable. For example, Klineberg
provided two other checks on the hypothesis that differences in the
intelligence of Negr~es who left the South at different times might
account for differences in IQ related to length of residence in New
York. He investigated\reasons for migrating from the South, and found
nothing in these reasqns to support the hypothesis that factors leading
to migration might be expected to correlate with intelligence. He also
studied records of southern schools attended by Negro children, and
found no systematic 1,ifference in the relative class standing of children
who subsequently migrated and of those who did not. Both these lines
of evidence suggested that, at least.during the period covered by these
interlocking investigations, intelligence was not a selective factor in

Negro migration from the South.
Another factor that may be confounded with length of exposure
to a variable is age. It is likely that those individuals who have been
exposed to a given experience for a longer time are older, and it may be
the difference in age rather than in exposure to the assumed causal
variable which accounts for differences in the dependent variable. The
solution here is to "control" for age-that is, to compare individuals of
the same age who differ in length of exposure to the variable. Newcomb
did not use this approach, presumably because of the slight age dif-
ferences within a college population. In the Klineberg study, where
all of the children were twelve years old, age was not a factor. However,
in a study that covers a wide age range, this is an important point to
check; and it is relatively easy to do so.
SUBSTITUTES FOR RANDOM ASSIGNMENT OF SUB JECTS TO

EXPERIMENTAL AND CONTROL GROUPS
As we have pointed out, one of the ways in which the experimenter

protects himself against fallacious inferences. is by random selection
of his experimental and control groups, with or without supplementary
matching. Probability theory makes it possible to determine to what
extent groups randomly selected from the same population are likely
to differ by chance, either in initial position on tht::dependent variable
or in characteristics or past history that might be expected to inHuenc~
subsequent position on the dependent variable. Tests of statisti01l
significance, which take into account the probability of chance differ-
ences, make it possible to say, with a specified degree of certainty,
whether a difference between the two groups on the dependent vari-
able after exposure of one of them to the experimental treatment is
greater than would be expected by chance, and thus can legitimately be
attributed to the effects of the experimental variable. The inves~igator
using a nonexperimental design has a variety of substitute ways .qf try-
ing to achieve this objective.
EVIDENCE OF lNITIAL COMPARABILITY OF GROUPS. The investigator
may gather data from which he can infer whether individuals who have
undergone different experiences and who now show differences on the
dependent variable were comparable before they were exposed to the
experiences in question. Newcomb used this procedure, among others,
in the study referred to earlier. He compared Bennington students
with students at two other colleges which did not have the same
general atmosphere of liberalism and of concern with public affairs.
He found Bennington seniors markedly less conservative than those
at the other two colleges. Before he could infer, however, that this
difference resulted from differences in college atmosphere, he had to
check on the possibility that students who attended Bennington might
have been less conservative even before they entered college. Compari-
son of attitudes of freshmen at the three colleges served this function,
and indicated that Bennington students at the beginning of their
college life were only slightly less conservative than freshman at the
other colleges; thus he could conclude that the groups of students
had initially been quite similar.
Frequently itis impossible to obtain measures of individuals in the
early stages of the experiences whose effects one wishes to determine.
However, if enough is known about other factors likely to influence
or at least to be associated with the dependent variable with which
the investigator is concerned, he may be able to make tentative esti-
mates of initial position on the dependent variable by gathering in-
formation about these related characteristics. This was one of the
chec~s used by Wilner, Walkley, and Cook (1955) in their study of
the attitudes of white tenants toward Negr~es in public housing
projects with different occupancy patterns. Earlier studies had in-
dicated that such characteristics as education, general political ideology,
religion, and previous experience with Negroes are related to attitudes
toward Negroes. Accordingly, these investigators included questions
on these matters in t~eir interview. On the basis of the replies, they
constructed for each respondent an estimate of her probable degree
of favorable attitude toward Negroes at the time she moved into the
project; they then cOmpared tenants in integrated and segregated
projects in terms of this estimate of initial attitude. Obviously, evi-
dence of this sort is' fjlr from conclusive unless the customary correla-
tion between the dep~ndent variable and the characteristics On which
the investigator has information is known to be sufficiently high fm
the purpose.
There is usually a possibility that people who have undergone

different experiences chose them. Such self-selection almost certainly
means that the groups being compared were not initially equivalent;
they may have differed in ways that would strongly influence their
position on the dependent variable. Therefore it is important, when-
ever possible, to check on the likelihood of self-selection. Thus, Wilner
and his associates included among their checks evidence on this point:
some from the tenants themselves, some from other sources. They
asked white tenants whether, at the time they applied for admission
to the projects where they were living, they knew that the projects also
had Negro tenants. If an individual did not know that a project was bi-
racial, there was little chance that his attitude toward Negroes would
enter into his decision about moving in. From housing officials they
obtained the following information: whether all housing projects in
the community had the same racial occupancy pattern, thus red~cing
the opportunity for choice; records of refusals by white applicgnts of
apartments next door to Negroes, or of moving out of such apartments;
records of refusal and move-out rates in apartments not near a Negro
famIly; etc.
Occasionally an investigator will be fortunate enough to find a
situation in which there was no possibility of self-selection. For ex-
ample, the Information and Education Division of the U. S. War
Department (1947) found that white soldiers in units that had com-
panies made up of white and Negro platoons were much more favor-
able toward having white and Negro platoons in the same company
than were white soldiers in all-white units. Given the nature of military
assignments, there waS no possibility that white soldiers who were I
initially more favorable toward Negroes had placed themselves in

mixed units.
COMPARISON OF MATCHED S'uBGROUP~. Suppose that the evidence
collected indicates that the groups whi'ch have undergone different
experiences were not initially comparable. This was the case in the
Deutsch-Collins (1951) study of attitudes of public housing tenants.
These authors found that the residents of their integrated and segre-
gated projects differed in religion, education, and political attitudes-
characteristics which other studies had found to be related,to attitudes
toward Negroes. Moreover, the nature of the differences was such that
the tenants in the integrated projects might be assumed to have been
somewhat less prejudiced before moving into the projects than were
those in the segregated ones. The authors state:
. . . it is clear that in the analysis and presentation of our results
it wi1l be necessary to eliminate or control the effects of these
population differences in order to attribute causal significance to
the effects of the occupancy pattern.
To do this, they compared matched subgroups within the projects.

For example, considering separately' the politically liberal tenants
interviewed, they examined the proportion within the integrated and
the segregated projects who showed friendly feelings toward Negroes in
the project. They found that 31 per cent and 72 per cent of the political
liberals in the two integrated projects, as compared with 0 per cent and
17 per cent of the political liberals in the two segregated projects, were
classified as having friendly feelings toward Negroes in their project.
Separate examination of "middle-of-the-road" respondents, of con-
servatives, of those of different religions and different amounts of
education, all showed the same results: among respondents matched
in these other characteristics, those in integrated projects consistently
showed more friendly feelings toward Negroes in the project. 19
RESTRICTION OF THE SAMPLE. Rather than ta19ng a heterogeneous
sample and comparing matched subgroups within it, the investigator
may ensure that his subjects are matched in certain respects by includ-
ing in his sa~ple only persons with certain characteristics. This pro-
cedure may be used in experimental studies als~; there it is done usually
for reasons of administrative convenience or to increase the sensitivity
of the experiment by ruling out factors that might influence the de-
pendent variable so strongly as to obscure the effects of the independent
variable in which t~e investigator is interested (see pages 102-104).
In nonexperimental tudies,
I
restriction of the sample has another func-
tion: disentangling of the independent variable in which the investi-
gator is interested frbm other variables with which it is commonly as-
sociated, so that any:effects that are found can justifiably be attributed
to that variable. In fxperiments, of course, this separation is achieved
19 Procedures used in analysis to compare such subgroups are discussed in
Chaoter 11. '
by random assignment of cases to experimental and control groups and

by control over some of the other contemporaneous variables.
For example, suppose we wish to investigate whether having a
quiet place at home in which to study affects high school students'
grades. Having a quiet place to study is likely to be part of a cluster
of factors. Overcrowding is greater among low-income families; thus
there is less possibility of quiet study space in their homes. At the same
time, lower-class families are likely to put less emphasis on scholastic
achievement than are the parents of middle- and upper-class children.
Moreover, in lower-class homes there is likely to be less opportunity for
;' children to gain the type of knowledge that would contribute to
J " academic achievement. If one simply selects a cross-section of high
school students, asks each of them whether or not he has a quiet place
to study at horne (or gets this information in some other way), and
then compares the grades of those who have a quiet study plac~ and
those who do not, he may corne to a quite misleading conclusiOn. The
difference in grades may be due not to the difference in study arrange-
ments but to the difference in emphasis on scholastic achievement, in
opportunities for gaining scholastically relevant knowledge, or to some
combination of all these "factors associated with socioeconomic class.
If the problem were amenable to experimental investigation, the
experimenter might select a cross-section of students and arrange for
some of them (randomly selected) to have a quiet study place, others
(also randomly selected, but perhaps also matched in socioeconomic
:;tatus) not to. Lacking this possibility, the investigator may rule out
the effects of class differences by; limiting his study to students of dnly
one socioeconomic level. He might study only lower-class children, or
only middle-class ones, comparing the grades of those with and without
quiet study arrangements. In thus limiting his sample, he would have
disentangled his independent ~ariable from certain c1ass-iinked vari-
ables with which it is usually associated. He would not, of course, have
ruled out such complications as the possibility that within the lower
class, or within the middle class, parents who are most concerned with
academic success provide quiet study arrangements for their children.
In other words, if he found a relationship between quiet study space
and school grades within his restricted sample, he could feel confider~t
that the differences in grades were not due to differences in socio-
economic status, though they might be due to some other factor or
factors which he had not controlled.
A simple logical point is relevant here. The phrase "not due to
differences in . . ." means, quite literally, "not due to differences
in . . ." It does not mean that socioeconomic status plays no part in
the observed relations. Thus, if his study included only lower-class chil-
dren, a finding that children who had a quiet place to study received
higher grades might hold only for that socioeconomic group; among
middle-class children, it is possible that some other relationship, or no
relationship, would be observed. The point is worth making because it
is commonly, albeit incorrectly, assumed that holding certain variables
constant in a study eliminates their effect. Consider, for instance, the
following example: One member of a pair of identical twins is raised
in an underprivileged socioeconomic environment, the other under
much better circumstances. At the age of twelve, a difference of
twenty IQ points is found between them. It is now asserted that, since
their heredities were identical, the IQ difference is due to the difference
in environment-and, by implication, that the common heredity had
nothing to do with it. Yet it is conceivable that, if that common
heredity had been different in some significant respect from what it
actually was, the two children might have reacted differently than
they actually did to their respective environments"and a different IQ
difference, or even no difference, might have been found.
While this procedure of restricting the sample has the advantage
of ruling I)ut variations in other specified characteristics as contributory
conditions of the observed variations in the dependent variable, it has
the corresponding disadvantage of limiting the population to which
the findings can be generalized. Thus, in our example of the relation
between study space iand school grades, if we limit the subjects to
lower-class children attending public schools, any findings about the
effect of study space on grades cannot with any confidence be presumed
to hold for middle-cl~ss children, etc. In order to have confidence in
the geperalizability o~ the findings, the study must be repeated with
groups of subjects who have other characteristics. This process of repe-
tition with other groupsj
has, of course, an additional function: i1
helps to specify the contingent conditions under which a relationship
between two variable~ obtains.
USING A COMBINATION OF ApPROACHES
Since none of these substitutes for experimental procedures pro

vides as much protection as does random assignment of subjects to
situations in which the invest1gator controls the application of the
independent variable, it is always advisable to plan as many different
approaches as possible to provide a basis for making inferences of
causality. Most of the studies we have mentioned used several checks;
we have not described them all. The subject matter of a study will
often determine which checks are most relevant and most feasible.
Most of the devices for making possible causal inferences in non-
experimental studies introduce specific requirements in the analysis of
the data. Therefore a section of Chapter 11, which deals with analysis
of data, is devoted to a discussion of analytic procedures useful when
one wishes to draw inferences about causality from such studies.
Summary
In these two chapters we have been concerned with the require

ments for research procedure posed by different kinds of research
problems. We have stressed the fact that not even carefully controlled
research will give knowledge with absolute certainty. Yet despite its
limitations, scientific method, more than any other procedure known
to man, can minimize misinformation. The skepticism, the alert self
criticism, the constant testing of hypotheses by empirical research, the
awareness of its own limitations ~hich characterize scientific end~avor,
make research results, if intelligently used, the most dependable source
to which One can turn for information.
In Chapter 3 we pointed out that research is not simply a matter
of experimentally testing well-formulate'd hypotheses. The develop-
ment of fruitful hypotheses does not occur in a vacuum, nor is It solely
a matter of good fortune or ingenuity~ It can be aided by carefully
planned exploratory and descriptive studies, which have the purpose of
accumulating the background information necessary to a pertinent
formulation of the problem. We;. have indicated that such studies,
particularly in a young science, call for the most imaginative efforts
SUMMARY 143
the scientist has to offer and are not to be scorned because they lack
the elegant rigor of the experimental study. The function of descriptive
studies in assessing the characteristics of a given situation and in test-
ing hypotheses that do not involve causal relationships has also been
examined. In the present chapter we have discussed in detail studie~
designed to test causal hypotheses. In this context we have presented
the logical considerations involved in drawing inferences about causal-
ity, and have described various procedures for gathering data that can
provide a basis for such inferences.
It is now appropriate to turn our attention from the structure of
a research study to the methods of giving substance and body to the
research design-the data-collection procedures. In the next chapter, we
shall discuss some general considerations of measurement that affect
all data collection. Chapters 6, 7, and 8 will then discuss methods for
the collection of primary data-observational methods, questionnaires
and interviews, and projective techniques. In Chapter 9, the use of
secondary sources of data-statistical records, personal documents,
communication content-will be treated. Chapter 10 will consider
techniques for placing individuals on scales on the basis of data
collected by any of these methods.
/
5
SOME GENERAL PROBLEMS
OF MEASUREMENT
Variations in Scores on Measuring Instruments
The Validity of Measurements

/
The Reliability of Measurements
Scales of Measurement
Summary
I PROPERTY OF I
71z~ Kansas S.ate University of I
Agriculture & Applled Science I
I, T C M India. -'
Measurement . . . is more than the pedantic pursuit of a
decimal place. Its vital and absorbing aspect emerges most
clearly perhaps when it becomes a question of measuring some-
thing that has never been measured. Or better still, something
that has been held to be unmeasurable. s. s. STEYENS
; T H E QUALITY OF RESEARCH depends not only on the adequacy of

the research design but also on the fruitfulness of the measure-
ment procedures employed. Basic to any meaningful measurement are
an adequate formulation of the research question and clear definitions
of the concepts involved~tiier words, one musfust-kio.w_what-iL-
""~_ - '1 ..",....__-" -
is he wants to measure .
..._ . SuppoSe11iat we are faced with the problem of determining the
effects of a visit to the United States on the attitudes of British tourists.
In order to secure relevant data, we must know what questions we want
to investigate. Specifying these questions requires, among.other things,
consideration of the: concept attitude. One may include in"his defini-
tion of attitude various aspects-for exalT\ple, beliefs about the nature
of an object, person, or group; evaluations of it; tendencies to behave
toward it in a certain way; views about appropriate policy with respect
to it. One !liay also include in his definition such other characteristics
of an attitude as the salience of the objedfor the individual, the extent
of differentiation in his view of the object, his time perspective with
respect to it, etc. Having thus specified his definition of attitude, the
investigator is in a position to formulate his research problem more
clearly by deciding whid~ aspects he wishes to focus on. He may wish
, to determine whether a stay in the United States changes the British
visitors' beliefs about the nature of life in America, or whether it leads
to changes in the extent to which they are willing to generalize about
146
GENERAL PROBLEMS OF MEASUREMENT 147
Americans, or to changes in the degree of their liking and respect for
Americans, or in their evaluation of American foreign policy.
Such specification of what is to be measured is a. prerequisite to
deciding how it is to be measu"fed_:_that -is,to the 'establishment of
measurement procedures. A measurement procedure consists of a
technique for collecting data plus a set.of rules for using these data. The
purpose of the various data-collection techniques is to produce trust-
worthy evidence that is relevant to the research questions being asked.
(As we have noted in the preceding chapters, a period of exploratory
research is often needed in order to find out what kinds of data
actually bear on the question or constitute adequate indicators of the
concepts.) The purpose of the accompanying_ rules is .to. .facilitate the
use of these dataTnrriakiIlgSpecific "StateIT;;:rts about_the charact~ristics
of the phenomenon to -which th.e d;t;-are believed to be relevant. The
m~ment procedures constitute the ''Working definitions" of the
concepts being used in the study, as discussed in Chapter 2.
Data may be collected in many different ways: by observation of
behavior, by questionnaires or interviews, by projective techniques, by
examination of existing rec~rds. The rules for using these data to make
statements about the phenomenon in which one is interested may be
built into the data-collection technique, or they may be developed as a
supplement to it An attItude -questlon-naire that yields a score placing
an individual along a favorable-unfavorable scale not only provides
for collection of the data necessary for an estimate of the individual's
position, but also includes the rules for making that estimate. On the
other h~nd, an unstructured interview on the same subject may gather
the necessary data, but a coding system (that is, a set of rules for using
the data) is necess!lry in order to estimate the degree of the individual's
favorable or unfavorable attitude. The discu!,sion in this chapter and
subsequent ones will clarify this distinction. _
The data-collection techniques and the rules for using the data,
to be useful, must produc'e information that is not only relevant but
fr~e of systematic errors; that is, they must produce valid information.
Suppose that the study of British visitors is to focus on their evaluations
of Americans. A study us~pg techniques which, .for example, led to a
substantial proportion of persons with favorable feelings being classi
fied as unfavorable could not e~erge with sound concfusions.
148 GENERAL PROBLEMS OF MEASUREMENT
A good measurement procedure must also be reliable; that is, in-

dependent but comparable measures of the same O15Ject"""\or attitude,
or wnatevef) should give similar results (provided, of course, that,
there is no reasontO-Delieve tnaTtne object being measured has in fact
changed between the two measurements). A measuring stick, for ex-
ample, is a highly reliable instrument. Under ordinary circumstances,
a table which it shows to be 30 inches long on one day will appear 30
inches long the next day and the next; the variations are likely to be
negligible for most practical purposes. However, a measuring tape made
of elastic would be extremely unreliable; the table might appear 28
inches long one day and 32 the next, depending on how taut the elastic
was stretched. Similarly, an instrument established to measure feelings
toward Americans would be considered unreliable if persons who were
classified as "favorable" on one measurement were classified as "un-
favorable" on a second measurement made at such a time and under
such circumstances that there was no reason to believe their feefings
had in fact changed. .
In addition to being valid and reliable, a research instrument
should.v be capable of making distinctions
_ _ .
~~~
fine enough for the purpose
_ __ _
it is to serve. InstrumentS-differ in the specificity or exactness with

(' which they attempt to localize the position of-any individual in respect
to the characteristic being measured. For example, a very crude measur-
ing instrument might distinguish only two positions in the British
visitors' evaluations of Americans: "favorable" and "not favorable."
A somewhat more finely graduated instrument would distinguish
"f.avora bl" " an even more. fi ne1y gra d ua t ed
e, " neu t ra,I"",un favora bl'e;
measuring technique would distinguish varying degrees of "favorable"
and "unfavorable." If our measuring instrument could distinguish only
two categories, our study would be unable to reveal many socially
important shifts in attitudes; for example, from "unfavorable" to
"neutral" or vice versa, or from "slightly favorable';. to "mar~edly
favorable." .
In addition to demanding the ability to make fine distinctions,
research objectivespften call for data-collection procedures that permit
us to state how much people differ in a given characteristic. In a later
section of the chapter it will become clear that such statements can be
GENERAL PROBLEMS OF MEASUREMENT 149
meaningfully made only if the measurement procedures involve scales
that have equal units as well as certain other characteristics.
One further point should be noted. Measurement of an individual,
object, event, etc., in terms of a given attribute presupposes that the
individual or object can appropriately be described in terms of that
attribute. An attempt to measure a given attribute may be irrelevant for
some people, objects, etc. For example, it would not make much sense
to ask how favorable a person is toward logical positivism if he is not
aware of such an approach to the philosophy of science. Unfortunately,
it is not always readily apparent whether measurement in terms of a
given attribute is or is not relevant for a given individual. Especially in
the case of questions of opinion or attitude, many people can be in-
duced to give answers on matters about which they really have no
opinion or attitu~e. Clearly, the results of such measurements have
little or no meaning. There is no simple solution to this difficulty. By
being alert to it, however, the investigator either may incorporate into
his measuring instruments provision for ascertaining whether the
measurement is or is not relevant for a given subject, or he may decide
to omit measures tliat seem likely-to be irrelevant for a considerable
number of subjects. Again, a period of exploratory research may help
to provide a basis for judging whether measurement of a given attribute
i~ relevant for a given group. /
Variations in Scores on Measuring Instruments

Measurement always takes place in a more or less complex
situation in )Vhich innumerable factors may affect both the character-
istic being measured and the process of measurement itself. One at-
tempts to control qr keep constant the more important of these vari-
ables and hopes that the variation of uncontrolled factors will operate
so as to cancel out one another's effects. The statement that a body has
a certain length, fOTj example, is accurate only in relation to an assumed
set of conditions among other characteristics of the system (tempera-
ture, velocity, etc.) of which length is an attribute. To the extent that
I.
these characteristi~s are related to length and, to the extent that they
change. one w()uld~ of course, expect both the measuring rod and the
length of the object being measured to change. Similarly, the measure-

ment of any psychological or social characteristic presupposes a con-
stant set of known conditions among the factors relevant to it and to
the process of measuring it. Unfortunately, one's knowledge and one's
attempts at cOntrol are seldom completely adequate. As a consequence,
the results of measurement reflect not only the characteristic being
measured but also other unknown factors that affect both the char-
acteristic being measured and the process of measurement.
Thus the variation among individual scores on a measuring device
administered to a group of subjects arises from a number of different
contributing factors. Some of the variation may be conceived of as
being due to true differences among the individuals in the character-
istic being measured; some of it represents "errors" in measurement.
The basic problem in evaluating the results of any measurement is that
of defining what shall be considered true differences in the character-
istic being measured and what shall be considered as variations due to~
error in measurement,!
Let us briefly consider some of the possible sources of differences
in scores among a group of individuals: -
1. True differences in the characteristic which one is attempting
to measure. In the ideal measuring situation, all of the differences in
scores among individuals would be due to their differences in the char-
acteristic one is attempting to measure. For example, if one were at-
tempting to measure attitude toward religion, all of the differences in
scores would be due to the individuals' differences in this attitude;
none of the differences would reflect chance variations or the effects
of other attitudes. '
2. True differences in other relatively stable characteristics of the
individual which affect his score. Few of the techniques available to
/
the social scientist provide "pure" measures of any given characteristiC.
Such general variables as intelligence, education, information, social
status, and various personality characteristics frequently "contaminate"
the results of an attitude questionnaire or of an observer's ratings.
Hence the scores of the individuals in a group will reflect not only
differences in the characteristic being measured but also differences
1 For a more detailed discussion of this approach to the sources of variati"ons
in scores. see Thorndike (19~9).
VARIATIONS IN SCORES 151
in other characteristics. For example, Edwards (l957b) has shown that
the number of people who accept or agree with a given statement on
a questionnaire is highly correlated with the "social desirability" of the
position presented in the item. Thus, differences in scores on instru-
ments which ask the respondent to indicate his agreement or disagree-
ment with statements that are subject to considerations of social de-
sirability may reflect differences in willingness to admit holding
"undesirable" positions, as well as differences in the characteristic
the instrument is intended to measure. For example, scores on an
attitude test may be influenced not only by individuals' attitudes
toward the object in question, but also by their willingness or unwill-
ingness to admit holding opinions. they know to be unpopular. Sim-
ilarly, differences in scores on tests of "personality" may reflect
differences in willingness to admit 'having feelings that are generally
considered "neurotic" or behaving in ways that are socially disapproved,
as well as true differences in the feelings and behavior asked about.
3. Differences due to transient personal factors. VariQus personal
factors such as mood, state of fatigue, health, mental set, distractibility,
etc., may vary even within a short tperiod of- time. For the most part,
one would expect the state of the person to exert its influence on his
responses primarily through the way he defines the situation 'of meas-
urement. For example, if he is fatigued, his response to the measure-
ment situation may be: ''I'll get it over with as quickly as possible; it's
too much of a bore to bother with.'~ Under appropriate conditions of
motivation and rapport, these transient personal factors often have
negligible effects. ,
Measuring instruments differ in the extent to which performance
on them is affected by transientpersonalfactors. The usefulness of an
instrument for measuring some characteristics otber than these trans-
ient ones is, of course, oecreased to the extent that scores on it are
influenced by such factors.
I
4. Differences due to situational factors. Variations in the situation
in which measurement tkkes place often playa large role in contribut-
ing to the differences in scores among a group of individuals. For ex-
ample, an interview witl;i, a housewife'may'be markedly affected by the
presence of her husband: The anonymity or lack of anonymity provided
by the situation, the rapport or lack of ease, the seriousness or playful-
ness, the various distractions, etc., all tend to affect responses of the
subject. If the situations of measurement vary from individual to in-
dividual or from one measurement to another, a considerable variation
in scores is likely to result from such factors quite apart from the true'
differences among individuals with respect to the attribute being
measured.
5. Differences due to variations in administration. Inadequate and
nonuniform methods of administering a measuring instrument may
contribute to variations jn scores. Interviewers may add questions,
change wording, revise the order, omit questions, etc., in such a way
as to make one interview noncomparable with another. A bored test
administrator may improvise his own instructions; a satiated coder
may glance at rather than read the item to be coded; a tired observer
may not be able to keep recording the constantly changing 'group
process. All of these variations in the use of a measuring instrument
may markedly affect both the consistency with which a given coller,
obs~rver, etc., rat~ the responses of various individuals and the con-
sistency of rati-qg fkom on~ coder, observer, etc" to another.
, Bothl the !situation
< t', '
iii which the measurement is made and the
method of ad~inistiatlon may influence the orientation with which
the subject answers-for example, whether he responds in terms of what
he believes to be true, of what he thinks the measurer considers the
"right" answer, etc.
6. DiHerences due to sampling of items. Any measuring instrument
necessarily taps only a sample of items relevant to the charact~ristic
being measured. Thus, an attitude questionnaire contains only a rela-
tively few' items' 'from the universe' of relevant items 'that m~ght haVJ
heen included. If we conceive of a score broadly, as a measure of at'
htude, rather than narrowly, as the score on a specific qu~stionnaire,
jt is apparent that the variations in attitude as masured by different
questionnaires will be, in part, dependent on the nature of the sample
-of items included in the questionnaires. For example, in one question-
naire dealing with attitudes toward Negroes, the particular items in-
cluded may happen to be those on which a given individual is more
jlikely to respond f~voiably than he would on another questionnaire
.consisting of a different sample of items.
It is obvious that, if other things are equal, a one-item question-
:naire is likely to be fl less adequate sample of the total universe than a
VARIATIONS IN SCORES 153
questionnaire with thirty items. Similarly, ratings based on a few ob-
servations or made by a single observer are not as trustworthy as ratings
based on many observations by several observers. Increasing the number
of items (provided the added items are equally appropriate to the pur-
poses of the given questionnaire), or the amount of relevant material
on which a score is based, makes it likely that the variation in scores
attributable to this source will decrease.
7. Differences due to lack of clarity of the measuring instrument.
If individuals understand the items in a measuring instrument differ-'
ently, variations in their responses may reflect these differences in inter-
pretation rather than true differences in the characteristic one is
3ttempting to measure. Frequently the categories in a coding or
observational instrument are complex and ambiguous; different coders
or observers may interpret the categories differently and assign similar
responses to different categories~\'Interview questi.ons may ~e so long,
or phrased in such a complex way, that some respondents do not un-
derstand them; the responses of these subjects can hardly constitute
an adequate indication of the characteristic.or attitude at which the
questions were aimed. Words such as free enterprise or liberty, which
are emotionally colored or which have special connotations not com-
mon to all people measured, may set off differential reactions not di-
rectly r~lated to toe characteristic which the instrument aims to meas-
ure. Even apparentlysimple questions may be unclear if their context
is am'Qiguous. Take, for example, t~e following question used in a
survey of a coll~ge community: :'Duririg the last week, did you visit
the home of any faculty member?" If the interview took place im-
mediately after a week of vacation, some respondents might interpret
the question to mean "during the last regular week of classes," others
as "the preceding :week-i.e., during the vacation." Simplicity, con-
creteness, and a high degree of specificity are to be desired in measurillt;
instruments. 2 i
8. Differences, due to mechanical factors. Circumstances such as
broken pencils, check marks in the wrong. box, poorly printed instruc-
tions, lack of space to record re5ponses fully, play their role in prevent-
ing the most effective functioning of a measuring instrument. MallY
2 This statemrnt does not apply when the characteristic the investigator is
trying to measure is the way in which a subject interprets an ambiguous situation
-as i< the case in m!?~t Dfoiective techniques (see Chapter 8).
sources of error may be eliminated by adequate attention to mechan-

ical factors in the presentation of stimuli and the recording of responses .
It is impossible to list the many mechanical details that must be given
thought; they range from the layout of a questionnaire to the hand-
writing, or typing, of an interviewer.
9. Differences due to factors in the analysis. Commonly overlooked
is the possibility of errors in the processes of scoring, tabulation, ma-
chine analysis, statistical computation, etc. These processes can be
easily checked, but unless they are, large differences in scores due to
such errors may be introduced into the data.
These, then, are some of the major factors that influence the
results obtained from any measurement process. An examination of
the list above indicates many sources of "error"-i.e., many _influences
on the score other than the influence of the characteristic one is trying
to measure. It is ,customary to classify errors as constant (systematic,
or biasing) or random (or variable). A constant error is one introduced
into the measurement by some factor that ..systematically affects the
characteristic being measured or the process of measurement. Factors
such as those discussed in (2) above are of this type. When such a
factor is unnoticed, its effects are not taken into consideration in evalu-
a,ting the results of the measurement. Random error is due to those
I tia~si~nt aspects of the person, .of the situation of measurement, of the
m~asu'rementpror;edure, etc., that are likely to vary from one measure-
ment to the next, evenJ though the characteristic one_ is trying to
me~sure has not changed. A random error reveals itself in the lack of
cOl1si~tency of repeated or equivalent measurements of the same per-
son, object, or event, or of the same group of persons, objects or events.
As will be seen in the following sections of this chapter, estimates' of
validity are affected by both types of error; estimates of reliability
usually take into account random errors only.
The Validity of Measurements

Certain basic questions must be asked about any measuring in
strument: What does it measure? Are the data it provides reievant to
the characteristic in which one is interested? Do the differences in
THE VALIDITY OF MEASVREMENTS 155
scores represent true differences on the characteristic one is trying to
measure, or do they reflect also the influence of other factors?
The va!_!diry of. a m~suring instrument may be d_efined as the
exte~hicb differences in ~c~~es <?n _it reflect .!F.~~iff~_re.nf~l1g.
individuals, ~oup2_, ~r_ situ~ti.2.~s ip the charg_cteristic w~ich it seeks
to measu~or true differences in the same individual,z .group~ or sit-
errors.
-
uation from one occasion to another, rather than constant or random
.... - -- - -- - -- . _ , . , _ .
Several of the factors discussed in the preceding section as con-

tributing to variations among individual scores may lead to constant
errors. It is obvious that true differences among individuals on endur-
ing characteristics other than that measured by the test, which affect
scores on the measuring instrument, will produce constant errors. So,
too, may factors in the instrument itself or in the situations in which
it is administered. For example, measurement procedures that rely
heavily on the complex intuitive processes of observers frequently have
constant errors introduced by selective perception, recall, or recording.
Note that the constant errors under discussion apply to individ-
ual scores and, by implication, to comparisons of averages of groups of
scores. For instance, suppose that we administer a test of racial prej-
udice to children in a setting where prejudice is' frowned upon. There
is, in this situation, comparatively little pressure on the unprejudiced
children to distort their answers to the test but a good deal of pressure
on the prejudiced children to give unprejudiced answers. Thus the
score of an unprejudiced child may refle~t his true attitude quite
accurately, while that of a prejudiced child may be shifted to some
unknown degree from his true position. As a consequence, both the
prejudiced and thT unprejudiced children will tend to come out alike
-equally unprejudiced. Repeated administrations of the test or of
parallel forms of it under the same conditions will not cause the dis-
tortion of the replies of the prejudiced children to show up as random
error, since there {vill again b.e distortion in the same direction. Sup-
pose now that we want to use. the test to evaluate the effectiveness of
a course in human, relations, by comparing the scores of children who
have had the cou;se with a control group who have not. If both prej-
udiced and unprejudiced children give unprejudiced answers, the test
cannot reveal any effects which the course may have had.
Since we usually do not know an individual's true position on the

variable we are attempting to measure, there is no direct way of deter-
mining the validity of the measure. (If there were some other source
of information as to true position on the variable, there would often be
no need for another measure of the variable, unless the available
method of measurement was extremely cumbersome or expensive or
in some other way impractical or inappropriate to use.) .
LIn the absence of direct knowledge of individuals' true position
on the variable being measured, the validity of an instrument is judged
by the extent to which its results are compatible with other relevant
evidenc~. What constitutes relevant evidence depends on the nature
and purpose of the measuring instrument. The purpose of SOJIle tests
:is to provide a basis for specific predictions about individuals; for ex-
ample, whether they will be successful in a certain type of job, whether
they need psychiatric care or are likely to need it in the future, etc.
Other tests, however, are not used in this way. Although they ar~ de-
signed to measure specific characteristics of individuals, they do not-
at least in the present state of our knowledge-lead to definite predic-
tion~ about how individuals will. function in given situations. ThIs
difference in the purpose of tests leads to a difference in the type of .
~vid~pce considered relevant for estimating validity. In the case of
fests intended as a basis for predictions in terms of some specific cri-
terion (such as success on a particular job), evidence about individ~
uals' position on that criterion provides a basis for estimating the
validity of the. tes( Investigation ~f validity in these terms ~ay be
described as pragmatic; validity is jll_dged in terms or the accuracY-2!
predictions made on thebasisortli; te~t-;; ;~~i_~.
InstrJ)ments designed to measure characteristics that do not lead
to sPecific predictions cannot be evaluated so. directly. Other evidence
must be sought to provide a basis for judging whether the instrument
adequately measures the concept it is intended to measur~.1'hk-Iess
direct approach has been described as construct validation. These
'.' approaches to valid;tion arc diSZ;;ssed in th~-following ;~ctions.3 .
,
3 For somewhat different, but related, ways of describing approaches to vali-
dation, see Cronbach (1949), Cronbach and Meehl (1955), and Technical Rrm-
mendations for Psychological Tests and Diagnostic Techniques (1954).
THE VALIDITY OF MEASUREMENTS 157
_-
PRAGMATIC V ~ldDITY 0
One approach to validation is to ask: Does this measuring instru-

ment work?' Can I male decisions better with its 'help than ~ithout it?
Does it h~p me 'fo distinguish individuals in_!erms of some -criterion?
-The investigat~r "ffiay wish, for' e~mple, to distinguish between indi-
viduals who, at the time of measurement, are "well adjusted" and those
who are in need of psychiatric care.~ test thaLh..clps him distinguish
indiv!gm!ls who differ. in their present status is ~ai_d to have concurrent
. . .vali(flty. Onthe otherhanc1~ the in~;tigator may wish to predict which
~e likely, in the future, to-need psychiatric care. The ade-
quacy of the test for distiQguishing individuals who will diff;In the-
-----tuture
,.___.> - -
In both_ui.stances,...the_ap..:.~_
may be called its predictive validity. __...........
proach to validatio!!~p"ragm~ one.~
~gm.?Jic <1:p'p'roach t~lid~~y', the int_e!:esUsjn_the_useful-
E..ess_.Q.lthe.measuring instrument as an indicator or a predictor of some
other behavior or characteristic of the individual. ~e, tests
that requ"i're th~ individual to reproduce a complex design by means
of blocks have been shown to be useful in identifying individuals with
organic brain disorders]rhe test-user is not interested in the individ-
ual's design-reproducing ability per se; he is interested in performance
on the test only as an indication of possible brain damage. He does not
need to know why the test performance is an effi"cient indicator of the
chafl>cteristic in which he is interested. 4
LJ\'hat is essential in this approach to validation is tliat there
be a reasonably valid and reliable criterion with which the scores on the
measuring instrument can be comparecf]rn general, the nature of the
predictions and the techniques available for checking them will deter-
mine what criteria are relevant. In validating the design-reproduction
test as an indicato~ of ,brain damage, for example, the obvious criteri3
would be other w~ll-established indicators of brain damage, or post-
mortem examinations, to' the extent that these can feasibly be used.
In the problem suggested earlier, of identifying individuals in need of
psychiatric care or' likely to need such. care in the future, the mest
adequate criterion :presently available is probably that of independent
4 However, from ihe point of view of advancing scientific knowledge, as distinct
from the ability to make useful predictions, he is likely to be interested in discov-
ering why his test works. This point will be discussed in more detail later.
diagnoses by 'competent psychiatrists, but this is not an infallible cri-

terion. If the purpose o~ ~t is to. pr:edict success in-college,._t~
sritexiQn i...likely_ tQ~be. cqllege grades. If the aim is prediction of job
"<
success, one may have to rely on supervisors' ratings as a criterion, un-
less the job is one in which more objective records of quantity and
quality of output can be secured.
Ddeally, of course, the criterion with which the scores on the
measuring instrument are compared should itself be perfectly valid
and reliable,jChecking predictions against a criterion which may be
irrelevant, inaccurate, or unreliable provides a dubious evaluation of
the measurement procedure. In practice, however, the investigator
frequently finds that no thoroughly tested criterion is available; he
selects the one that seems most adequate and tries to keep in mind its
limitations and, if possible, to supplement it by additional criteria.
The development of an adequate criterion for checking predis-
tions and thus evaluating the usefulness of measuring instrumen~s is
an important part of research, which seldom receives the time or atten-
tion it deserves. Frequently, even though no perfect criterion is avail-
able, tne reliabiljty and validity of available criteria can be improve.4J
In a study in wh'ich supervisors' ratings are to be ~sed as the criterion
for validating a test predicting success in a g~v~n job, ~~mple,
different. s,up~~ors may use suc1iClifferentbases of judg~n~ that
-their:_ratings are not comparable. The reliability and validity of their
riling;-can be increased by l'earJUf specificati~; of the kinds of be-
- havio~__!1gta.r_e...to..b.e"Tco~sidereg irr' raJingjob ,p-'e;formll~~~~ri!Y
~~fining points on the rating scale~ by providing a traIning period
_in WhTc1i two or more sll:perviso-rs rate the same individuals' and then
discuss discrepancies in ~heir placement, etc. Generally, pooling of
ratIngs by two or more observers who are rating the same subje~s
tends to increase both reliability and ~alidity.5
CONSTRUCT VALIDITY
'(frequently, however, the investigator is interested in the test per-

formance not as a simple predictor of behavior but as a basis for infer-
5 For further discussion of methods of improving the reliability of ratings, see
Chapter 10; pages 352-353.
ring the degree to which the ind_iyidual Rossesses some. characteristic
presumed T<'f"'beTeBected in._tb_e test performall.~he presumed_
_-
characteristic being reflected is not something which can be .point._<:!d_
to or identified with soIiIe'spe'Ciflclfnd
_._.,
6
~... . of beh~vior; rather, it is an ab-
straction, a construct. Therefore the process of validating this kind
~ing-instrumk.nt "'isi=ererfecf to as conStruct validatlon.
Many of the measures used in the social sciences deal with con-
c:!
~u~of intel~ig<:nceL.<?f~attitudes,_ ~~t_!1.orita~ia~lsm, of
introversion-eXtroversion, of ~i~~y.J or of more global personality pat-
terns, erc05f1lilS sort. qronbach and Meehl P955), who first~~
explicit the concept of construct validity, p-ointed out~he defini-
tions of such co_nsb.:ucts-CQ[!sis..Un_p_grt ~tS.2LP-.I2Rositions about
-'their relatiol1ships. to other via~les-other constructs or directly
~havim,.Thus, in examining constru~t validity, it Tsap-
propriate to as~ such guestTOii'Sas=: Whafp~edi~tIons would one make,
o""i11lle"15'iiSlSOf these sets of propositions~ about the relationships to
other variables of scores based on a measure of this construct? ~the
measurements obtain~d by using this instrument consistent with these
prealcbons? .~ ='" . - - -- --'
-----Three closely related points should be noted in connection with

these questions. ~t, the predictions are oLa sQm~nL
nature, and serve a somewhat different function(from those involved
li1cIetermining__llli!gmatic valiClity. Consider a prediction of how In-
dividuals will vote in a national election. A measuring instrument
may have the specific aim of making this prediction possible. In this
case, the interest is in the accuracy of the prediction, and there need
be no concern about the psychological attributes involved in the re-
lation between an 'individual's behavior on the test and his voting
behavior. But a preaiction about voting may also be made in connec-
tion with examining the construct validity of a test of political conserv-
atism. Here the reasening runs: "I believe that this test measures-polit-
ical conservatism (a construct). Given the generally accepted view
of the position of :American political parties, I should expect that
people who score a~ less conservative on this test will be likely to vote
Democratic; those who score as more conservative, to vote Republican."
6 See Chapter 2, pages 41-42, for a discussion of concepts and constructs,
terms which we have used interchangeaH:r
But there is no necessary expectation that the correlation between test

score and voting behavior will be very high, because it is recognized
that voting for one or the other party is not equivalent to being less
conservative or more conservative. The Democratic and Republican
parties do not clearly represent lesser or greater conservatism; in some
respects official Republican stands are less conservative than those of
Democrats. Moreover, even if the two parties were unequivocally
identifiable in this respect, one would expect other influences in addi-
tion to the individual's conservatism to enter into the determination
of his vote: family tradition, religion, socioeconomic status, voting
intentions of his friends, expectations of specific gains if one or the
other party wins.
Second, in the case of pragmatic vali~a.tion, the ability of the
measure to distinguish in ter~f1h~le criterion toward whi"ch it
is di~ected is th: t~st of its validity; ~1;:, case of constrEct validatlop,
~ pr~d~~~tIl?! would be ~a~on tl!e bas~2f Jh~s~
~ltions, in w~!ch the cons!!!!ct i.~E.~?!~~d.lor as many of them as
possible) enter into the consideration of validity. Thus, in the case of
the conservatism measure, besIdes predicting--;hether an individual
will vo~e Derriqcratic or Republican, one might predict preferences
for candidates within a given political party; for example, one might
predict that Republicans who score as less conservative on the test win
favor a given candidate, those who score as more conservative, anot~er
candidate. One might also make and test predictions about relation-
ships to socioeconomic status, to education, to stands on specific issues.
Failure to confirm anyone of the prydictions would, of course, call into;
question the validity either of the measure or of the underlying hypo-
theses. However, even if each of the correlations proved to be quite
low, their cumulative effect would be to suppqrt the validity of the test
'and its underlying theory.
":...Ibir.d, 2_aminati<;m of construct validity i~alidation not
oril y of the I!?-~asuring instrument "'but ot the_theory underlying it . .!.L
'th~'p;;dictions ;-renot supported~v~stigrtor iilliJ.""liaye no cl~r
guide as tQ..1:Y!!~r. ili~hod'comlllg is ii{th;~easur~g instrume~t
, or in the theory. Consider, for example, a study focused on the l1ypoth-
7sis th;t-P;;r~-;;~al association with members of an ethnic group other
than one's own is likely to lead to more favorable attitudes towarp
that group. Suppose that the findings do not show the predicted reo
lation between extent of personal association and attitude change.
Shall the investigator conclude that his. measure of attitude was not
valid, or shall he conclude that, under the conditions of the study, the
hypothesis was incorrect? He will probably be led to re-examine the
construct attitude and the network of propositions that led to his
specific prediction. The result may be a refinement of the construct,
with more detailed hypotheses about its relation to other variables, and
changes in the measuring instrument. For example, a number of in-
vestigations focusing on the hypothesis that personal association with
members of an ethnic group other than one's own is likely to lead to
more favorable attitudes have come out with quite different results.
Analysis of a ~umber of these studies indicates that the differences in
results may be accounted for partly in terms of the specific aspects of
attitude tapped by the different measuring instruments, and suggests
that the hypothesis might be refined to specify which aspects of at-
titude are most likely to be affected by personal association.
~p.h.elland Fiskt: .(959), have su~estec!_thatthe investigation
of construct validity can be made more rigorous by increased at_t~!!.!!2!!..
to the adequacy of the measu~ of, the construct in_. gtI(~s_tiO!l..1_bfore
its relationshi s to er_variables_are..considered. They propose that
a kinds of evidence about a measure are needed before one is justi-
fied in examining relationships to other variables: (1) evidence that
different measures of the construct yield similar results, and (2) evi-
dence that the construct as thus measured can be differentiated from
other constructs. In order to secure such evidence, one must measure
the construct in question by two or more diff~rent methods; one must
also measure the characteristic or characteristics from which he wishes
to differentiate his construct, using the same general methods he has
applied to his centralI construct.
For example, Burwen and Campbell (1957) were interested in an
assumption commori to several different psychological theories-that,
on the basis of early experiences within the family, an individual de-
velops a generalized attitude toward authority figures. Before attempt-
ing to study the reJation between "attitude toward authority figures"
and early family experiences, they concentrated on developing a num-
ber of measures of "attitude toward authority figures." Ratings of
attitude toward authority figures were made on the basis of each of the
following methods: interview questions about father and present
superior officers (the respondents were Air Force trainees); a list of
traits to be ,checked as descriptive of father and of immediate superior;
written character descriptions of photographs of middle-aged and older
persons (intended as symbolic authority figures); stories about scenes
containing symbolic authority figures; an autobiographical inventory; ,
an attitude survey; and a sociometric questionnaire.
Each of these methods was also used to measure a second charac-
teristic: attitude toward "nonauthority figures" (present colleagues, a
past fellow worker, "symbolic peers" represented by pictures of young
persons). This second characteristic-attitude toward nonauthority
figures-was measured in order to determine whether the attitudes
expressed toward authority figures were indeed specific to persons in
authority or whether they were expressions of attitudes toward people
in general. If there were a high positive correlation between attitudes
expressed toward authority figures and toward nonauthority figures-
that is, if individuals favorable toward authority figures were' also
favorable to nonauthority figures and those who were unfavorable
toward one wer~ also unfavorable toward the other-one would con-
clude that what was being tapped by the first group of measures was not
a specific attitude toward authority figures but a more general attitude
toward people. On the other hand, if there were little or no correla-
tion, or a negative correlation, between the measures of the two types
of attitudes" one would conclude that the first set of measures was in-
deed getting'\at attitudes specificallY,directed toward authority figures.
As it t~rned out, the measures of attitude toward authority figures
showed so 'little agreement that there seemed no basis for believing
that any consistent attitude had been tapped; thus there was no point
in trying to determine whether these 'measures, were getting at a spec,ific
attitude that could be distinguished from attitude toward nonauthority
figures. Ratings made on the basis of the interviews showed a high
correlation between attitude toward father and attitude toward supe'
rior Dfficers; had this been the only method used, the investigators
might have concluded that they had successfully measured a general-
ized attitude toward authority figures. However, the ratings based on
different methods showed little agreement with one another; more-
over, techniques other than the interview showed little correspondence
between attitude toward father and attitude toward superior officers.
In such a situation, one faces the question whether the measuring in-
struments are invalid or whether the construct one is attempting to
measure (in this case, "attitude toward authority figures") is somehow
faulty. In this study, the investigators reasoned that the number of
different methods they had used provided a basis for concluding that
the difficulty was with the construct rather, than the measuring instru-
ments. Although they recognized that anyone or more of the measures
might have been invalid, they thought it unlikely that all of them were
inadequate indicators of the construct "attitude toward authority."
In view of the fact that no two of their measures showed high agree-
ment, they concluded that their findings required a modification of
the assumption that each individual has a generalized attitude toward
authority which reflects his attitude toward his father.
From this discussion, it is apparent that construct validity cannot
be adeguatel>; tested by_a_Dy_singk.l~rocedure. Evidence from a number
'of sources is relevant: correlation with other tests and with other
~ehavior, internal consistency-of ite1!l~ ,stability over ti;;e;~' 'Ho~
eVIdence-from each of these sources bears on estimation of the validity
of the test depends on the relationships predicted in the theoretical
network in which the construct is embodied. The more different rela-
~p-s tested..allci.coufirmed,jhe. ,greater the support both for the
measuring instrument and for the unded)illJgjJ1~r.i.' . -
RELATIONSHIPS BETWEEN DIFFERENT ApPROACHES TO VALIDATION
....The "pragmatic" and "C~C1!Rp'roaches t~ation are not

_ElUtually exclusive. A ~le test or measuring in~trument maybe used
for a number of diffe~_t_p~urposes, aEd fQLellch .p}1rppse the appro-
~estilw-validity.mustb~<:!mployed. As already pointed
out, estimates of 'Rragplatic validity may enter into the evaluation of
constrUctva}i~n !the other hand, alJnough-m@y_ m~a_s~s shown
to have pragmatic validity have been arrived at purely on a trial~and
error basis, there is nOI reason for not mvesti'gatini'why these measures
work-that iSt-~onside~ing_.ilie. g>nstructs invglveq a_n~~i.r "relation
to the crit~iQn yariables. Such investigations may lead to the clarifica-
i
l64 GENERAL PROBLEMS OF MEASUREMENT
tion of concepts and eventually to the construct validation of these

measures or to the development of alternative measures which have
both construct and pragmatic validity.
In fact, there are good reasons for not remaining content with a ,
measurement procedure that has been validated only pragmatically. As
long as the why of its working is not understood, one has no assurance
that the conditions of its working stilI hold for any particular applica-
tion, and with every application one can only proceed with a naive
faith that it will still work this time. Moreover, to the extent that the
pragmatic approach limits itself to discovering empirical correlations
without any concern for an underlying theoretical explanation of the
relationships, it is an uneconomical procedure. It permits no general-
ization to other problems; it results in knowledge which is isolated and
barren rather than interconnected and logically fertile (Margenau,
1950).
Is THE V ALlDITY OF A MEASURE EVER SELF-EVIDENT?
In the examples we have been discussing, t],.1e ..data provided by the

measuring instruments have. been. used as an indicator of some attribute
~f th~ individual which~is~not .itselL.directly"'mea~~reLThus, scores
based on answers to a questionnaire may be ~sed to diagnose or to pre-
dict mental illness, or to infer the extent to which the individuai
possesses some characteristic not directly revealed in the test. Bul
sometimes measures are based directly on behavior of the kind ~n which
the investigator is interested. Perfprmance tests are frequently of this
sort: reading speed is measured by computing how much of,a passage
a person reads with comprehension in a given time; ability to solye
arithmetic problems by success in solving a sample of such problems;
job performance by rating the quantity ar~d quality of the work pro-
duced. Observations of behavior may also have this characteristic, if
they are used descriptively rather than as, a basis for inferences about
underlying dynamics. Thus, an investigator interested in studYIng the
behavior of a group leader may record a sample of his behavior; one
I interested in interaction among group members may record the ex
changes among them.
"'" _ ----- _- -
Such measures, which focus directly on behavior of the kind in
...
~ich the tester is interested,. are often said to have "face validity"; t.bet
)Sl the relevance of the measuring instrument to ~J.Qne. is trying to
measUreTs apparent "on the face of it." Whether such an assumption
IS Tusfified in any gi;en~ case is ultimately a matter of judgment. But
in making this judgment, two major questions must be considered: (1)
whether the instrument is really measuring the kind of behavior that
the investigator assumes it is, and (2) whether it provides an adequate
sample of that kind of behavior.' Frequently, in the case of achieve-
ment and proficiency measures-where consideration of "face validity"
is most appropriate...::..one is''justified 'in !the'assumption that the be-
havior which appears to be involved in the test is the behavior it actually
measures. For example, if one is interested in evaluating the adequacy
of an in~~3~~p-~rformance.as a stenographer, ana!ysis of theJ~_tters
she types, from the point of view of accuracy, spelling, neatness, speed,
etc., w~9.-seeffi to prov~de clearly reieV"ant -evlaence. Ifowever, it
may occasionally happen that a test which appears to measure one
kind of behavior is in fact measuring another. Suppose, for example,
that we give a set of arithmetic problems to a group of eighth-graders
'md obtain a wide range of scores. We think that we have measured
performance in arithmetic. But it may happen that the arithmetical
operations involved can be performed by all members of the group; the
differences in scores may stem from differences in/ability to understand
the language in which the problems are presented. The investigator
must always be alert to possibilities of this sort.
The second consideration-whether the test provides a.n adequate
s;:tmple of the kind of behavior with which it is concerned-reguires, in
principle;a~complete specification of the u;;iye~~e of behavi; i~ q~es
tion and of allpossible test items that might be used to measure It. Fre-
quently this is impossible, since the number of potential test items
may approach infinIty. What is possible, and essential, is careful con-
sideration of exactly,I what the behavior is that one wishes to measure,
and of the. variety of ways in which it might be measured. Suppose, for
example, that one wishes to test reading comprehension. Obviously,
one cannot assemble all the material that has been written in a given
langu~ and select;a sample, either randomly or on any other basis.
The investigator rr1ust make a selection without knowing the total
universe. But he can, and should, consirler whether the passages in-
eluded in his test refer to topics which may be more familiar to some
individuals than to others, and which may thus test knowledge of the
topic rather than reading comprehension; whether they involve peculi-
arities of style that may present more difficulty to some individuals
than to others; etc.
The Re1iabilityot Measurements7

As pointed out earlier in this chapter, scores on measuring instru-
ments usually reflect not only the characteristic which the instrument
is attempting to measure, but a variety of constant and random errors.
The evaluation of the reliability of any measurement procedure con-
sists in determining how much of the variation in scores among indi-
viduals is dut.:.to inconsistencies in measurement. When independent
but comparable measures of the' same thing are obtained, they will
yield the same results to the extent that the measurements an! free
from random-or variable errors.
If we knew that a measuring instrument had satisfactory validity
for the purpose for which we intended using it, we would not need
to worry about its reliability. If an instrument is valid, it is reflecting
primarily the characteristic which it is supposed to measure, with a
minimum of distortion by other factors, either constant or transitory~
thus there would be little reaSOn to investigate its reliability-that is,
the extent to which it is influenced by transitory factors.
However, an investigator is seldom in the position of knowing in:
advance that his measure has satisfactory validity, unless this has been
demonstrated in earlier studies concerned with the same characteristic.
Moreover, it rar~ly happens that an instrument can be shown to have
such high validity that no improvement is needed. In the case of prag-
matic validity, the coefficient of correlation with the criterion measu~e
is usually substantially below 1.00; the lack of complete correlation may
oe due to the fact that the instrument does not measure exactly the
characteristic reflected by the criterion measure, or to variable errors.
III the criterion, or to'variable errors in the measuring instrument. In
7 For more detailed discussions of reliability, see Gulliksen (1950), Guilford

(1954), and Tryon (1957b).
THE RELIABILITY OF MEASUREMENTS 167
these circumstances, it is important to determine the extent of variable
error. in the measuring instrument-and also in the criterion measure,
if that is possible. In the case of construct validity, no simple direct
determination of validity is possible; in these circumstances, evidence
')f the extent of variable error is a necessary part of the evidence con-
t:erning validity.
Unless satisfactory validity or reliability has already been demon-
strated, the reliability of a measuring instrument should be determined
before it is used.in a study, rather than after. If a research instrument
is plagued by variable error, the likelihood of achieving significant re-
sults is minimized. Rather than go ahead with unreliable instruments,
it may be prudent to delay the research and try to increase their re-
liability.
METHODS OF DETERMINING THE RELIABILITY OF MEASUREMENT
Evaluation of the reliability of a measuring instrument requires a

determination of the consistency of independent but comparable
measures of the same indlVlouar,g;oup, or situation. Clearly, it would
be desirable to have many repeated measurements of the same indi-
viduals, under the same conditions, as a basis for estimating random
errors of measurement. In the study of human ;behavior, however,
this is not often feasible. Not only may oft-repeated measurement
create annoyance; it may also affect the characteristics one wishes to
meas~re. When this is likely to be the case, reliability may be estimaten
on the basis of as few as two measures for each individual in a sample
of the population on which the measurement device will be used-or
even on the basis of one measure if it can be subjected to internal analy-
sis. Enough measuremen'ts to provide an adequate basis for evaluation
are obtained by increasing the number of individuals measured rather
than the number of measurements
I
of each individual. The usual pro-
cedure in computing reliability involves calculating some index of
agreement between the results of the repeated measurements.
Different methods bf estimating reliability focus on different
sources of variation in scores. Some are concerned with the stability
of individuals' position from one administration of the measure to
another; in other words, they focus primarily on fluctuations in the
characteristic being measured or on change in transient personal or

situational factors. We shall use tne term stability coefficient to refer
to this type of estimate. Others are concerned with the equivalence of
individuals' position on different instruments intended to measure the
same characteristic, or on a given instrument as administered by dif
ferent people at essentially the same time or as scored by different
judges. In other words, they focus primarily on unreliability due to
sampling of items or to variations in administration or analysis. Still
others.are..concerned with both stability and eq uivalence. 8
0A~~LI;YThe .stability of r~sults of a measuring instrument is d~
termm:ed on: the baSIS of the conSIstency of measu!:es on reReated "<lFP!I-
cation~_r.tis.important, however, to distinguish between inconsistency
due-to genuine changes in the characteristic being measured and in-
consistency due to changes in extraneous factors. The characteristic
being measured may fluctuate -from one application of the mea~ure to
~nother. This is true of such physical phenomena as temperature, blood
pressure, the weight of individuals. Many of the phenomena with
which social science is concerned-for ex;rmple, leadership behavior,
attitudes, morale-show such variation. Inconsistency of this type
should not be interpreted as unreliability of the measuring instrument,
but it complicates the problem of determining the stability of the in-
strument itself.
Even though there are genuine fluctuations in the characteristic
we are attempting to measure, it is often reasonable to assume that
there is some "typical" position for a given individual, object, etc,
around which the fluctuations centter. Such an assumption is mad<for
example, in computing the mean temperature in a given city during
a given season, or in determining an individual's weight. It is a frequent
assumption in dealing with social and psychological characteristics of
individuals or groups. When we are interested in determini~g this
"typical" position, we must consider the extent to which any given
measurement is likely to-deviate from iand"hQ.W many meaSUrements
may bt; needed t~ yIeld "~~~"ble averag~timate\ In this sense we
8 The consideration of reliability in terms of stabilify a;d equivalence of scores
was suggested by Cronbach (1951). Earlier terminology (test-retest.lT\ethod, etc.)
reflects the fact that procedures for estimating reliability originated, ,for the most
part, in connection with the development of intelligence and aptitude tests.
may be concerned with the stability of measures even though the issue
is not one of unreliability of the measuring instrument.
Inconsistency in repeated measurements may also, of course, be
due to inadequacies of the measuring instrument. A scale that shows
an individual's weight:as 120 pounds and two minutes later shows the
same individual as weighing 140 pounds probably has some mechanical
defect. An observer mayan one occasion categorize a long answer by
a teacher to a student's question as "discussion," on another as "lec-
ture." An individual may reply differently on two occasions to the same
item on an attitude scale, even though his "attitude" has not changed.
The appropriate method for determining stability is comparison of
the results of repeated measurements. This is true whether the source
of instability is genuine fluctuation in the characteristic being meas-
ured or random error due to inadequacies of the measuring procedure.
\Vhen the measuring instrument consists of observations, a consider-
able number of repeated observations may be made. When it consists
of an interview, questionnaire, or projective test, usually only two ad-
ministrations are used.
Let us i1l11strate both of these procedures. Suppose we are inter-
ested in the percentage of time which a particular group leader spends
in lecturing, as compared with other possible activities. A single ob-
server may be assigned with a stop watch to observe the leader for a
fifteen-minute period while he is in charge of the group and record
the amount of time he spends lecturing during this period. From this
record the percentage of time spent in lecturing may be easily calcu-
lated. Then the same observer may make a similar record of the group
leader's behavior on another occasion when he is in charge of .the
same group. When the process is repeated several times we have a
series of numbers, each representing the percentage of time which
this group leader devoted to lecturing on a particular occasion, accord-
ing to this observer. We may take as our index of the stability of this
percentage the range 'of percentages, their standard deviation, or some
other measure of the Ivariability of the series. Lack of stability may be
due to variations in the leader's behavior or in the observer's recording
of it, or both. I,
\Ve may also be interested in the lecturing proclivities of a number
of group leaders. If each one is observed for fifteen minutes by the
same observer under comparable conditions, it is possible to order them

in terms of the proportion of time which each spends in lecturing. How
stable is this ordering? We could follow the procedure described in the
last paragraph to determine the stability of the percentage of time
spent in lecturing by each of the group leaders, but this would not tell
us anything directly about the extent to which the order of the group
leaders remained the same in this particular characteristic from observa-
tion to observation. The stability of ordering of individuals in a group
is usually measured by a coefficient of correlation or some other index
of agreement between the scores or ratings received by these individuals
on the first administration of a measure and the scores or ratings
received by the same individuals on a second administration of the
same measure. In the particular example we are discussing, the "score"
of an individual would be the percentage of time he was observed lec-
turing.
-- ~ ~
In the case of _al}. interview, questionnaire, or projective test,
-.~._..--
essentialIy the same procedure is used to estimate the stability of the
measure, except thafiE is usually applied only twice, in whitisknown
a~~eSt:Brocedure. The identical mtervIeW, etc.! is_g~o
the same Iildrvfciuals at different times under equivalent conditions,
and the results'dftfie two measur~~ents are comparea.Rowevei; in
-~---- -'~' -..
the case of measuring procedures such as these, which require a great
deal of participation by the individual subject, there are additional com-
plications. The very process of remeasurement may intensify differences
in transient factors; for example, anxiety, interest, and, motivation may
be lower during the second administration of the test simply because
the individual is, ~)ready familiar with it. To the extent that such
chaIigeSoccur, the test 'which is given--Onthe second adminIstration,
although objectively identical with the earlier one, may actually
/ ,
represent a quite different test situation. Moreover, the ~
~'re~Il)ber the re_spon~es he gave t~e_ first tes!_ (parti~ if the
subject may
,
time interval ~twe~n the two Tests is short) and, in the second test,
maygiveagain the responses he remembers (or misremem bers) having
made earlier rather than responses which are spontaneous or tnought
through anew in the second situation.
There is the further possibility that the initial measure has actually
changed the characteristic being measured. (The reader is reminded
of the discussion of the "before-after" experimental design in the
preceding chapter.) An interview, a situational test, an attitude ques-
tionnaire may raise questions a person has never thought about and
may heighten interest and stimulate the development of definite
opinions; thus, for example, a "don't know" response may be replaced
by a definite agreement or disagreement.
In addition to the possibility of changes brought about by the
initial measurement, there is-as with all types of measurement-the
possibility of genuine change between the two administrations of the
test. As a result of influences unrelated to the testing, some subjects
may have acquired more information, or undergone a shift in attitude,
during.th~ inter~al b~t1:~ DYQ admiI1istrationi <;lithe test.) -1..:.
When there is both the possibility that the initial measure may
affect the results of the second measurement and the possibility of
genuine changes brought about by other factors, the common practice
is to try to steer a course between waiting long enough for the effects
of the first testing to wear off and not long enough for a significant
amount of real change to take place. If the second measurement is
administered before the effects of the first have worn off, the estimate
of stability will not be trustworthy because the results of the two
measurements will not be independent; the error is likely to be in the
direction of an overestimate of stability. On the other hand, if genuine
changes have occurred, the resulting coefficient will be an under-
estimate of the stability of the instrument itself. No hard and fast rules
can be offered for judging the optimal interval; much depends on the
specific nature of the test. Fortunately, one can expect the effects to
wear off most rapidly at the beginning, with a decreasing rate as time
goes on.9 In other words, there are diminishing returns for waiting
over longer and longer periods of time. Two weeks to one month is
co~ly_conside~~ be a suitable interv~~ fo~ many psychologial
tests. If i_? douhl, Mw~.v~r, it _is better to w.!!iLa19l!g~_uqtheLthan a
sliorterperiod of time, since with increasing time such errors as occur
are likely to be in tIie dir~ction of unclerestimati"Z>n of the stability pf
theii1s1~umeiiE':"ratlier tha~_ ~ve~esti~ati~ One is safer ,,:ith a~
underestImate than, an overestnuate:" m--the former case, the mvesti"
9 S~e the curves of forgetting in any standard textbook of psychology.
gator knows that his instrument is at least as stable as the coefficient

indicates; in the latter case, he does not know where he stands.
In summary: The stability coefficient indicates the extent to which
the measure reflects relatively enduring differences among individuals
in characteristics that affect the measure.1 o To the extent that the
coefficient is below the maximum value possible, it is assumed either
that there are genuine fluctuations in the characteristic being measured
or that the measure is subject to random errors due to transient per-
sonal factors or other conditions that have changed from one adminis-
tration to the next. The stability coefficient does not take into account
the sampling of items as a source of unreliability. If\tlie personnel
administering and scoring the test are the same on the repeated fIP-
plications, the stability coefficient does not take account of possible
variations in the use of the instrument by different administrators or
analysts.
EQUIVALENCE. Estimates of equivalence concern the extent to
which, ~lfferent ~nvestigators using the instE':llle?t ~o measure ~
..._-._
saJ;Ile iPdiyiduals at the same time, or Clifferent instruments ,ap-pJied to
~ ----
the same individuals at the same time, yield consistent results.u We'
shall illustrate the first condition (different investigators ~ing the
same instrument) by our earlier example of the behavior .of group
leaders; the second (different instruments) by the examp-Ie of a test
of information, ability, or attitude made up of a number of items.
In our earlier discussion of the measurement of le~turirig pro-
clivities among group leaders we assumed a single observer who was
responsible for all the measwements made. But what if this observer
is biased, or otherwise unreliable? The notion of a reliable measurement
procedure requires that it yield comparable results from adqlinistrator
to administrator, provided each has been properly trained; in other
10 Note that these "relatively enduring differences" may be in characteristics
other than thoSe the test is attempting to measure; i.e., they may involve constant
errors.
11 It is convenient to use the phrase "at the same time"; however, the two
measures are not necessarily administered simultaneo.usly. In fact, in the case of
different instruments, it is often impossible to administer them literally at the
same time. What is meant is that, in estimating equivaleRce, the time interval be-
tween the measures is short enough so that there is no reasonable expectation that
the characteristic may have changed.
words, the sources of variation discussed in item (5), page 152,
should be minimized. ~e can estimate the extent of variation by
having different trained observers watch our group leaders at the same
time, and by having each record independently the time during which
he thinks the group leader is lecturing.
From such records, indices of equivalence can be calculated in
several different ways. If we have a number of observers watching a
single group leader, and. each reports independently the percentage of
time the leader spends in lecturing, we can use the range or the
standard deviation of these percentages as our index of the degree of
equivalence of the different observers. Another possibility is to divide
the period of observation into time units of equal length; 15 seconds
might be a convenient time unit for an observation period 15 minutes
long. Each observer would record for each time unit whether the
major activity of the group leader during that unit was or was not
lecturing. The index of equivalence for two observers would be the
percentage of time units for which they agreed that the group leadeI
was or was not lecturing.
If our major interest lies in the extent to which/different observers
agree on the ordering"of several group leaders (each observed for only
one session) with respect to proportion of time spent lecturing, we
would use a correlation coefficient as an index of the equivalence of
any pair of observers.
I
Estimates of equivfllence of individuals' position on different in-
struments intended to measure the same characteristic focus on varia-
tion in scores due to differences in sampling of items. There are many
possible items we might select to measure any given characteristic; to
what extent is the mea~ure of the characteristic we would derive from
one set of items equivalent to the measure we would derive from
I
another possible set? In discussing such estimates, we shall use as

illustration a test mad~ up of a number of different items to which
the subjects respond. L
The construction 'and selection of test items is a fairly arbitrary
procedure, and there is'a great range of possible items for measuring a
given characteristic; consequently there is usually little interest in
knowing the extent to which a single individual's responses vary from
item to item of a particular test. Indeed, most tests are constructed to

ensure a great range of responses for each individual; e.g., in, an
attitude test it is expected that every individual will find some items
with which he agrees and some with which he disagrees, and in an
achievement or ability test it is expected that every individual will find
some items that he can pass and SOme that he will fail. Interest centers
on the extent to which the ordering of individuals is the same from
one sample of items to another sample, and the index of equivalence
which is calculated is typically a correlation coefficient or some closely
related statistic.
The principles involved in estimating the effect of item sampling
on equivalence are most easily seen in the case of alternate forms
administered at the same time. In this procedure, supposedly equiva-
lent forms of the same test are given to the same individuals at the same
testing session. Although the two forms contain different items; the
items are intended to measure the same underlying characteristic.
The correlation between scores on the two forms of the test indicates
the extent to which the two forms are actually measuring the same
characteristic in a consistent fashion.
Obviously, this procedure does not take account of day-to-day
fluctuations in the person or in the situation of measurement, since
both forms of the test are administered during a single session. Some
transient differences, however, undoubtedly enter. Ther~ -may be
I, shifts in attention during the testing period; increasing bol~om or
fatigue may influence the respopses to the second test; responses. to
items in the first test may affect responses to items in the seconq test.
Nevertheless, unless the measuring devices are extremely long, these
changes are likely to be less than the random changes that would
occur over a longer period of time; thus c;oefficients computed on this'
basis do not fully take into account the effe~t of variable errors of this
sort. On the other hand, since there is little likelih?od of genuine
change~ in the characteristic during one testing session, this method of
computing reliability avoids the problem of confusing real change with
random error. .
The split-half method may be thought of as a special case of the
method o~mate forms ad;;i~istered at the same time. In this
procedure, a single form of a test is administered once to a group of
lndividuals; the items on the test are then divided into two halves, and
the scores on the two halves are correlated to provide an estimate of the
extent to which they are equivalent. In other words, the two halves are
treated as alternate forms of the same test. The resulting coefficient
is an indication of the internal consistency of the test; again, a high
coefficient of equivalence is taken as indicating that the individual's
position is not affected by the particular sampling of items in either
half of the test but would be substantially the same on any test made
up of items from the same universe. As in the alternate-forms approach,
a coefficient of less than 1.00 JUay_reflect randopl..errors- in responses
to individual items as ;ell as nonequivale~~items.
Traditionally, it has been held th-;rtmtlie"split:half method the
test Qr measurement should be split into equivalent halves, each of
which repI:esents the total test in all significant respects. 12 The usual
method of obtaining presumably equivalent ha1ves is to assign the
even-numbered items to one half, the odd-numbered to the other. The
correlation between the scores on the two parts is then regarded as
an estimate of the equivalence coefficient of a test half as long as the
at
originat test. From thIS, anestim;te the coefficient of equivalence
for the entire test-known as the corrected split.-halt..reliability-can
be com?u~ed by n:eans of the S.p.:.:rmanJ~~(This pro-
cedure IS dIscussed 10 greater detaIl on-p-a:ger83.)
More recent thinking holds that, if all items in the test are in-
tended to measure the same characteristic, random rather than sup-
p~sedly equivalent halves should be compared, and a new method of
computing a coefficient of equivalence has come into use_13 The index
result~ from this ~ethod, called coefficient alpha, has, among its
other properties, tha~ of being the average split-half correlation for all
possible ways of dividing the test into two parts. This satisfies the
requirement of rand9mness with respect to the items composing the
halves of the test. Coefficient alpha gives an exact coefficient of
I
equivalence for the full test. The rationale of the method cannot be
developed here, but the interested reader may consult the references
cited.
12 See Thorndike (i 949) and Guilford (1954).
13 See Cronbach (1951) and Tryon (1957b).
Whatever the procedure used for estimating split-half equivalence,

chis approach, like that of alternate forms administered at the same
time, does not take into account the day-to-day fluctuations in the
person and in the conditions of administration.
In summary: The coefficient of equivalence indicates the extent
to w~ic:_h meas~remel]t-.of_~h_e same i~di~Is at th~ same time agree.
The measurements may be made by different observers using the same
~--
same characteristic. To the extent that the coefficient --
instru~~nt, OrQi"different inst~ments intended''''''to measure the -
- IS less than
the' 'maX1mu~alue possible, it is assumed, in the case of different
observers using the same instrument, that there are variations in the
use of the instrument by different observers. In the case of different
instruments made up of a number of items, it is assumed that the items
are not equivalent measures of the same characteristic or that there are
random errors in responses to individual items, or both. In either case,
the coefficient of equivalence does not take into account instability
over time as a source of unreliability.
STABILITY AND EQUIVALENCE. As we have pointed out, indices of
the stability of scores take account primarily of fluctuations in personal
and situational factors as sources of unreliability, while indices of
equivalence take account primarily of variations in the administra-
tion, content, and circumstances of the measurement procedures. The
method of alt.ernate measurement procedures administered. at dif-
fer~nt tin:es is~~ttempt to ~ak~ .aCcOiiiltof the ~~~ned effe~\of
these van.ous sources of unrelIabilIty. A group of mdlVlduals receIve
one form of a test .Iat One time (or ard rated by one observer in a par-
.
e.
ticular situation); after a lapse of time they receive a different forqi
the test (or are rated by a different observer in another situation).
Correlation of scores or ratings on ,the two occasions provides an
ovef:alIluclex of the reliability of the measurement procedure.
In the case of questionnaires, interviews, etc., this procedure has
the advantage over the te~~ethod of being less affected by
memory and practice. Nevertheless, the fact that the items in the two
forms are different does not mean that the results of the second testing
are completely independent of the first. Having taken the first, one's
'lttitude and approach to the second may be different; one has had
the impact of practice with or of stimulation by items that resemble
the new items; and even effects of remembering or misremembering
may enter as a consequence of the similarity of items. As a precaution
against distortion of the results by the specific effects of taking a
given form of the test first, the administration of the two forms is
usually counterbalanced; that is, half of the group is given Form A
first and Form B on the second administration, while the other hal
takes Form B first. However, this does not rule _out the more general
effects of having taken o~e 'for~, whichever it is. The problem of
setting a suitable time interval between the two testings is not es-
sentially different, therefore, from that encountered in estimating
stability by the test-retest method.
As in the repeated-observation or test-retest method, there is the
possibility that genuine changes in the characteristic being measured
have occurred in the interval between the two test administrations. But
again, provided that the results of the two testings are reasonably in-
dependent, the effect of this possibility is to make the obtained co-
efficient an estimate of the minimal reliability of the measuring
instrument.
Since the method of alternate measurement procedures adminis-
tered at different times takes into account more sources of variation
than the other methods we have described, it will ordinarily give a
lower-but a mo.re accurate-estimate of reliability tban either a co-
efficient of stability or a coefficient of equivalence.
Which method of estimating reliability an investigator will use
in any given research depends not only on the value of different tech-
niques for his purposes but also on the practical possibilities open to
him and the resources ~hat can be devoted to the development of the
measurement proceduf,es. Sometimes it is not possible for him to
reach the same group bf subjects twice, or the cost of doing so may
seem prohibitive; in s~ch a case he has no choice but to base his
estimate of reliability ;on the equivalence of Scores. Sometimes the
measuring instrument does not lend itself to the internal analysis that
may be needed for a measure
I,
of equivalence. Whatever the method
an investigator uses, he, should be aware of its implications and its
limitations.
178 GENERAL PROBLEMS OF MEASU'REMENT
WHAT Is A SATISFACTORY RELIABILITY?
There is no simple answer to the question of satisfactory reliability,

It depends both on one's purpose and on the method by which
reliability has been estimated. In general, any lack of reliability in a
test lessens its validity; correspondingly, it lessens one's ability to
demonstrate relationships between variables or to make precise distinc-
tions among individuals who are similar in the characteristic one h
trying to measure.
RELIABILITY AND VALIDITY. To the extent that scores on a measur-
ing instrument are influenced by random errors, they are not accurate
indicators of an individual's position in terms of the characteristic one
is trying to measure. In other words, to the extent that a measure is un-
reliable, it lacks validity. An exception must be noted, however. When
the estimate of reliability consists of a split-half equivalence coefficient,
low reliability does not necessarily detract from validity; paradoxically,
it may even increase validity. In order for split-half equivalence to be
high, all items of the test must be highly cQrrelated; that is, they must
all provide a measure of essentially the same characteristic or of char-
acteristics that vary together. To use the technical term, they must be
~mQg~~.~~;..'13ut for some purposes, a test that taps a number of
different characteristics may be more valid than one that measures a
single characteristic.
Suppose, for example, that we are concerned with selecting can-
didates for training as nurses. Assume that we have eliminated by
other methods those who are c1ea,rly not equipped for-the profession,
on such grounds as low intelligence, poor physical stamina, gross per-
sonality disturbances, etc. We want a test that will help us select,
from those who have met the basic requirements, the ones who are
most likely to be successful as nurses. It seems obvious that a number
of characteristics are likely to be relevant; these may include, for ex-
ample, ability to withstand strain, abilit~ to accept and follow direc-
tions, perhaps such characteristics as insight into self and others, sym-
pathy, or optimism. We might, of course, develop a test for each
characteristic th<j.t we believe, Or know, to be relevant. But for prac-
tical use in selecting candidates, such a battery of tests might not be
feasible; we might prefer to combine into a single test of "Probable
success in nursing" items that measure a number of the relevant char
acteristics. In such a test, the correlations among different items would
probably be quite low; thus the estimate of equivalence computed by
a split-half technique would be low. Nevertheless, the test might be
very effective in identifying candidates who will do well as nurses. How
can this come about?
From the point of view of predicting a particular behavioral out-
come, the ideal test consisting of a number of items should have the
following properties: Every item should have a reasonably high cor-
relation with the criterion and a zero correlation with every other
item. If we were able to construct such a test and measure its reliability
by the split-half or related methods, it would turn ou~ to have ex-
tremely low reliability. Thus, it would seem that the specifications for
maximal pragmatic validity require low reliability, a seeming contradic-
tion ,of the proposition stated earlier, that reliability is a necessary
condition of validity. It should be noted, first of all, that our ideal test
would still have to have reasonably high test-retest reliability to be
useful-i.e., to have pragmatic validity. In the second place-and there
is an important less~n here about what good pragmatically valid tests
measure-the low split-half reliability is not what it seems. Our ideal
test is not really one1esf-arail, but a battery of one-item tests. The
pragmatically valid test, thtls, does not measurea'/characteristic, but a
conipoSite of m~-ny' characteristics.The determination of a split-half
reliability coefficient for a test so set up is meaningless. If one-item
tests are themselves not likely to be very reliable, this is compensated
for by touching on the many facets of the complex. An error of under-
estimation of one facet would tend to be counterbalanced by an error
of overestimation of another. In principle, if each item were expanded
into a more reliable, multi-item test and the scores from each com-
ponent test assigned10ptimal rather than equal weights, the pragmatic
validity of the total battery would increase.
BETWEEN VARIABLES. Random errors in
,
DISCOVERING RELATIONSHIPS
measurement of a variable-that is, unreliability of the measuring in-
strument-obviously reduce the possibility of discovering how that vari-
able IS related to anbther.14 Suppose that we have constructed a ques-
14 Again, we mustj ~ote that this does not app)y tn measures where low split-
half equivalence: does not reduce pragmatir v~lidltv.
tionnaire to measure worker morale in the hope that it will help us makt
predictions about the rate of absenteeism under specified conditions~ If
the questionnaire were completely unreliable-for example, if workers
whom it classified as having low morale were just as likely to show up
as having high morale on a second administration ten minutes later-
it would be impossible to observe a relationship between morale and
absenteeism, even if the two were in fact cIosdy related. If the ques-
tionnaire is not completely unreliable, we may be able to demonstrate
that some relationship exists between morale and absenteeism. How-
ever, if we hope to discover how close the relationship between the two
variables is, it is necessary t'o have highly reliable measuring instru-
ments.
DISTINGUISHING AMONG INDIVIDUALS AND AMONG GROUPS. All the
methods of estimating reliability that we have described, and most of
the others in common use, consist basically in determining whether
measurements at different times or by different forms of the ins~rument
place individuals in the same position in relation to the total group
tested. No matter what the subject matter 6 the test or the method of
estimating reliability, the question being asked is essentially: Do the
results of the two testing situations agree in where they place Charlie
(and Joe and Mary and each of the others) in relation to the average
score of the group? Charlie and Joe and Mary and each of the others
may score ten points higher in one testing situation than in the other,
but this will not show up as unreliability if each is in the same p'0sition
relative to the others on both measures. Nor will different changes in
scores for different individuals aff~ct the estimate of reliability unless
they change the position of the individuals in relation to One another.
Suppose that on the first measure Charlie scored 30, Joe, 40, and Mary
50; and that on the second measure Charlie scored 33, Joe" 4CJ\and
Mary 47. Since the relative position of the three would not be changed,
these shifts would not appear as unreliability. But suppose that on
the first measure Charlie has scored 39, Joe 40, and Mary 41. If Charlie
again gained three points on the second measurement and Mary again
" lost three points, their relative position would change; thus the
changes in scores between the two testing situations would appear as
unreliability.
From the nature of these operations used in estimating reliability
follow several conseq uences: ~ (~ : I.J'
1. The reliability at a measurement procedure is always contingent
on the degree at _un_5;!rmitX at the given characteristic within the
population beIng measured. Small shifts in individual scores may lead
to changes in relative position in a group where the scores of many
individuals are close to one another, whereas the same shifts may not
lead to changes in relative position in a group where individuals differ
markedly from one another. Thus, a test with a low reliability in a very
homogeneous population may have a high reliability in a very hetero-
geneous population. Tests are sometimes published with deceptively
high estimates of reliability, computed on the basis of administration
to very heterogeneous populations, whereas the application of the test
may require the ability to distinguish among individuals in relatively
homogeneous groups.
2. High reliability- is more important it we wish to make fine
discriminations amQng)ndivig!Jals than it we merely wi~h to identity
pdopre;ho-;re at the _extrem~s. To demonstrate a sig;ificant difference
between two scores, the difference between them would have to be
approximately three times as great if the reliability coefficient were .10
than if it were .90; twice as great if it were .60 rather than .90; and
/
about 1.4 times as great if it were .80 rather than .90. Reliability is
obviously important for precise discrimination, and without it the fine
gradations of a measuring instrument are illusory.
3. Estimates at reliability apply to the average reliability ot Scores
of individuals in a group. They provide no estimate of the different
reliabilities of the sc;ores of each individual w!thin the group. It is, of
course, an approximation of unknown degree to assign the same reli- .
ability coefficient to ~cores of all individuals. Frequently, the reliability
of a score at one poiqt on a continuum is different from that at another
point; for example, ~ndividuals who have more intense attitudes may
be more consistent ~han individuals who are less intense (see Cron-
bach, 1949). The reliability of an average score is higher than the
reliability of the individual
I,
scores that go into the computation of that
average. If we are int~rested in group results, therefore, we can afford
to operate with measuring instruments of relatively low reliability,
compensating for the low reliability by increasing the size of the

sample. However, if we are interested in making statements or predic-
tions about particular individuals On the basis of their scores, reliabilities
below .90 are risky.
METHODS OF INCREASING RELIABILITY
The reliability of measurement procedures can often be increased

by taking appropriate steps with respect to the sources of error. Thus,
the conditions under which the procedure is applied can sometimes be
highly standardized, with attention to illumination, noise level, tem-
perature, presence of observers, etc., if these are considered relevant.
Undesired variations in the administration of the procedure can be
minimized by using only sufficiently trained, instructed, and motivated
personnel. One may specify that subjects are to be interviewed only
while they feel in the pest. of health, are suffidently relaxed, and after
appropriate steps have been taken to assure rapport. The greater the
control desired over these sources of unreliability, the greater the re-
. sources one needs to have at one's command.
There are two very powerful methods of increasing the reliability
of a measurement procedure that involve the selection and accumula-
tion of measurement operations rather than any change in the con-
ditions under which the measurement operations are made. These
methods can be illustrated most clearly when the "measurement opera-
tion" is the administration to a subject of a particular test item which
he either passes or fails; but the same principles apply when the meas-
urement operation is the administration of another kind of test item,
or the rating of some aspect of the subject's behavior by a particular
observer.
The first method of increasing reliability is to add measurement
operations of the same type as the ones with which we startea, and
assign the subject a score based on the sum of the results of all the
measurement operations. In the testing situation, this means increasing
the length of the test. In the observational situation, it means Increas
ing ~~rs, or the number of o~casi?~n whic~ ea_c;h
subject is observed, or both. If the' correlation between the results of
anyone measurement operation and any other measurement operation
is approximately the same within the series of operations we are using,
there is a very simple formula, known as the Spearman-Brown. fOIrnula,
which enables us to predict approxima.teIy t~ffect of increasing the
n umber of measurement operations: 15
- _- -----Nr
rNN = 1+ (N-l)r
A close examination of this formula shows that we can make
the reliability of a measurement procedure approach as closely to 1.00
as we wish, provided that we are able to add measurement operations
indefinitely without changing their nature in any important way. If the
correlation between the proportion of time spent lecturing by a group
leader during one fifteen-minute period and the amount of time spent
l.ecturing during another fifteen-minute period is .10, a measure of lec-
turing proclivities based on observation during only one such period
win be of little use. But if we can base the measure on the average re-
sults of observation during five fifteen-minute periods, its reliability
can be expected to increase to around .36. If we can average the results.
for ten fifteen-minute periods (of the same sort as we have been using),
the reliability of our measure will probably increase to around .53, while
if we can use as many as 100 such periods of observation, we can expect
to attain a reliability of more than .90. Exactly Jhe same principle
applies when we add individual items to a psychological test.
The Spearman-Brown formula assumes that the oorrelation be-
tween any O'ne--rneasurement operation, or "item," and any .other item
is ~pproximateli the' same for all pairs of items being considered; to
the extent that we aiM items or meas-;;-rement operations that do not
correlate with the o~hers, our actual results will fall short of those
predicted by the formula.
An alternative method of increasing reliability starts by assuming,
15 In this formula r is the correlation between anyone measurement operation
and any other, N is the number of measurement operations, and rNN is the correla-
tion between the sum ot average of N measurement operations of this particular
sort and the sum or average of another N operations of the same sort. Essentially,
r is the measure of reliability'for a "test" consisting of one measurement operation,
while r NN is the predicte9 reliability for a "test" consisting of N measurement opera-
tions of the same sort. ,The Spearman-Brown formula is discussed in most text-
books on mental measurement. See, for example, Gulliksen (1950) or Guilford.
(1954).
that there are sizable differences in the correlation of items with one
another. The problem then is to select from the possible available items
or measurement operations those that correlate most highly with one
another, and to increase the reliability of the measurement procedure
as a whole by increasing its internal consistency.
This method has rarely been used outside the field of psycho-
logical testing (including attitude measurement), but in this field it ~as
been quite successful. The most commOn practice is to begin with ~_
fairly large collection ofltemr,calculate a scorebased on each item,'"'"
anl another score based on-;~sponses to the' total set of ite~s:-The~
"~ ..._.~....-.-
------ -
- < ____ ....,..,..".
tile score for eacli Ifem is "correlated with the total score, and those
.
- ----~ - _. -- ........ _.__.-....
ifems are select~~ ll).at correlate most lilghly with this score. These
ifems are CTivid"ed into tWo equivalent g~oups; two ~ewscores are
calculated based on the two groups of selected items; and these
scores are correlated to provide a measure of the 'reliability of -the
"purified" test. If the new reliability is not satisfactory, the test may be
further purified in the same manner as before, or'additional items may
be added ofthe type represented by the selected items. "
Rather thaf; Correlating the score for each item with that for the
total test, the goal of increasing internal consistency may be approached
by the followirig procedure: The subjects are divided into ~o groups
-a high-scoring_and a low-scoriiig- one-on the-b-asls_or their total
scores:"Inhe,n~mber of s~ts IS quite large; as-it properly should be,
one takes extreme groups-say, the top and the bottom twenty per
~tanirem IS consistent with the complete set of items, th~~-th{'
proportion of high scorers who answer the item in a ,specified Wqy
should be significantly different from the corresponding proportion ,of
low scorers. Those items are most consistent with the total setwhich'
yield the largest differences in the appropriate direction.
As an example, let us consid_er the procedure used in constructing
a scale of anti-semitism for use in the Authoritarian Personality inv,esti-
gation (Adorno et at, 1950). A questionnaire consisting of 52 items
referring to Jews was administered to a group of female college stu-
dents. Let us consider the results for five of the items on the test:,
A. One trouble with Jewish businessmen is that they ~tick
together and connive, so that a Gentile doesn't have a fair
chance in competition.
B. Colleges should adopt a quota system by which they limit
the number of Jews in fields which have too many Jews now.
C. Anyone who employs many people should be careful not
to hire a large percentage of Jews.
D. The trouble with letting Jews into a nice neighborhood
is that they gradually give it a typical Jewish atmosphere.
E. Most hotels should deny admittance to Jews, as a general
rule.
The respondents were asked not simply to agree or disagree with
each item, but to indicate the strength of their opinion, from "strong
support, agreement" to "strong opposition, disagreement." The reply
to each item was scored on a scale ranging from 1 (strong opposition
to anti-semitism) to 7 (strong anti-semitism), with a neutral point of
4. For each item; the mean scores of the 25 per cent who scored highest
and the 25 per cent who scored lowest on the total test were computed;
the difference between the two means was taken as the "discriminatory
power" of the item. The figures for our five items are given in the
following table:
Item Mean Score Discriminatory Mean for

Power Total Group
Upper Lower
/
25% 25%
A 5.86 1.38 4.48 3.45
B 2.89 1.00 1.89 1.67
C 5.30 1.l9 4.11 2.84
D 5.28 1.32 3.96 3.23
E 2.22 1.05 1.17 1.46
It is apparent that items A, C, and D distinguished sharply between
the high and low sc6rers. On these three items, those who scored high
in anti-semitism oni the total test took a mean position of slight to
moderate agreement, while those who scored low on the total test took
a mean position bet\veen moderate and strong disagreement. Items B
and E, on the other hand, showed much less difference between the
high and low scoreis; the entire group tended to disagree with thes(
items. Items A, C, and D were kept; items Band E were dropped.
It is important; to recognize that the process of eliminating item'!)
from a test or attitude scale by the criterion of internal consistency

always results in a change in the working definition of what is being
measured (see Chapter 2). This change is always in the direction of
narrowing and restricting the working definition. How far to carry this
process depends on the purpose of the research and the specificity of
the characteristic the investigator is trying to measure. The extreme
limit of the purification process would be represented by a set of items
that correlated perfectly with one another and differed only in dif-
ficulty or "acceptability." Such a set of items corresponds to the con-
ception of a completely homogeneous test or unidimensional scale,
which will be discussed further in Chapter 10.
Another important thing to remember about the method of in-
ternal consistency is that it is concerned solely with the equivalence of
---- -----
two sets of items or measurement operations. If canreauce unfeliatility
-..--.- ---
resulting from lacKor equival<::nce of items, but it does nothing to
reduce unreliability resulting from instability of a subject's responses
or variations in the conditions of measurement. The method of in-
creasing the number of measurement operatjons can be used to reduce
these sources of unreliability if it is possible to spread the measurement
operations out in time or to distribute them over a number of dif-
ferent conditions of measurement. If either of these procedures is
feasible, the investigator should ask himself: "Is what I really want
to measure the average level of this characteristic in a variety of situa-
tions, even if there may not be much consistency in the characteristic
from one situation to another; or do I want to measure something more
specific?" The answer to this que~tion will determine- the approach
he adopts for increasing the reliability of his measurement procedure.
Scales at Measurement
The effec~iveness of our behavior both in science and in everyday
life depends on our ability to distinguish among objects and to make
differential responses to them. Many of our activities require no more
.than the distinguishing of objects possessing qualities that are rather
sharply demarcated from those of others. To take obvious ~xamples
from daily life, it is both useful and simple to notice the differances
SCALES OF MEASUREMENT 187
,
between a pear and an apple, between an infant and an adult, between
a tennis court and a swimming pool, between the ringing of a telephone
and a Beethoven sonata.
Similarly, in the social sciences many of the distinctions that are
made are qualitative in nature. For example, we distinguish different
languages, different types of social system, different nationalities, and
so forth.
However, both in the sciences and in everyday life, it is often de-
sirable to make distinctions of degree rather than of quality. In daily
life we are frequently faced with the problem of selecting among
alternatives: Which person is more intelligent? Which type of cloth
is sturdier? Which teacher is more interesting? In the interest of both
accuracy of judgment and the discovery of constant relationships
among characteristics that vary in amount as well as in kind, science
pursues the objective of replacing statements that simply affirm or
deny differences by more precise statements indicating the degree ot
difference.
Although there is little doubt that quantification facilitates the
establishment of scientific laws, it should be recognized that measure-
ment exists in a variety of forms. Sometimes measurement has been
defined so as to exclude methods of data collection/
that permit only
qualitative discriminations. For example, McGregor (1935) defined
measurement as "the process of assigning numbers to represent quanti-
ties." Other writers, such as Weyl (1949), Stevens (1946,1951), and
Coombs (1953), have included in their concept of measurement any
empirical procedure that involves the assignment of symbols, of which
numerals are only one type, to objects or even~s according to rules.
Measurement tS possible only because there is a certain corre-
spondence betweenjthe empirical relations among objects and events,
on the one hand, a~d the rules of mathematics; on the other. We use
empirical procedures to determine the relations among objects and
events. In the case of physical objects, these empirical procedures may
take the form of direct manipulation. Suppose that one has a number
of bars of iron and ~ticks of wood, which he wishes to distinguish from
one another. On the basis of criteria we need not go into here, he
identifies some of .them as being wood (that is, as being equivalent
to one another in composition) and others as being iron (that is, as

being equivalent to one another, but not equivalent to the sticks).
Now he wants to know mOre about the relationships among the
sticks in terms of their length. He wants to arrange them in order,
with the longest at one end, the shortest at the other. He wants to find
out whether the differences between pairs are equ~or 'example,
whether stick A is as much longer than stick B as B is than C. He warts
to determine ratios among them; for example, whether A is twice as
long as B. All these relationships can be determined by empirical
operations of placing the sticks next to each other in certain ways and
making certain observations.
In practice, of course, the measurer would be likely to use a yard-
stick rather than to carry out these operations directly with the pieces
of.wood. He is able to do this, however, because the properties that tan
be ascertained by the empirical operations correspond to the properties
of the numerical system in terms of which the yardstick is set up. For
example, if 'his empirical operations have shown that stick A is twice
as long as B, and if the yardstick shows B to ~e four inches long, it will
show A to be eight inches long. In other words, numbers yield to
mathematical operations which are analogous to the empirical ones
by which we can determine relationships among objects. This cor-
respondence between properties of the numerical system, when the'
various rules are laid down, and certain empirical relations among
objects, permits the use of the numerical series as a model to represent
characteristics of the empirical world.
The same correspondence be~een empirical relations and proper-.
ties of the numerical system is basic to measurement in the soc~al
sciences. Suppose that one is interested in studying attitudes toward (he ,
United Nations. If 'one wants simply to assert that two people differ
in their attituoes toward the U.N.,' without specifying how great the
difference is or whether one is more favonible than the other, he must
at least be able to distinguish different types of attitude; that is, he
must be able to identify certain attitudes as equivalent, others as not
equivalent. If one wishes to state that the attitude of otie person is
~ore favorable thap that of another (without, however, specifying how
~uch mOre favorable), he must be able to rank different attitudinal
positions as being more favorable or less favorable than other positions.
If one wishes to make the statement that A is as much more favorable
than B as B is than C, or that two experiences have produced equal
changes in attitudes, he must be able to determine whether the dif-
ference between two attitudinal positions is equal to the difference
between two other attitudinal positions. And if one wishes to make
some such assertion as "A is twice as favorable as B," he must be able
to determine the existence of an absolute zero of favorableness for the
given attitude, as well as equal units above the zero point.
These four types of statement correspond to four types of measure-
ment scales described by Stevens (1946, 1951) :16 nominal, ordinal,
interval, and ratio scales. 17 Let us indicate, briefly, the formal rules and
empirical operations that distinguish these various scales. The scales
are listed in ascending order of power. The "stronger" scales presuppose
the ability to perform the empirical and mathematical operations of
the "weaker" ones; thus, the ratio scale implies all the operations of
the nominal, ordinal, and interval scales as well as those that are unique
to itself, and it contains all of the types of information contained in
the preceding scales as well as certain information unique to itself.
NOMINAL SCALES
One of the dictionary definitions of nominafis "of, pertaining to,

or consisting in a name or names." A nominal scale is one that consists
of two or more named categories, into which objects or individuals or
responses are classified. The basic iequirement for a nominal scale is
that one be able to distinguish two or more categories relevant to the
attribute being considered and specify criteria for placing individuals,
etc., in one or another cat~gory. The only specified relationship be-
tween the categoriesI is that they are different from each other; there
is no implication that they represent "more" or "less" of the char-
acteristic being measured. The empirical operation is the decision that
I
-
a given object, indiridual, or response belongs in a given category or
that it does not; in other words, it is the determination of equivalence
16
- ..
Our discussion lof measurement scales is directly indebted to Stevens.
17 Coombs (1950, 1953) has described, in addition, the "partially ordered
scale," which falls logically between the nominal and the ordinal scale, and the
"ordered metric," which falls logically between the ordinal and the interval scale.
or noneguivalence, with respect to the attribute in question, between

the given object and other objects placed in a given category.1S Classi-
fication of ill_divid~a~ a~ing to nationality, for example, consilli:ires
a nominal scale.
For convenience, numbers may be used to identify the categories
in such a scale, but they are used only in the way that letters or words
would be; there is no empirical relation among the numbered categories
that correspondsto the mathematical relation betwee~ the numbers
assig~d. Therefore statistical techniques that make _uj_oLmathe-
matical relations among numbers (such as the computation of means
or correlation coefficients;fOr example) are inappropriate;. 9ne may us~
only such statistics as are appropriate to'counting; for example, the
number of cases, the mode, the coefficient of contingency, and the
chi-square. 19 Thus, in a study of the relationship between personality
and nationality, one might establish some criterion or criteria on the
basis of which he would categorize people according to personality
types, and would then determine whether a given personality type
tends to be associated with a given nationality more frequently than
the other types by counting the number of cases of each personality
type which fall into each nationality grouping.
Use ~of nominal scales is c~aracteristic of eXEloratory research,
where the emphasis is on unc_overing a relationship between two char-
acteristics rather than on specifying,with SOme degree of precision,
the mathematical form of the relationship; or where the focus of the
study is on the paftern of relationship among several characteristics of
the person. However, if the data permittthe ordering of people in tenns
of a characteristic, it is uneconomical to use a nominal scale, since such
a scale severely limits the nature of the conclusions that can be drawn.
18 The relation of equivalence (or equality) has the logical characteristics of
being transitive and symmetrical. In a transitive relation, if A stands in a certain
position with respect to B, and B stands in a similar position with respect to C,
then A also stands in that position with respect to C. For example, if A equals B
and B equals C, then A equals C; if A and B are placed in the same category on a
nominal scale, and Band C are placed in the same category, then A and Care
placed in the same category. In a symmetrical relation, if A stands in a certain
position with respect to B, B stands in the same position with respect to A. For
example, if A equals B, then B equals A; if A is placed in the same category as B,
B is placed in the same category as A.
19 The procedures used in computing these statistics may be found ,in any
standard statistics text.
ORDINAL SCALES
An or~inal scale defines the relative position of objects or individ

uals witli respect to a characteristic, with no implication as to tht,
distance between positions. The b~sic requirement for an ordinal scal(;
is that one be able to. ~etermine.z for. ~.ac~,individual or opj~cJ being
measured,
. - whether thaJjndiyidual has ,more of the attribute in gues-
I""-~.-- ..
tism than, another individual, or the same amount, or less; in other
words, one must be able to deterI1!ine the order of 20sitions.20 This
presupposes that one must be able to place each individual at a single
point with respect to the attribute in question. If the attribute or the
measuring procedure is such that o!}e can say, "In some respects John
has more of this attribute than~, but in other respects he has less,"
the two men cannot be placed in relation'to each other on a single
ordinal scale. Difficulties of this sort are not uncommon in the social
sciences. Take, for example, the attempt to order people in terms of
degree of favorable or unfavorable attitude toward some minority
group. John may have more negative stereotypes of the group than Bill
does, but less hostile feelings; they may engage equally in discrimin-
atory behavior. In sllch a cas~, in order to meet the requirements for
ordinal scaling, it is necessary either to rank"1he men on eacKof the
three dimensions separately, or to h;ve some rational~ for combining
positions on the three dimensions intq a single score representing de-
gree of favorable or unfavorable attitude toward, the group.21 The
ability to place a giv.en individual at a single position is a prerequisite
forthe more powerful scales as well.
An ordinal scale is like an elas~ic tape ~easure that is being
stretched unevenly; the scale positions as indicated by the numbers
on the tape are in a clearly defined order, but the numbers do not
provide a definite indic~tion of the distance between any two points.
20 The relation of order has the logical characteristics of being transitive and
asymmetrica1. Transitivity has been discussed in the preceding footnote; in this
case, if A is greater than Band B is greater than C, then A is greater than C. In an
asymmetrical relation, if A stands in a certain position with respect to B, then B
does not stand in that position with respect to A; if A is greater than B, then B is
not greater than A. The relation of ancestor to descendant, for example, is both
transitive and asymmetrical.,
21 This issue is discussed in more detail in the section on attitude scales in
Chapter 10.
The distance between 8 and 9 may be equal to, less than, or greater
than the distance between'l and 2: With ordinal ;cales we are limited
to statements of gre.:;_ter, _S9ual, or less; we cannot undertake to state
how much greater or how much less.
An exaIiIple of' an ordinal scale in the physical sciences is the
Mohs' scale of hardness, which is applied to minerals. The empirical
relation in this case is the ability of minerals to scratch one another.
A diamond is ranked highest on this scale, since it can scratch all other
known minerals but none can scratch it; however, the scale does not
assert anything about how much harder a diamond is than other
minerals.
When the operation by which objects or individuals are placed
on a scale involves direct comparison of the individuals in terms of the
extent to which ~hey P,Ossess the attribute in question, it is ea~l... to see
[that the scal~ reflJ~ts.only the order of positio~s and n~t the dis~~
, between them. T.lus IS the case, for example, m a spellmg bee, or m ,a .
teacher's ranking of children in terms of cooperativeness, or in the
judgment of the relative desirability of applicants for a job. Although
the individual who is ranked highesris given tn.e number }, the next
highest 2, etc., it is clear that there is no necessary assumption that
# 1 is as much higher than #2 as # 2is than # 3, etc. - -
.., 'However, when the data needed for placing, a~_ individual are
gathered by an instrument that yields a numerical score, the fact that
one may still be dealing only with an ordinal sc:ale is sometimes ob-
scured. Suppose that three indivi,duals taking a spelling test: or an at-
titude questionnaire, receive scores of 100, 80, and 60, respectively. pur
knowledge of mathematical relations may dispose us to think tha,t the
person who scores 100 is as much higher than the one who scores 80
as the latter is above the one who' scores 60, But unless we have reason
to believe that the distance between 80 and 100 represents the same
amount of the attribute being measured, as does the distance between
60 and 80, these scores indicate only that the first person ranks higher
than the second, and the second higher than the third.
, 0~e statistics ,a,ppIicable to dat~ that permit, only ra~k ordering
are lImIted. In addltJon to those applIcable to nommal scales, one may,
strictly speaking, use only such statistics as medians, percentiles, and
rank-order correlations. Within recent years there has been a rapid
expansion of statistical tests appropriate to data that are simply ranked
or ordered. 22
INTERVAL SCALES
On an interval scale, not only are the positions arranged in terms

of greater, equal, or less; the units, or intervals, of measurement are
equal. In other words, the distance between the positions labeled I
and 2 on the scale is equal to the distance between positions 2 and 3,
etc. The basic empirical operation for the establishment of such a
scale is a procedure for determining that intervals are equal. The
Fahrenheit and Centigrade thermometers are examples of interval
scales; the units represent equal amounts of change in the volume of a
column of mercury under a certain pressure.
For many of the attributes with which the social sciences deal,
procedures have not yet been devised that will give reasonable certainty
about the equality of intervals. Attempts, bowever, have been made,
particularly in connection with the measurement of attitudes. The
most frequent approach to determining the equality of intervals has
made use of the judgments of large numbers of p~ople about the loca-
tion of various positions. Intervals between adjacent positions have
been considered equal either on the basis of a consensus of judgments
that they appeared equal,23 or on the bAs
of statistical calculations
based on the distribution of judgments Ib~ut the location of various
positions.
If one has reason to believe that the units of his scale are equal,
then he is justified in making use of mathematical relationships among
numbers which correspond to this fact. For example, he may assert that
an individual whos~ score changes from 3 to 5 has shown as much
change as one whose I score shifts from 5 to 7. He
-_ may . -appropriately
compute means, standard deviations, and produc!-moment correla-
22 For a presentation of a variety of statistical techniques applicable to data
gathered by means of otdinal scales, consult Siegel (1956).
23 For a more deta,iled account of this approach, see the discussion of Thurstone
scales in Chapter 1;).
tions; in f,act, almost all the usual statistical metho~ are applicable to
an interval scale.
As in the scales previously discussed, the zero point on an interval
scale is a matter of convention. Its arbitrariness is indicated by the fact
that a constant can be added to all scale positions without changing
the form of the scale. The arbitrariness of the zero point is apparent
when one compares the Fahrenheit and Centigrade scales of tempera-
ture. In the latter, zero corresponds to the point at which water freezes;
in the former, zero is well below that freezing point. Because the zero
point is arbitrary, multiplication and division are meaningless; although
relations between positions can be stated in terms of the distance
(".i.e., number of scale points) between them, they cannot be stated
in terms of ratios. Thus, with data that meet the assumptions of an
interval scale (but not of a ratio scale-see below), one cannot state
that a person's attitude is twice as favorable as that of another person,
just as one cannot state that 20 0 F. is twice as hot as 100 F., or 20 0 C.
twice as hot as 100 C. However, difierences between values on an
interval scale can be treated in terms of ratIos. Thus, we can say that
an individual who shifted from a score of '3 on an inter_val scale to a
score of 7 has changed twice as much as one who shifted from a score
of 3 to a score of 5. This is because the point of no difierence provides
an absolute zero.
RATIO SCALES
A ratio scale, in addition to having the characteristics of an/interval

scale, 'contains an absolute zero. The empirical operations necessary to
establish a ratio scale inch;lde methods for determining not only
equivalence-nonequivalence, rank order, and the equality of intervals,
but the equality of ratios. Since ratios are meaningless unless there is
an absolute zero point, it is only with this type of scale that one is
justified in making assertions such as: "A is twice as heavy as B," or "I
spent half the amount that you did."
This type of scale, often called "fundamental measurement," i5
most commonly found in physics. Measures of weight, time intervals,
length, area, angles, etc., all conforn!'to ratio scales. So does the scale
of cardinal numbers itself-the scale we use when we count people,
dollars, eggs.
For most of the subject matter of the social sciences, we are far
from having devised procedures that satisfy the requirements of a
ratio scale. However, attempts have been made to construct such scales
in connection with the judgment of psychophysical attributes such as
loudness and pitch (see Stevens, 1951, 1957). As in the case of at-
tempts to construct interval scales for the measurement of attitudes, the
judgments of people have been used as the basis for determining posi-
tions on the scale. For example, a tone of a given magnitude is sounded,
and subjects are asked to select a tone that is half as loud, another tone
that is twice as loud, etc. The absolute zero point is established by the
threshold between no perception of sound at all and the barest percep-
tion of sound.
If one's data conform to the criteria for a ratio scale, all the rela-
tions between numbers in the conventional system of mathematics
obtain between the correspondingly numbered positions on the scale.
Thus, as in an ordinal scale, an individual, object, or event which is
placed at the position labeled 10 on the scale ranks higher on the at-
tribute being measured than does one which is placed at the position
labeled 5; as in an interval scale, the individual, etc., at position 15 is
as much higher than the one at 10 as that one/is above the one at
position 5; and, in addition, the one at position 10 can be said to possess
twice as much of the attribute as the one at position 5. With such a
scale, all types of ~atistical procedures are applicable.
THE PRESENT STATUS OF MEASUREMENT IN THE SOCIAL SCIENCEs 24
Whether various techniques used in the social sciences yield data

corresponding to o~e or another of these scales of measurement has
been the subject of considerable controversy. Disagreement centers
on whether a given" measurement technique, or most techniques, or
I
any techniques used in the social sciences meet the criteria for an in-
terval scale. There is little argument about nominal scales. A nominal
scale is usually recdgnizable as such, and there is little temptation to
24 For more detail~d discussioq, see Coombs (1953), Green (1954), Gulliksen
~1950), Stevens (1946,1951,1957), and Torgerson (1958).
-
use inappropriate statistics since, in the absence of distinctions of

degree, the mathematical rules governing order and distance are clearly
inapplicable. At the other extreme, there is general agreement that
except for the scale of number itself, which is used in counting fre-
quencies, none of the measurement techniques used in the social
sciences, with the possible exception of the procedures for measuring
certain psychophysical phenomena referred to on the preceding page,
correspond to ratio scales.
There is, however, considerable disagreement over whether an
ordinal or an interval scale provides the most appropriate model for
most measurement techniques currently used in the social sciences.
This disagreement manifests itself both in explicit statements and
implicitly, in the choice of statistical techniques. Some writers have
;taken the view that few, if any, of the techniques now in use provide
\ dkta that can be considered as appropriate to more than ordinal scales.
Others have taken the position that data from certain measurement
procedures-for example, those used in measuring IQ, or in certain
types of attitude scales-may properly be treated as conforming to
interval scales. Still others have taken the position that, altl!ough most
of the, measurements used do not go beyond ordinal scales, probably
little ;harm is done in applying to them statistics that are, strictly
speaking, appropriate to interval scales.
Statistics appropriate to interval scales continue to be widely used
i~ fh~ analysis of social science data, with or without the assumption
that the data actually meet the requirements of such~scales: However,
there is also an increasing use of statistics that are specifically appro-
priate to ordinal scales. '
If the experiences of the natural sciences may be taken as il. guide,
the discovery of ,precise relationships, among characteristics, as ex-
pressed in numerical laws,.is largely dependent upon the existence of
ratio scales. Without knowledge of such relationships, the measure-
ment of characteristics by indirect or derived procedur<;:s is difficult. But
indirect measurement may be more efficient than direct measurement.
For example, it is rather simple to measure the density of liquids on the
basis of the law that expresses density as a constant function of the
ratio between the weight and the volume of the particular liquid, but
it would be tedious and less precise by direct procedures. By a direct
procedure we might, for example, use a standard set of solid bodies,
which we would place in the various liquids. We would agree to call
one liquid more dense than another if we could find a solid body which
would float in one but not in the other. Following this procedure sys-
tematically for all liquids, we would assign the numbers 1, 2, 3, 4, 5,
etc., to designate the position of the liquids in the density scale. Clearly,
the direct procedure is more cumbersome and less precise than the'in-
direct procedure.
Only by the aid of'iindirect procedures can we measure the tem-
perature of the distant stars or the blood pressure in the arteries of living
things. Similarly, it seems probable that much of social science meas-
urement will always be indirect, and thus will depend on the develop-
ment of knowledge of regular relationships among characteristics. The
development of this knowledge is in turn partly dependent upon the
development of fundamental measurement. The history of science
attests to t~e fact that the development of measurement processes is
dependent upon the constant interaction of both empirical procedures
for measurement and theoretical concepts about what is being meas-
ured. Exclusive concern with either the empirical procedures or the
theoretical concepts is likely to be unproductive.
The opinion is sometimes expressed that the social sciences can
never hope to reach the precision of measurement achieved in the
physical sciences, because the very nature of the material with which
the social sciences deal does not permit such refinements as, for ex-
ample, the establishment of an absolute zero point. This may prove
to be a correct prediction, but it seems premature at this early stage
of the development of the social sciences. As'Stevens (1951) has
pointed out, the measurement of many physical qualities has pro
I
gressed from scale to scale:
I
When men knew temperature only by sensation, when things
were only "warmer" ior "colder" than other things, temperature
belonged to the ordinal class of scales. It became an interval
scale with the development of thermometry, and after thermo-
dynamics had used the expansion ratio of gases to extrapolate to
zero, it became a ratio scale.
Summary
In this chapter we have discussed some general problems of meas-

urement that must be considered in connection with any procedure for
gathering data. We have indicated that scores on any measuring in-
strument are determined not only by. the characteristic it is intended
to measure but also by a variety of o'ther factors, some relatively stable,
some transitory. A measurement procedure is considered valid to the
extent to which scores reflect true differences among individuals,
groups, or situations in the characteristic which it seeks to measure, or
true differences in the same individual, group, or situation from one
occasion to another, rather than constant or 'random errors. A measure-
ment procedure is reliable to the extent that independent applications
of it yield consistent results. I , "
It has been pointed out that measurement rests on the correspond-

ence between empirical relations among individuals, objects, etc., and
properties of the numerical system. In this connection, four types of
scale have been described: the nominal scale, consisting simply of dis-
tinguishable categories, with no implication of "more" or "less"; the
ordinal scale, on which positions can be identified in terms of "more"
or "less" but with no implication as to the distance between positions;
the interval scale, on which the distance between any two adjacent
positions is the same as that between any other two adjacent positions;
and the ratio scale, which has not only equal interyals but an absolute'
zero.
In the next few chapters we shall discuss specific approaches 'to
the collection of data: observation, interviews and questionnaires, pro-
jective and disguised techqjques, and the use 'of data already available.
6
DATA COLLECTION
1. Observational Methods
Unstructured Observation
/
Structured Observation
How odd it is that anyone shollfd not see that all observation
must be for or against some view, if it is to be" of any
service. CHARLES DARWIN
(l'X 7:.. ARE ALL CONSTANTLY OBSERVING-noticing what is going on

\_l' l' ~round us)We look out the window in the morning to see
whether the sun is shining or whether it is raining, and make our
decision about carrying an umbrella accordingly. If we are driving,~we
look to see whether the traffic light is red or green. We see a child dash
into the street in front of a car, and watch whether the driver manages
to stop or swerve to avoid hitting him. There.is no need to multiply
examples; as long as we are awake, we are almost constantly engaged
in observation. It is our basic method of getting information about
the world around us. .
Observation is not only one of the most pervasive activities of dai1y
life;'it is a primary tool of scientific inquiry. Observation becomes a
scientific technique to the extent that it (1) serves a formulated re-
search purpose, (2) is p1anned systematically, (3) is r~corded system-
atically and related to more ge~eral propositio'ns rather than being
presented as a set of interesting curiosa, and (4) is subjected to ch'ecks
and controls on validity a~d reliability.) .
1 The f2ct that observation as a research~technique must be purposefully iiJte-
grated with other steps in the research process ,does not, however, preclude, the
possibility that observations of great scientific significance can ,occasionally be made
by chance. The history of science is replete with discoveries based on haphazard
observations that were unrelated to an established research purpose; the ,discovery
of radium and of penicillin are two of many available examples. For a 'discussion'
of the role chance observations have played in scientific discoveries and many in-
teresting examples, see Beveridge (1950). The significant chance observation,
however, is largely a gift of the gods. Moreover, we must distinguish between the
chance observation that points to a hitherto unsuspected phenomenon or suggests
somE important hypothesis and the systematic follow-up observation that makes
something of this gift. .
2(.10
DATA COLLECTION: 1 201
Many types of data required by the social scientist as evidence in
research can be obtained through direct observation. Suppose that he
is interested in how members of different groups behave toward one
another when some activity brings them into contact; or in the manner
in which mothers rear their infants; or in comparing the quality of
housing occupied by different social strata of the population; or in
describing religious ceremonies and rituals. To obtain these and many
other types of data, he proceeds best by observing the appropriate sit-
uations.
~irecl: observation of behavior is, of course, not the only method
by which the scientist can obtain data; interviews and questionnaires,
projective techniques, and available records can be- substituted for the
scientist's own observatio) Compared with these other methods of
data collection, observation has some advantages and some disadvan-
tages. J....J.. u "
~rhaps the greatest asset of obseiVaHoriirtechniques is that they
make It possible to record behavior as it occ~An too many research
techniques depend entirely on people's retrospective or anticipatory
reports of their own behavior. Such reports are, as a rule, made in a
detached Illood, in which the respondent is somewhat remote from the
stresses and strains that influence what he does or/says in the ordinary
course of events, while he may be influenced by other stresses and
strains peculiar to the research situation. The degree to which one can
predict behavior from interview data is at best limited, and the gap
between the two can be quite largeUn contrast, observational tech-
niques yield data that pertain directly to typical behavioral situations-
assuming, of course,: that. they are applied to' such situatio~ Where
the social scientist has reason to believe that such factors as detachment
I
or distortions in recall may significantly affect his data, he will always
prefer observational imethods. Sometimes a study demands that what
people actually do ~nd say be compared with their account of what
they did and said. Obviously, two methods of collecting data must be
employed in 'such inquiries-observation
I <....
and interviewing.
_
Many forms of-behavior are so taken-for graiifed'bi the subjects
2 See Chapters 7, 8, and 9.
202 DATA COLLECTION: I
under investigation, are so much "second nature," that they escape

awareness and resist translation into words. Anthropologists, for in-
stance, in observing foreign cultures, often note facts that their best
local informants would never have thought of reporting. Not only rit-
uals and ceremonies but also everyday occurrences, such as the ~rea!
ment of a small child by his mo% are often of the kind that have to
be seen if their characteristic features are to be discovered.
Moreover, some investigations deal with subjects (infants, for
example, or animals) who are not able to give verbal reports of either
their behavior or their feelings for the simple reason that they cannot
speak. Such investigations necessarily use observation as their method
of data collection. Spitz and Wolf (1946), through the observation Gl
behavior of babies in a nursery, were led to the conclusion that pro-
longed separation of a chiI'd from a previously attentive mother may
lead to. a severe depression, starting with weepiness and culminafing
in ngid withdrawal. Observations of chimpanzees at the Yerkes Labora-
tories h~ve provided data on the social behavior of animals. For ex-
ample, a study (Hebb and Thompson, 1954) in which the experi-
menter played atsome times the role of a "timid man," at other times
th~. ,!gle of _a "b91.4 .l1.!~~~ spowed both marked indi~nces
in the re\lctions of different chimpanzees and striking common features
of their behavior. The apes behaved in ways interpreted as indicating
fear much more often when the "bold man" appeared (for example,
they moved to a far corner of the cage); toward the "timid .man" they
behaved much more often in ways interpreted as teasing (or, as the
investigators describe it, "behavior which . .' . enticed an innocent
clos_;dust to scare the hell out of him") .
~ddition to its independence of a subject's abilitt to report,
observation is also independent of his Willingness to repo~i)There are
occasions when social research meets with resistance from the person
or group being studied. People may not have the time, or they may
not be inclined, to be interviewed; they may resent being singled out
or being asked questions whose purpose is obscure to them; they may
object to being tested, out of fear that they may not come up to the
standards of their group; and so on. Although observation cannot al-
.ways overcome such resistance to research; it is less demanding of
DATA COLLECTION: I 203
active cooperation on the part of the subjects. To be sure, people under
observation may, if they know they are being observed, deliberately try
to create a particular impression; but, even so, it is probably more
difficult for them to alter what they do or say in a life-situation than
to distort their memory or report of what they have done or said.
On the other hand, observation has its specific limitations. We
have listed as an asset the possibility of recording events simultaneously
with their spontaneous occurrence. The other side of the coin is that
it is often impossible to predict the spontaneous occurrence of an event
precisely enough to enable us to be present to observe it. If an anthro-
pologist wishes to learn about marriage ceremonies by observation
rather than through interviews, he has to wait until a wedding to which
he has access takes place. If a social psychologist wishes to observe
behavior in extreme situations (such as during a disaster), his physical
and emotional endurance as well as his patience may be put to a severe
test. Even the observation of regular daily occurrences may become
difficult because of the possibility that unforeseeable factors will inter-
fere with the observational task. An observer attempting to collect data
on the games played by children in a playground is at the mercy of the
weather, alternative attractions in the street which may interrupt the
games, the possibility of fights, etc. Unless there are good reasons for
/
engaging in direct observation, the method of interviewing is often
more economical in such cases.
Furthermore, the practical possibility of applying observational
techniques is limited by the duration of events. Life histories, for ex-
ample, can hardly be obtained in this way. Moreover, some occur-
rences that people may be able and willing to report are rarely, if ever,
accessible to direct ,observation. Sexual beha~ior, a family crisis, or an
undisturbed familYtbreakfast are examples of events that, as a rule, are
not open to direct observation by an outsider.
One prevalent notion about a limitation of observational tech-
niques, however-the idea that observational data cannot be quanti-
fied-is a mjsconception.' Historically, observational data have, it is
true, most frequen,~ly been presented without any attempt at quanti-
fication. Pioneering work in the use of observational techniques was
done by anthropologists wh9 were studying small, isolated cultures
and did not feel a need for quantifying their observations. The richness
of their data, based as it was on their subtle and perceptive approac:;h,
has tempted other social scientists to adopt similar methods. In so I
doing, they have frequently taken over not only the subtlety of the -
approach but the neglect of the possibilities for quantification as well.
This is not to imply that all observational data must be quantified,
but it is important to note that they can be.
Observation Play serve a variety of research purposes. It may be
used in an exploratory fashion, to gain insights that will later be tested
by other techniques; its purpose may be to gather supplementary data
that may qualify or help to interpret findings obtained by other tech-
niques; Or it may be used as the primary method of data collection in
studies designed to provide accurate descriptions of situations or to
test causal hypotheses. Observation may take place in "real-life" sit-
uations or in a laboratory. Observational procedures may range from
almost complete flexibility, guided only by the formulation of the
problem to be stuc;lied and some general ideas about aspects of prob-
able importance, to the use of detailed formal instruments developed
in advance. The observer may himself participate actively in the group
he is observing; he may be defined as a member of the group but keep
his participation to a minimum; he may be defined as an observer who
is not part of the group; or his presence may be unknown to some or
all of the people he is observing.
In general, the degree of structure and the degree of participation
tend to vary with the purpose of the study. In an exploratory study,
the observational procedures are iikely to be relatively unstructured,
I
and the observer is more likely to participate in the group activity

than he is in a study focused on accurate description of a situation or
on testing a causal hypothesis. However, these characteristics do not
necessarily vary together. The investigator in an exploratory $tudy
may be clearly identified as an observer watching the group, or his
presence may be unknown to the group. For example, an investigator
interested in developing hypotheses about young children's social be-
. havior may watch a nursery school group from behind a one-way
screen; or he may be present with the group, may openly take notes, but
may refrain from any interaction with the children. Participant ob-
DATA COLLECTION: 205
servers may use highly structured observational instruments; for ex-
ample, in an investigation of some aspect of psychotherapy, both pa-
tients and therapists may fill out formal rating scales. And, of course,
there are degrees of structure, rather than a sharp distinction between
"unstructured" and "structured" observation.
Whatever the purpose of the study, four broad questions con-
front the investigator: (1) What should be observed? (2) How should
observations be recorded? (3) What procedures should be used to try
to assure the accuracy of observation? (4) What relationship should
exist between the observer and the observed, and how can such a rela-
tionship be established?
Since the ways of answering these questions differ somewhat de-
pending on the nature of the study and the extent to which observa-
tional procedures can be structured, we shall discuss these questions
separately for relatively unstructured and relatively structured observa-
tion. Because of the frequent use of unstructured observation in ex-
ploratory studies and in situations where the observer participates in
the activity of the group, we shan consider problems specific to explora-
tion and to participant observation in the section on unstructured ob-
servation.
One key issue, however-that of the need for attention to the
correctness and adequacy of observation-is so b~sic, regardless of the
degree of structure of the procedures lused, that it seems appropriate
to discuss it here.
Because all of us are constantly /6~erving in the course of our
daily lives, we may be inclined to ~hink\ that no special training is
needed in order Jo become a scientific otserver. But the observation of
everyday life is haphazard.
, _We. pay attention to some things, not to
others. We may observe with a purpose, as in the case of the traffic
light or the weathe;; we may o'bserve out of curiosity; or we may ob-
serve simply becau~e our eyes are open,' our other sense organs are
responsive, and stiqluli are impinging upon them. Our observations
may agree with those of other people who have observed the same event,
or they may, be q~i,~e different. The selectivity, the inaccuracies, and
the omissions of observation have long provided interesting and in-
structive demonstriltions for social science classes. One of the first
such experiments was conducted by Miinsterberg (1908) early in the

twentieth century:
I stood on the platform behind a low desk and begged the
men [in a psychology class J to watch and to describe everything
which I was going to do from one given signal to another. As
soon as the signal was given, I lifted with my right hand a little
revolving wheel with a colour-disk and made it run and change
its color, and all the time, while I kept the little instrument at
the height of my head, I turned my eyes eagerly toward it. While
this was going on, up to the closing signal, I took with my left
hand, at first, a pencil from my vest-pocket and wrote something
at the desk; then I took my watch out and laid it on the table;
then I took a silver cigarette-box from my pocket, opened it,
took a cigarette out of it, closed it with a loud click,.and returned
it to my pocket; and then came the ending signal. The results
showed that eighteen of the hundred had not noticed anything"
of all that I was doing with my left hand. Pencil and watch and
cigarettes had simply not existed for them. The mere fact that
I myself seemed to give all my attentio_n to the colour-wheel
had evidently inhibited in them the impression of the other side.
Yet I had made my movements of the ldt arm so ostentatiously,
and I had beforehand so earnestly insisted that they ought to
watch every single movement, that I hardly expected to make
anyone overlook the larger part of my actions.
On the basis of a number of similar experiments and of analysis of
testimony presented in legal proceedings, Munsterberg commented:
The sources of error begin " . . before the recollection sets
j in. The observation itself may be defective and illusory; wrong,
associations may make it imperfect; judgments may misinterpret
the experience; and suggestive influences may falsify the data
of the senses.
Innumerable subsequent experiments have confirmed these con-
clusions. They point to the need for careful training of observers and
systematic procedures for checking the reliability of their observations.
Since the nature of the training and checking procedures is somewhat
different in two types of observation, they will be discussed sepa-
rately in the following sections on unstructured and structured ,observa-
tion.
UNSTRUCTURED OBSERVATION 207
Unstructured Observation 3
The technique of unstructured observation has been contributed

mainly by social anthropology, where it has frequently taken the form
of participant observation. In this form of observation the observer
takes on, to some extent at least, the role of a member of the group
and participates in its functioning. Later in this section we shan dea1
with some of the problems and isssues that are involved when the
observer is a participant in the process he observes. However, much
of the discussion is relevant to both participant and nonparticipant
observation when the observer has no predetermined set of categories
to use.
THE CONTENT OF OBSERVATION
The first question the observer must face is: What should be
o~ "Everything" is an unachievable goal, si~e n"ot even the 0
best observer, or the best team of observers, can be expected to provide

a complete record of even seemingly simple events. For example, in
the course of an arithmetic lesson_ in a school classroom, a great deal
is going on. The teacher is engaged in certain activities. Each of the
thirty children is doing something slightly-cIifferen Presymably
learning is taki'i1g])kice, but there are also movements and talk not
directly related to the learning process. It is impossible to record all the
dctails. From the start, one must face the difficult problem of selection.
In more highly structured studies, the formulation of the research
problem inqicates what kinds of data will be most relevant. But in an
exploratory study, where unstructured observati9n is most likely to be
used, one does not knciw in advance which aspects of the situation will
prove to be most relevaht. _
A meaningful ansl"er to the question of what to observe is, indeed,
difficult to give. Since, unstructured observation is often usedc a~n
----
exploratory technique,' the observer's understanding of the situation is
,
3_For'amore detailed'discussion of many problems arising in this type of obser-
vation, with emphasis on/situations where the observer participates in the group
activity, see Whyte (1951,). Much of the material in this section is taken verbatim
from that source, with Dr., Whyte's permission.
likely to change as he goes along. This, in turn, may can for changes
in what he observes, at least to the extent of making the content of ob-
servation more specific; and often the changes called for may be quite
radical. These changes in the content of observation are not undesir-
able. Quite the contrary; they represent the optimal use of unstructured
observation.
Suppose that an observer wishes to explore child-rearing practices
in a foreign culture. tI..e wilLp~bIy begin by observing situations in ~
- -wliich-mOf:~cf child are together. In the course of his initial ob-
servations he may discover that such situations are much less frequent
than he anticipated because mothers in the particular culture go out
to work while fathers or older siblings take care of the infants. As soon
as he has satisfied himself about this fact, the focus of his observational
efforts, will of course, shift to the persons entrusted with the rearing
of the young.
The shift in focus often goes hand in hand with narrowing~ the
scope of observation. Suppose an observer wishes to explore social rela-
tions among the families in a suburban community. He may begin by
observing street life, shopping centers, tile local drugstore; he may
attend club meetings and lectures, watch the crowds in front of the
local theater, visit sessions of the local governing body, mix with
parents waiting for their children a! the close of the school day, etc.
His initial observations may reveal that street life is hurried and un-
conducive to social interchange in the particular community; that the
resident families give their shopping orders by telephone or send .a
maid to the store; that the drugstore is a center of activities only for
adolescents; etc. Probably he will exclude these situations from his
schedule after an initial period of observation and narrow the focus of
attention to those more rewarding for his purpose.
However, although narrowing the range of situations to be ob-
served facilitates observation, it stiIlleaves the crucial part of the ques-
tion-what to observe-unanswered. Among the features of a social
situation that has been recognized as rewarding, which should be
noted?
No hard an~ fast rules can be laid down; the observer must always
be prepared to take his cues from unanticipated events. Nevertheless,
it may be helpful to provide a check list such as the one which follows.
The list indicates significant elements of every social situation; it
suggests directions of observation that may otherwise be overlooked.
l. The participants. Here one wants to know: Who are the par-
ticipants, how are they related to one another, and how many are there?
There are various ways of characterizing the participants, but usually
one will want to know at least the following about any person who is
being observed: age, sex, official function (e.g., "teacher," "doctor,"
"spectator," "customer," "host," "club president") in the situation
being observed and in the occupational system of the broader com-
munity. One will also want to know how the participants are related
to one another: Are they strangers or do they know one another? Are
they members of some collectivity, and if so, what kind-e.g., an in-
formal friendship group, a fraternity or club, a factory, a church?
What structures or groupings exist among the participants-e.g., can
diques, focal persons, or isolates be identified by their spatial group-
ings or patterns of interaction?
2. The setting. A social situation may occur in different settings-
e.g., a drugstore, a busy street intersection, a factory lunchroom, a
nursery school, a slum dwelling, a palatial mansion. About the setting
one wants to know, in addition to its appearance, what kinds of be-
havior it c::ncourages, permits, discourages, or prevents. Or the social
characteristics of the setting may be described in'terms of what kinds
of behavior are likely to be perceived as expected or unexpected, ap-
proved or disapproved, conforming or deviant.
3. The purpose. Is there some official purpose that has brought
the participants together, or have they been brought together by
chance? If there is an official purpose, what is it-e.g., to attend a
funeral, to compete in a boat race, to participate in a religious cere-
r '
mony, to meet as a committee, to have fun at a party? How do the
participants react ~o the official purpose of the situation-e.g., with
acceptance or with!rejection? What goals other" than the official pur-
pose do the partidpants seem to be pursuing? Are the goals of the
various participants:compatible or antagonistic?
4. The soci~l behavior. Here one wants to know what actually
occurs. What do th,~ participants do, how do they do it, and with whom
and with what do they do it? With respect to behavior, one usuaIIy
wants to know the' following: (a) what was the stimulus or event that
initiated it; (b) what appears to be its objective; (c) toward whom or
what is the behavior directed; (d) what is the form of activity entailed
in the behavior (e.g., talking, running, driving a car, gesturing, sitting);
( e) what are the qualities of the behavior (e.g., its intensity, persist-
ence, unusualness, appropriateness, duration, affectivity, manner-
isms); (t) what are its effects (e.g., what behavior does it evoke from
others )?
5. Frequency and duration. Here one wants to know the answer
to such questions as the following: When did the situation occur?
How long did it last? Is it a recurring type of situation, or unique? If
it recurs, how frequently does it occur? What are the occasions that
give rise to it? How typical of such situations is the one being ob-
served?
It should be emphasized that this list is not meant to apply in its
entirety to every situation observed. Frequently it is impossible to
obtai'n enough clues to permit such a comprehensive description. Or
the course I6f events may be too rapid to permit consideration of all
dimensions of a social situation. Or some aspect of an occurrence may
need the entire attention of the observer, to the virtual exclusion of
everything else. The list has its greatest advantage in planning the
content of observational activities.
RECORDING UNSTRUCTURED OBSERVATION
In recording unstructured observation, two questions require

consideration: When should the' observer make notes? How should
notes be kept?
The best time for recording is undoubtedly on the spot and during
the event. This results in a miniI)1um of selective bias and distortion
through memory. There are many situations, however, in which note-
taking on the spot is not feasible, because it would disturb the natu-
ralness of the situation or arouse the suspicions of the persons observed.
This is, of course, especially likely to be true in participant observation.
I Furthermore, copstant note-taking may interfere with the quality of
observation. The observer may easily lose relevant aspects of the sit-
uation if he divides his attention between observing and writing.
Some anthropologists have designed mnemonic devices that may
be adapted by other investigators to their needs. E. J. Lindgren (1935),
for example, says:
I found a few simple devices for remembering things ex-
tremely useful in ethnographic work. Thus during a shaman cere-
mony which may go on for ten hours and presents, as I have
said, peculiar difficulties, I associate the first outstanding in-
cident which occurs with a word beginning with "A," the next
with a word beginning with "B," and so forth. If I am not able
to return to my tent before one or two o'clock in the morning,
when I am generally too exhausted to record much, I can at
least write down these key words to guide me when writing up a
fuller account next day.
In situations in which immediate, detailed note-taking is not
feasible, the memory of the observer may be too heavily taxed if
recording is postponed until the observational period is over. For such
situations it is well to acquire the habit of jotting down significant
key words in an almost imperceptible manner, using a small sheet of
paper, the back of an envelope, or other inconspicuous material. If the
amount to be recorded is so great that this method does not satisfy
the observer, he may well decide, if it is at all feasible, to retire from
an on-going situation for a few minutes every h6ur or two to make
more detailed notes.
However the observer records his immediate impressions, he
should write up, as soon as possible -after a period of observation, a
complete account of everything in the situation that he wishes to re-
member. This write-up will ordinarily be in narrative form. Each
record of a period of observation contains a great deal of information.
As the records of ob~ervation accumulate, it will become difficult or
I
impossible to remember which records contain information on a par-
I
ticular topic and wh~ch do not. Some indexing system is essential if
the observer is to av~id wasting hours searching through his notes for
items he remembers vaguely but cannot locate. The indexing system
should be adapted toI, the purpose of the research, so no flat rules can
be given here. However, the index should probably contain at least
the following information: number or date of observation notes (or
of interview, in the case of participant observation 4 ); group chi~fly in-

volved (unless the study is confined entirely to one group); names of
persons observed or interviewed, and perhaps also of persons discussed
by them; and a brief summary of what is covered in the notes.
As the inquiry becomes more clearly focused, the investigator
may develop a more elaborate indexing system. For example, during
the course of a study of a street gang, he may tentatively decide that
the analysis is likely to center around such topics as relations with other
gangs, relations with adults, relations with girls, attitudes and be-
havior concerning work, attitudes and behavior concerning the use of
drugs. Each of these topics may then become a category in the index.
If the observer has made carbon copies of his notes (always a sound
procedure), he may wish to cut up one copy in order to file the various
items under the relevant index categories. However, in order to avoid
losing the context of a given item, which may be important in its
interpretation, it is desirable to keep one copy of the observation in the
original chronological narrative form. Notes may be made in the
margin showing the index categories to which the various paragraphs
are relevant.
INCREASING THE ACCURACY OF OBSERVATION
When one first observes in situations where immediate recording

is I}ot feasible, he is likely to find that by the time he has an oppor-
t1.i'r'iity to write his observations, he remembers little of what has been
said and done, or remembers it in a vague and confused way. With
experience, however, the ability to remember increases, and the ob-
server comes to feel that he is able to record significant parts of con-
versations almost verbatim. However, such an impression is no substi"
tute for an independent check. 1.n order to check the accuracy- .and
completeness of the observer's record, it would be interesting to com-
4 An observer who participates as a member of a group will naturally talk with
other members of the group. Some of these talks imay serve the function of inter-
views, in that the other members will give the observer information abou~ past
events, their reactions to present events, etc. Such interviews will usually- be un-
structured (see Chapter 7); the observer may combine them with more strictly
observational notes in' his records and in the analysis. For a discussion of the rela-
tive advantages of observation and unstructured interviews, and of wayS of com-
bining the two, see Dean (1954).
pare it with a record made by sound-recording equipment. However,
this is not often feasible; as far as we know, no such comparison has
ever been made. 5 In any case, a sound recording captures primarily
what is said; it cannot portray gestures and other nonverbal behavior.
To supplement sound recording by a motion-picture record would
ordinarily be prohibitively expensive, even if it were feasible on other
grounds.
The next-best solution is to have two or more people observe the
same event. When two or mOre observers are watching and recording
in the same problem area, they have opportunities to compare their
findings and check bias. It is desira'ble for them to make independent
records first, so that the written records can be compared. This is an
excellent way to discover one's blind spots. When both are participant
observers in the situation, each will have opportunities to observe the
other in action and to check on how people are reacting to him. This
provides each observer with a valuable check on himself. Even if it is
not possible to use more than one observer throughout the course of
a study, it may be desirable to do so in the early stages, or at least
during a pilot period in a setting similar to the one which is to be
observed in the investigation.
The use of more than one observer, however, will in itself provide
no clues to biases of interpretation that are common to all. When
observers have a common cultural background and similar training-
as is usually the case-it is inevitable that they will share certain gen-
eral ways of perceiving and interpreting events. Whether these ways of
perceiving and interpreting constitute, a serious source of potential
bias depends to a large extent on the nature of the data being collected.
If, in observing a street gang, one is interested,primarily in such ques-
tions as which members are central, which ones peripheral, the ob-
server's point of view may make little difference. But if one is inter-
ested, for example, ih the amount of aggression or friendliness the
members show toward one another, the observer's point of view may
I ~
have an important ihfluence. Behavior that a middle-class observer
5 Studies of intervie~ing in which tape recordings were used have shown that
even when the intervievjer is recording responses as they are made, omissions and
distortions occur (see Hyman et aI., 1954). It seems reasonable to suppose that
they would occur even more frequently when observations and conversations are
recorded from memory.
interprets as aggression may seem like good-natured horseplay to one

who has himself grown up in a street gang. Where interpretations of
this sort are important to the study, it is desirable (though often not
feasible) to have two observers from quite different backgrounds ob-
serve the same situation.
Whether an observer is working alone or as a member of a team,
he may increase the objectivity of his observations by indicating, as he
writes up his notes, which statements ref~o actual events and which
represent his interpretations. This is no easy task. Interpretation of the
meaning of a situation must be present to some extent in the observer's
mental set during the act of observation; otherwise it would be im-
possible for him to perceive the relationships among discrete move-
ments, gestures, statements, and objective conditions of the situation.
When the observer participates with the people he is observing, at
least to the extent of asking them questions, his own tentative inter-
pretation of a situation may lead him to ask certain questions. These
questions, in turn, may channel and focus the informant's replies,
possibly even suggesting to him certain interpretations of events that
h__emight not otherwise have thought of.
Notwithstanding the difficulty of the task, efforts in the direction
of trying to separate observations from interpretations can be made
and will be rewarded by greater understanding of the situation. An
overdose of interpretation in an observer's record may seriously inter-
fere with the validity and reliability of his ultimate conclusions. One
way of detecting the intrusion of interpretation is to have two ob-
servers record the same event according to the same'system. If their
accounts differ, it is not hard to determine whether t,he differences re-
sult from the inclusion of more detail by one or from the intrusion of
interpretation on the part of one or both. (Agreement between them
does not necessarily indicate tha"t interpretation has not entered. They
may both have made the same interpretation). In the case of reports
by informants, the investigator may become more sensitive to the pos-
sible influence of his Own tentative interpretations on the informant's
statements if he records the questions he asked and his rea~ons for
-asking them.
The participant observer faces especially severe difficulties in
maintaining objectivity. He is likely to develop friendly reiations with
some of the people he is studying. He may find himself personal1y
concerned with a story told by a certain informant. Full recording
helps to restore objectivity. As the story is recorded, the observer has
something like a catharsis experience himself. He is still interested in
the informant and his story, but he can now look at the record more
objectively. Instead of feeling, with the informant, that the informant
did the right thing and the other fel10w was wrong, the observer be-
gins asking: Why did the informant act as he did? He seeks to explain
instead of defending or accusing. A full record of interviews and ob-
servations is as important in maintaining the objectivity of the re-
searcher as it is in providing research data.
Becoming involved in a situation may lessen the sharpness of
observation not only because the investigator identifies with his in-
formants but also because he becomes accustomed to certain kinds
of behavior. To get access to intimate data, the observer allows himself
to become absorbed in the local culture, but then this very absorption
process makes him take for granted behavior that he should be trying
to explain. Whyte (1951) reports that, as he began the investigation
which he reported in Street Corner Society, everything he saw and
heard was new and strange, and all sorts of questions arose in his mind.
But at this point he did not know enough or have enough data to ask
good questions and get any answers. As he be~me part of the com-
munity, the richness of the data increased; yet he found in himself an
increasing tend~ncy to take for granted the sort of behavior that waS
taken for granted by the people he was observing. .
How can this problem be met? The observer can hardly avoid
becoming accustomed to the setting he is studying and thus becoming
blind to valuable data. He can help himself to some extent simply by
being aware of the tendency to take things' for granted. If he writes
progress reports at: fairly frequent intervals, he will find, as he reads
I
them over, points ~t which explanations are weak or at which they are
needed and not pro~ided. It is still more helpful to describe and explain
his observations r~gularly to someone outside of the situation. The
)
outsider does not take as much for granted, and his questions are a safe-
guard against growing blind spots.
Reference to i check list, such as the one on pages 209-210, may
also be helpful in ,overcoming blirrd spots, especially if it is reviewed
with the attitude, "Have I been overlooking anything about this item
which is relevant in the context of this study?"
It is also possible to overcome blind spots by deliberately breaking
up the perceptual field so that the factors that lead it to be seen in a
particular way lose much of their force. The natural way of seeing a
situation (and the most valid for most purposes) is to see the action
centered around the principal characters. But sometimes the real
center of the action is not the obvious one. For example, one of our
associates has informally described a family that he has known fairly
intimately for many years. It had always seemed obvious to him that
the mother was the central character in the group. She was the manager,
the disciplinarian, the one who gave direction and set limits to the
activities of the children. The father seemed like a negative quantity.
He rarely spoke. When he came home, no one seemed to notice him.
There was never any exchange of greetings. He would be there, reading
a book; then you would notice that he was no,longer there, with hardly
more than a softly closing door to mark his departure. The children
developed certain behavioral disturbances, and itwas in trying to under-
stand these that our colleague eventually real~ed that he had totally
misperceived the family constellation. Actually, that entire family
gravitated around the person of the father. The mother was constantly
interpreting the wishes of the father, regulating things so that they
would fit in with the father's notions of how things should be. The
children were. very much aware that the ultimate source of approval
and disapproval was the father. And both mother and children attri-
buted a mystical power to the fath'~r's few words; they shared a belief
that even his most casual remark was bound to come true.
Sometimes one discovers that a parent who has been dead or
missing for many years is, nevertheless, the real center of a situation.
In any group, important leadership functions are not necessarily vested
in the manifest leaders; ther~ may be a variety of behind-the-scenes
sources of power without formal leadership status-individuals who
crystallize opinion, individuals who take over the organization of
. actions in emergencies, individuals who can block particular lines of
action, individuals who take the center of the stage on certain social
occasions, etc. By deliberately refocusing on individuals who do not
appear to be central in the group, one may gain new insights about im-
portant relationships.
A very different kind of check on the accuracy of observation and
interpretation may come from the people who are being observed, if
the investigator establishes the sort of relationship with them which
make it possible for him to take them into his confidence about the
research. Whyte reports, for example, (1951) that in gathering mate-
rial for Street Corner Society, he had innumerable research discussions
with "Doc," one of the key figures in the group he was observing, and
that Doc read every page of the first draft of his manuscript. Usually,
of course, the participants in the situation cannot check on the validity
of theoretical interpretations, but they can ten the observer whether he
has caught the meaning the situation and the behavior have for them.
Rosenfeld (1958) has suggested that the situation of being a par-
ticipant observer is likely to create inner conflicts within the investi-
gator which may interfere with objectivity. She points out that,
especially if the group being observed is undergoing an emergency of
some sort, there is strong pressure on the observer to become an active
participant, to the extent of abandoning at least temporarily his de-
tached position as an observer. If he does not do so, he may feel guilty
about not helping when help is needed. On the other hand, if he does
enter completely into the activities of the group,/he becomes anxious
about losing his identity as a scientist. In order to re-establish his posi-
tion as an objective investigator, he may lean over backward to separate
himself from the group he is observing; in doing so, he may become
susceptible to sources of negative bias and distortion. Rosenfeld
suggests that the first step in safeguarding against bias arising from
inner conflicts is to be aware of the conflicts and of the nature of one's
defenses. With this awareness, one can develop specific safeguards
appropriate to the mature of the conflicts and the situation being
studied.
THE RELATION OF OBSERVER AND OBSERVED
The need to prepare both oneself and the field carefully for ob-
servation in a real-life setting cannot be emphasized too strongly. For
here, more than in many other techniques, mistakes in approach are
heavily penalized. If, in an inquiry based on a sample, a faulty approad.

to a respondent leads to a refusal to be interviewed, another respondent
can, as a rule, be substituted without much harm to the inquiry. (If
there should be many such cases, one would, of course, become con-
cerned about the possibilities of sampling bias.) In field observation,
however, a faulty approach to a key person may have dire consequences
for the entire inquiry. Since the method is applied in the actual life
sphere of the subjects, where people are in contact with One another
and exchange views or pass rumors, the observer is inevitably talked
about, and his mistakes cannot remain isolated incidents.
Befor:e he approaches anyone in the community or group he wishes
to study, the observer-to-be must decide whether l;le will reveal the fact
that he is a research worker or whether he will attempt to enter the
situation under some other guise. Ordinarily, it seems preferable to
make known the fact that one is doing research. First, this is often the
simpler:procedure; it is not easy to turn up suddenly as a member of a
street ~ng, or as a skilled carpenter, or in some other role that provides
a strategic vantage point for observation. Second, identifying oneself
as a research worker frequently increases one's. opportunities to get
information. While one can learn much about the life of a factory
worker by taking a job in a factory, introspecting on his own experience
and watching the behavior of others, he may need to supplement this
by getting people to explain what they are doing or why they are doing
it. He needs to be able to ask questions-the sort of questions that
would not be asked by a bona fide factory worker. Third., the investi-
gator who proposes to enter a situation without revealing his research
purpose has an obligation to ask himself whether there is{lny possibility
that his disguised activities will harm any of the people in the situation
and, if so, whether the potential results of nis research are valuable
enough to justify their acquisition under thC!se circumstances. 6
6 This is only one of many ethical questions that may arise in connection with
research in the social sciences. At every stage in the process-from the selection
of a topic to the reporting and application of results-consideration of ethical
implications may be relevant. In raising and answering such questions, the inves
tigator would do well to turn to the" publication of the American Psychological
Association (19)3) entitled Ethical Standards of Psychologists, especially pages .
1'13-124; to the discussion by Rogers and Skinner (1956); and to the volume of
the T(lumal of S(lcial Issues ~evcted tr '.\.'';]!'~~ lmd the Social Scientist" (Benne
anc S'i>~n!K'~. ! 5~\
Sometimes, however, an investigator may decide that knowledge
of his research interest would interfere with the behavior he needs to
observe in order to answer his research problem, and that disguised
observation would not have damaging effects on the people observed.
For example, in a study of the sociopsychological effects of long-term
unemployment in an Austrian village (Jahoda-Lazarsfeld and Zeisl,
1932), the study design required an observation of the standard of
maintenance in the households of the unemployed. Because th(
Austrian unemployment aliowance was linked to an official investiga.
tion of the actual needs of the unemployed, the families were under-
standably reluctant to admit curious strangers into their homes. To
overcome this difficulty, a large collection of used clothing was brought
to the village. On the pretext of having to establish the most urgent
clothing needs of every family, research workers presenting themselves
as members of a voluntary welfare group found entry into the homes,
where they were welcomed with open arms when their mission of
distributing clothing became known.
Two recent examples of data gathered by disguised observers that
probably could not have been secured by any other means are presented
by Sherif and Sherif (1953) and Festinger, Riecken, and Schachter
(1956). In the former case, the gang activities of boys in a summer
camp were observed by a resean;h worker disguise,d as a laborer on the
camp grounds. In the latter case, a sect that had predicted the end of
the world was observed by investigators posing as converts to the sect.
Entrance into a community as an admitted research worker may
require somewhat less ingenuity than entrance under some other guise,
but nevertheless it requires careful staging. Depending on his specific
task, the observer will wish to be in good rapport with many strata of
the population. If h~ studies a factory, for example, he will have to be
accepted by both management and workers. Unless he wishes to run
I
the risk of being identified with one side in the industrial set-up, he
must, so to speak, ~tage a simultaneous entry into both sides of the
plant. The problem becomes even more 'complicated in community
studies, in which there are, as a rule, many more than two sides to
be approached sinjultaneously. The essential task here is to avoid
premature identifi~ation with one side or faction of the community.7
7 For a more detailed discussion of this problem, see Merton (1947).
settings, the investigator can arrange the major aspects of the situation
in such a way as to suit his research purposes and reduce the danger of
unexpected interference from disturbing factors. Few of the prob-
lems that arise with observation in a community setting need trouble
the observer if he can arrange and control the situation. Here his
observational activity is often, but not necessarily, reduced to noting
the presence, absence, or intensity of clearly specified types of behavior,
much as the animal experimenter observes a rat's behavior in a specific
way under controlled conditions. Of course, such control of the situa-
tion is appropriate only when the investigator already possesses a great
deal of information about the phenomena he wishes to study .
.' Katz, Goldston, and Benjamin (1958), for example, created a series
'of controlled situations to test predictions 'about Negro-white social
interaction. The hypotheses were suggested by findings in field studies
of interracial contact and in experiments on the dynamics of small
face-to-face groups. Male college students, white and Negro, were
"hired" to work together in groups of four (two whites and two
Negroes) for several three-hour sessions. M~mbers of each group re-
ma'ined together throughout their employmenf and had no contact
with other groups. The subjects were given various group tasks (osten-
sibly materials that were being developed for vocational aptitude tests),
which included mental problems, mechanical construction, human
relations problems, map drawing, and a game that required a high
degree of coordination of effort among the four participants. Two
types of hypothesis were tested. First, there were predictions having to
do with the effect of the general 4ifference in social sfatus between
Negroes and whites in our culture on the content and direction of
communication between them: that the ~hite students' would tend to
ignore the Negroes, and that the Negro students would speak less than
the white participants and would direct most of their remarks to the
latter. The second group of predictions concerned the effects on
interracial behavior of two experimental variables: group reward versus
individual reward, and high group prestige versus no prestige. It was
hypothesized that group rewaJil and group prestige would tetid to
reduce the divisive, effects of disparity in social status-specifically, that
these experimental conditions would bring about greater friendliness
Ilnd cooperation between men of the two races, less behavioral restraint
STRUCTURED OBSERVATION 223
on the part of Negroes, less "bias" in the direction of communications
of both Negroes and whites, and higher group productivity.
In order to measure amount and direction of relevant types of
behavior, it was first necessary to develop a set of reliable categories for
systematic observation and recording. Many weeks of preliminary work
with pilot groups preceded the establishment of a satisfactory classifica-
tion scheme. It consisted of twenty-eight interaction categories for
describing such facilitative and disruptive behaviors as giving help,
advice, information, and encouragement; rejecting another's sugges-
tions, hoarding materials, disparaging another's contributions, ex-
pressing anger, and so on. Direction of behavior was recorded by noting
the initiator and recipient of every social action. Specific categories
were developed for the various tasks, so that the unique effects of each
work situation could be ascertained. The observers were in a room
adjoining that in which the subjects worked, behind a one-way screen;
a recording device made it possible for them to hear what was being
said. Although the subjects could not see the observers, they were told
that they were being observed; the reason given them was that, in
order to improve the aptitude tests, it was necessary to have a detailed
record of how people went about working on them.
On the whole, the results of this rather elaborate investigation
tended to support hypotheses about the effects of status disparity and
of group reward; predictions about the effects of group prestige tended
to be contradicted.
THE CONTENT OF OBSERVATION
Structured observation, as has been ind~cated, is used mainly in

studies starting with relatively specific formulations. Normally, there-
fore, there is much less freedom of choice with respect to the content
to be observed than In unstructured observation ..
\
Since the situation and the problem are already specified, the
observer is in a position to set up in advance the categories in terms oj
which he wishes .to analyze the situation. When he starts out, he i3
likely to have a cO~$iderable number of categories. As he tries out his
instrument, he may find both mechanical problems in observation a,'J
failures in reliability. To meet these problems, categories are drop;)_J
or combined or formulated more clearly. By the time the "real" observa-

tions are to be made, either in a field setting or in an experiment, the
categories will have been defined well enough to provide reliable data
on the questions to be asked. For example, Bales (1950) at first used
more than fifty categories in observing group situations. As his ex-
perience with reliability accumulated and as his theories developed, the
number of categories gradually decreased. The final version, shown
opposite, was a set of twelve standard behavioral categories applicable
to a wide range of group situations. Behkvior of any group member,
or of the leader, is coded in terms of careful definitions for each cate-
gory.
During the first attempts to use an observational instrument, it is
us~ful to check, by means of interviews with the persons observed,
whether, according to their own account, they are doing or feeling
1"- what the observer has described them as doing or feeling.
THE OBSERVER'S FRAME OF REFERENCE. An observer may categorize
the beh~yior, of, a person in terms of the assumed reactions of other
group menioers, or he may categorize it in terms of the intention the
speaker probably had. The procedures developed by Bales instruct the
observer to do the former. The method developed by Steinzor (1949)
requires the observer to do the latter. Obviously either or both observer
sets can be used. What is important is to make a decision about the ap-
propriate frame of reference and to train observers accordingly.
Thelen and Withall (1949) reported substantial agreement
among observations of the same classroom groups by three observe!
teams, each working within a different frame of reference. The ob-
servers categorized the behavior of a teacher according to whether it
was "learner centered" or "teacher centered." Several observers con-
centrated on ,the objective behavior of the teacher. An equal numbel
made inferences about the intents' and attitudes of the teacher. The
third group was made up of the class members themselves, who in-
dicated how they felt at various points by pressing levers attached t(1,
their desks. (The lever pressings were recorded by a kymograph.)
Observations made according tp these different frame~ of reference
'were in close agreement in identIfying the teacher's behavior as teacher-
centered or learner-centered. It appears, .then, that at some, levels of
data collection, the nature of the observer's set 'may make little dif-
THE BALES SYSTEM OF CATEGORIES FOR RECORDING
GROUP INTERACTION9
1 Shows solidarity, raises other's status,
gives help, reward:
Social- 2 Shows tension release, jokes, laughs,
Emotional A shows satisfaction:
Area:
Positive 3 Agrees, shows passive acceptance,
understands, concurs, complies:
B
4 Gives suggestion, direction, imply-
ing autonomy for other: .
5 Gives opinion, evaluation, analysis,

'--
J
expresses feeling, wish:
6 Gives orientation, information, re-
Task peats, clarifies, confirms:
Area: abedef
Neutral
j
7 Asks for orientation, information,
repetition, confirmation:
8 Asks for opinion, evaluation, analy-
c sis, expression of feeling:
9 Asks for suggestion, direction, pos- I+-
sible ways of action: /
10 Disagrees, shows passive rejection,
formality, withholds help:
Social-
Emotional D 11 Shows tension, asks for help, with-
Area: draws out of field:
Negative 12 Shows antagonism, deflates other's ,
:status, defends or asserts' self:
KEY:
a Problems of Communication A Positive Reactions
b Problems of Evaluation B Attempted Answerr:.
I
c Problems of Control C Questions
d Problems of Decision D Negative Reactions
e Problems of Tension Reduction
f Problems of ReIntegration
9 Reproduced from Bales (1950).
DATA COLLECTION: I
ference. The more one is interested in details, however, the more im-
portant the choice of the proper frame of reference may become.
It often happens that certain data can be coded only in retrospect.
Whether or not a remark precipitates group tension, for example,
can be determined only in the light of the events following that re-
mark. To handle this type of categorizing, some studies have made tape
recordings of the observed situation; others have demanded that the
observer pause periodically to go back over his notes in order to make
codings in the light of subsequent events.
TIME UNITS. The amount of time included in one notation by
an observer may range from a few seconds to several hours. The central
problem in setting up time units is to determine what a psychologically
meaningful unit would be. For example, it may.not be sensible to
make a rating of the constructiveness of a child's play with a certain
toy every two minutes. Such a rating may have to be based on t.he
complete sequence of events in a child's use of a toy. A typical way of
meeting this problem is to use more than one observer. One of them
watches for those acts that must be poted a~ they occur, such as brief
comments or small bits of motor behavior. Another takes a larger view,
)loting those behaviors that would be distorted, by strict adherence to
time or other sampling citeria, and codes thes~ in terms of an index or
rating scale whenever he feels that he has enough data for his purposes.
WHAT IS AN ACT? The definition of an act is difficult when one is
attempting to categorize the verbal behavior of a person. Is an act a
sentence, a pause for breath, a complete thought, or the least _notice-
able difference between one idea apd another in a given speech? An
act is even more difficult to define when recording ~otor behavior,
since the person being observed seldom separates his movements one
from another as neatly as the categories on the observation schedule.
The problem is still further confounded 'Yhen dealing with group
yhenQmena. Is the act of a group a speech by one member, a decision
\reached, a function carried out, an event, a set of events, a program
item, the completion of an item on an agenda, a mood shift, or what?
The definition of an act will be determined by the frame of reference
used as well as by the size of time units recorded. "
The most frequent practice in coding verbal behavior is to code
each complete thought separately. Thus, one sentence may includr
thoughts that fit several categories or none at all. The observation of
motor behavior is usually concerned with the general nature of the
behavior, such as sitting, walking, slouching, gesturing, handling, etc.
The recorded acts of a group, on the other hand, are usually decisions,
or the completion of items on an agenda or of certain phases of the
meeting.
RATING SCALES VERSUS ALL-OR-NONE CATEGORIES. If the research pur-
pose requires only a record of the objective facts of behavior, without
any further qualification, all-or-none categories are usually adequate.
"Speaking" versus "not speaking" is an example of an all-or-none
category; the observer simply places a tally on his sheet when a given
person speaks, and makes no tallies for this person until he speaks
again. A series of discrete categories intended to describe the behavior
of the speaker may take the form of all-or-none tallies. For example, in
one study the observer noted, for each act of a youth-group leader,
whether the leader was acting in the role of a comrade, a policeman,
a referee, an educator, or a coordinator.
Often, however, observers are asked to describe the behavior of
a group member by making ratings on scales. In watching the leader
above, for example, the observer rated each leadership act in terms of
the degree of freedom implied and the strength/of the influence ex-
erted on the group members by the act.
PRESERVING THE PATTERN OF THE PHENOMENA. The personal inter-
actions or group phenomena in any situation make up a number of
themes which it may be important to preserve in their original pattern.
In many studies, of course, it is sufficient to get a record of acts apart
from their context. An example is that of tallying the frequency of
remarks addressed to a deviant group member. In other studies, how-
ever, the main function of the data is to specify the nature of some
general pattern of behavior in the group, such anhe teaching method
being used, the rise ~md fall of group tension, or the establishment of
group procedures forihandling the feelings of members.
In part, this i~ a problem of analysis. However, a reliable observa-
tional device may t\:\rn out to provide only fragmented information,
which cannot be combined into a meaningful picture. It is necessary,
therefore, that the tryout stage of any instrumeI\t include attempts
to code the data in order to make sure that information of the desired
type can be obtained.
A study of leadership among nursery school children in Hungary
(Merei, 1949) provides an example of how a complex set of observa-
tions recorded by a number of different observers can be combined to
give a total pattern. Merei was interested in the relationship between
leader and group, particularly in whether leadership at this early age
is maintained when a child leader is put into groups which have various
degrees of cohesiveness and with which he is not familiar. Some of
the criteria Merei used for identifying leadership behavior were: giving
orders more frequently than following orders; being imitated more
frequently than imitating; attacking more often than being attacked.
Categories for recording the behavior of each child were established
according to these criteria. In addition, categories were developed to
describe other dimensions of the group situation. It was decided that
several observers were needed to note all of the relevant dimensions of
behavior. Some of the observers concentrated' on describing the activi-
ties of the group by five-minute intervals; others recorded the behavior
of the leader, his relation to other children and to the on-going activity,
for the same intervals. Others singled out the following dimensions of
behavior: group formation and isolation; imitation; order-giving and
response to orders; taking the initiative; ownership and change of
ownership of toys; etc. Typical forms of behavior were itemized und~r
each of these categories. In the ownership category, for instance, the
fOllowing forms were distinguished; the child asks for somethi.ng and
receives it; the child asks for something and does not receive it; the
child takes an object away from another child; etc. Ea9h form of be-
havior was given a special written symbol to facilitate speed of reco~d
ing. When the observers had mastered this symbol language, the actual
recording could proceed at considerable s,J?eed. Synchronizing the
various protocols at fivecminute intervals made it possible to reconstruct
the complete scene when the data were analyzed.
RECORDING OBSERVATIONS
There is no one best method of recording observations, although

some procedures yield certain kinds of data that .others cannot. The
simplest and most economical device that will yield the required data
is the One to use.
The most frequently used system of recording is one that provides
the observer with a number of duplicated sheets containing the list of
categories to be coded and the cells in which they are to be marked.
An economical variation was used by Lippitt and Zander in an unpub-
lished study that required field observations of group leadership. Each
of the observers had a ring-binder notebook containing sets of cards of
the type used in machine analysis, on which the observational catego-
ries had been printed. The observers made a pencil mark on the propel
card opposite the appropriate category. These cards were then run
through an automatic punch machine which was sensitive to graphite
and which punched the cards wherever a mark had been made. The
cards were then ready for machine sorting without the usual expense,
time loss, and danger of error involved when a punch-machine operator
transfers material from code sheets to a card.
Mechanical recording instruments have been used in some studies.
Chapple (1949) has devised an "interaction chronograph," on which
the observer indicates who is talking by pressing a button for the
duration of each person's statement. The machine is so designed that
it will automatically summate, at the end of the observation period,
certain frequencies of interaction among group members. Thelen has
developed an "audio-introspectometer," which allows observers or
members of the 'group to record their feelings or other observations
by placing a desk-lever at one of several precoded positions.
Bales and Gerbrands (1948) have described an "interaction
recorder" which they have found useful in dealing with a large number
of categories. This d~vice consists of a box containing a roller mechan-
ism. On top of the Ibox, at the left, is fastened a sheet showing the
categories in terms ?f which observations are 'being made. A roll of
paper, as wide as theisheet showing the categories, is moved from right
to left across the tOR of the box by a roller me~hanism. At any given
moment, a blank P?rtion of this paper appears at the right of the
category sheet; the observer makes the appropriate entries on this blank
section of the paper. Suppose, for example, that he is observing a
small group in term~' of the categories shown on page 225. Let us say
that member #1 makes a remark that the observer categorizes as
"gives opinion"; he writes a "1" on the paper in the position next to

this item on the category sheet. As the roller continues to move the
paper, the portion with this mark now disappears into the box, and
an unmarked portion of the paper is again available to the observer.
Here he records the response of another member to # l's opinion-
perhaps a "5" in the position corresponding to the item "disagrees."
Since the strip of paper on which he records his entries is constantly
moving, the final record shows the sequence of behavior.
Sound recordings and motion pictures have been used when it is
necessary to describe the over-all nature of an event or to code certain
actions of members or of the group in terms of the frame of reference
provided by the entire event. Tape recordings have often been used
for this purpose. Here, however, one loses the direction of the remark
and the gestures, nods, and bodily postures which are often needed
for a full comprehension of the behavior of the speaker as well as
that of the recipient of the remark. Silent motion pictures have limited
usefulness because remarks are omitted. Sound movies are so expensive
that they have seldom been used.
Although such devices as tape recordings, 'motion pictures, and
television may be very helpful in giving an over-all view of a social event,
their use does not in itself solve the problem of gathering data for
!systematic purposes. Relevant categories for recording behavior must
he established, time units decided upon, methods set up for recording
who ini~iated an action and to whom it was addressed. Observations
may be recorded either at the time the event occurs (that is, simul-
taneously with the making of the motion picture or the tape record-
I
ing), or coding may be done from the movie Or the recording. But at
some point, if the data are to be used in a quantified way, th'ey must be
recorded in terms of such aformal scheme.
/
fNCREASING THE RELIABILITY OF OBSERVATION
In one sense, all that has been said thus far is relevant to the
problem of obtaining adequate reliability in the use of an observational
instrument. That is, wise and consistent procedures iu the develop-
, ment of such an instrument will greatly enhance reiiability, assuming,
of course, that the observers have been trained to interpret their in-
structions similarly and have practiced enough to develop the skills
necessary for proper categorizing and recording. There are some
special problems, however, in achieving reliable and valid observation
that are worth separate consideration.
One problem grows out of inadequate definition of the kinds of
behavior that are to be taken as corresponding to a given concept.
Berkowitz and Guetzkow (1949), for example, have pointed out that
in categorizing different groups in terms of "pleasantness of group
atmosphere," one observer might be especially sensitive to the per-
sonal liking of the group members for one another, another to the
informality of the relationships among members, another to still a
different dimension of pleasantness of atmosphere.
Another factor that may lower the reliability of even well-trained
and skilled observers is the degree of confidence one must have in
one's judgment before marking a given category. If observers are re-
quired to rate the presence of "ego need" behavior (an example used
by Berkowitz and Guetzkow) even when little behavior relevant to
such a rating occurs, one observer may rate the extent of it as being
greater than it "actually" is simply because he himself has a predisposi-
tion to perceive evidence of "ego need" behavior; another observer
may rate it as lower than it "actually" is because he requires more evi-
dence or, perhaps, greater confidence before making a decision about
the presence of "ego need" behavior.
One of the greatest sources of unreliability is the constant error
introduced by the observer because of distortion of his perceptions by
his own needs Or values. An observer who sharply disapproves of certain
leadership practices, for example, will have difficulty in preventing a
bias; he may code more of the leader's behavior as falling in the
categories he disapprovbs of than would another observer who feels
less strongly on the matter. Adequate training and practice can over-
come this in most persons, though not in all.
I
Although,research observers have been able to rate a group reliably
on as many as fifty categories and even mOre, there is a point at which
the load can seriously hamper reliability. The major result of overload-
ing is that the observd,cannot record all the relevant data and may
unwittingly record some aspects more adequately than others, thus
giving a biased account. This may result from fatigue, which causes
the observer to slow down and later spurt; from avoidance of more
difficult categories in order to keep up the pace; or from any of a
number of other reasons. Overloading can be prevented by standardized
rest periods, by distributing the job among several observers or by
mechanically recording the session (if such recordings are suitable).
Obviously, one important method of increasing reliability is care-
ful training of observers. A well-developed observational procedure can
be spoiled by differences among users or by failure to understand the
rules for its use. It is necessary, therefore, that the investigator plan to
invest a good portion of time in training the observers. The elaborate-
ness of the training depends on the complexity of the observer's task.
One study, which used almost one hundred observers in several dif-
ferent parts of the country, determined the length of the training
course by the amount of time it took the trainee to obtain adequate
agreement with the observer-trainer. When he had achieved sufficient
skill, he was allowed to begin observations.
A typical training program begins with an explanation of the pur-
poses and theory involved in the given study and then moves on to an
explanation of the categories and the rules fortheir use. The purpose
of each category in relation to the theory and to specific hypo'theses is
pointed out. After the trainees have had an opportunity to ask ques-
tions, they try to use the schedule on a group that is attempting to
demonstrate phenomena of the type the observers will be expected to
code when the actual data collection begins. The trainee naturally
encounters difficulty in selecting proper categories, sampling, keeping.
abreast, deciding how to categorize marginal cases, etc. These difficul-
ties are ironed out by discussion and further practiC)::. Next comes a
tryout"in.a pilot study oh a group similar to the one the trainee will later
be expe'cted to observe. Here again difficulties arise which ~n be cor-
rected. At this point, or a little later, it may be helpful to use tape re-
cordings or motion pictures in orq.er to check events that were coded
differently by different observers. Now the observers are ready Jor reli-
ability tests, followed by subsequent practice and more,reliability tests
until the trainer is satisfied that the observers have become useful and
I comparable measuring instruments.
Here too, as in the case of less structur~d observation, checking
and increasing reliability does not eliminate the possibility of a con-
stant bias shared by two or more observers.lO There are no simple tech-
niques for dealing with this problem. If it seems likely that such con-
stant biases may seriously affect the findings, it may be desirable to
have two or more observers with different backgrounds record the same
events, at least during a preliminary period.
THE RELATION OF OBSERVER AND OBSERVED
Unless he is concealed behind a one-way screen, his presence

unknown to the group, the nonparticipant observer faces the same
kinds of problems as the participant observer in establishing relations
with the people he is observing. He, too, must carefully prepare his
entry into the situation and make sure that all members of the group
are willing to accept him. Since he is usually conspicuously engaged in
recording behavior on prepared schedules, using a timing device and
perhaps other technical aids, it is ordinarily not possible to disguise the
fact that he is doing research. Hence, it is -all the more important that
he obtain the group's full agreement to the inquiry. In some situa-
tions, however, only the 'COnsent of the group leader is necessary, and a
rather vague explanation may satisfy the group members. Whenever
possible, irivestigators usually attempt to create an impression of the
observer as a neutral, psychologically non visible person.
Conceivably, the entry of an observer into a group, however un-
obtrusive, may introduce another variable into the situation that may
change the behavior being observed. In children's groups, for example,
the presence of adult observers may be more potent than the particular
experimental variable that is supposed to be operating. Deutsch (1949)
found that college students who were in a competitive situation said
that they were moreiaware of the observers than comparable students
who were in a cooperative situation. It is important that some thought
I
be given tq ways in which the observer's presence may influence the
outcome of the research, and to developing techniques that reduce
this possibility.
On the whole, however, people seem to get used to observers if
the behavior of the observers convinces the group members that they
10 See pages 213-214.
are no threat. Deutsch also found that the members of small groups
were much more aware of the observer's presence (as indicated on a
rating scale marked after each weekly meeting by the group members)
at the beginning of their experience with them than they were after
they had been observed for three meetings. Many investigators believe
that it makes little difference in the observed behavior of the group
members whether the observer sits in the room with the group, behind
a one-way screen with the group aware that he is there, or behind a
one-way screen with the group left to wonder whether he is there or not.
7
DATA COLLECTION
II. Questionnaires and Interviews
Comparison of Interview and Questionnaire

Question Content /
Types of Interviews and Questionnaires
The Sociometric Method
Visual Aids in Interviewing

A Concluding Note
If we want to know how people feel: what they experience and
what they remember, what their emotions and motives are
like, and the reasons for acting as they do-why not ask
them? G. w. ALLPORT
O BSERVATIONAL METHODS, as we have seen in the precedinE

chapter, are primarily directed toward describing and under-
standing behavior as it occurs. They are less effective in giving informa-
tion about a person's perceptions, beliefs, fe~lings, motivations, antic-
~
ipations, or future plans; and certainly they provide no information
about past behavior or private behaviors, "Such as sexual activity or
dreaming, which are, by their very nature, either unfeasible or impos-
sible to observe. To obtain such information,. the interview, the ques-
tionnaire, and the projective method have been devised.
In the interview and questionnaire approach, heavy reliance is
placed on the subject's verbal report for information about the stimuli
or experiences to which he is exposed and for knowledge of his behavior;
usually the investigator has not observed the events discussed. The
subject's report mayor may not be taken at face value; it may be, in-
terpreted in the light of other knowledge about him or in terms of some
psychological theory; inferences may be drawn about aspects of his
functioning which he has not reported. Regardless of the amount and
kind of interpretation, however, the starling point is the subject's self-
report. Thus these approaches can ordinarily obtain only material that
the subject is willing and able to report.
For many years now, a controversy has been raging in the.,psycho-
logical literature about the validity of verbal reports. The question is:
How do we know a person is really hungry when he says, "I am hun-
gry"? There are many weighty issues involved in this question, which
236
DATA COLLECTION: II 237
we shall not discuss. 1 Let us point out, however, that in everyday life
we accept many verbal reports as valid. For example, if we ask a friend
what he thought of a certain play and he says "It's terrible," we ordi-
narily believe that his statement of leeling is correct; whether his feel-
ing is appropriate to the play is, of course, another matter. However, in
everyday life we also realize that in certain circumstances verbal reports
are not to be trusted. For example, whenever we have reason to suspect
that a person's truthful self-report would be embarrassing, humiliating,
Dr degrading, or would in some way place him in an unfavorable light,
we are likely to entertain some reservations about a report that shows
him in favorable light. Or whenever we have reason to believe that a
person is using a verbal report to ingratiate himself, to gain respect or
prestige, to amuse or astonish, or in some other way to create a certain
social effect, we are likely to place little confidence in it. In other words,
when the circumstances in which the report occurs lead us to suppose
that the subject's motivation or the pressures to which he is exposed
are such as to prevent a candid report, we are not likely to give it much
credence.2
Not only may people be reluctant to report openly their beliefs,
feelings, motivations, plans, and so on; they may be unable to do so.
As psychoanalysts have pointed out, we are not aware of many of our
important beliefs and motivations, and hence cannot report them.
Moreover, self-report frequently requires sell-diagnosis. Even such
seemingly simple questions as "Are you shy with strangers?," or
"Would you rather go to a party or stay home and read a good book?"
require the individual to make a judgment about himself on the basis
of many past events. Feelings, beliefs, and motivations become appar-
1 For an excellent discussion of these issues, see the "Symposium on Opera-
tionism" in the PSYChological Review, 1945,52,241-294.
2 The research of Parry and Crossley demonstrates how the answers even to
factual questions may be linfluenced by the desire to appear "respectable." As sum-
marized by Katz (1951 H "They found that people consistently exaggerated their
registration and voting behavior. The exaggeration varied from thirteen percent
who falsely claimed to nave voted in the 1948 election to twenty-eight percent
who made fictitious claims to voting in local elections. One third of those who
reported contributing to the Community Chest were speaking of pious intentions,
not actual contributions. jTelephone and home ownership were accurately reported.
Similarly, car ownership 'was not appreciably inflated, but ten percent of those
reporting a driver's license did not have one. Again, ten percent claimed to have
library cards when in act'they had none."
238 DATA COLLECTION: II
ent to the self in an intellectually comprehensible form only as the

end result of an involved process of inference. With respect to com.
plex social attitudes, many people have never learned to make the
inferences necessary to an adequate verbal report; they cannot indicate,
in any systematic or analytic manner, their attitude toward their hus-
band or wife, for example, or toward "progressive education," or toward
a minority group.
Nevertheless, every person has a unique opportunity to observe
himself. To the extent that he can and will communicate his knowl-
edge about himself, he provides the investigator with information that
could otherwise be obtained, if at all, only by more time-consuming
methods. Despite the limitations of self-report, it is frequently both
possible and useful to get an individual's own account of his feelings
toward a psychological object, his image of the object, his views of ap-
propriate behavior toward it, etc.
Comparison of InterVIew and Q~estionnaire
Although both interviews and questionnaires place heavy reliance

upon the validity of verbal reports, there are important differences be-
tween the two methods. In a questionnaire, the information one ob
tains is limited to the written responses of subjects to prearranged
questions. In an interview, since the interviewer and the person inter-
vjewed are both present as the questions are asked and answered, there
is opportunity for greater flexibility in eliciting information; in addi
tion, the interviewer has the opportunity to observe both the subject
and the total situation to which he is responding. Let us detail further
some of the general characteristics of the questionnaire and interview
approaches, with their respective advantages and disadvantages.
ADVANTAGES OF QUESTIONNAIRES
By its very nature, the questionnaire is likely to be a less expensive

prpcedure than the interview. It requires much less skill to administer
than an interview; in fact, questionnaires are oft~n simply mailed or
handed to respondents with a minimum of explanation. Further, ques-
INTERVIEW AND QUESTIONNAIRE 239
tlOnnaires can often be administered to large numbers of individuals
simultaneously; an interview, on the other hand, usually calls for ques-
tioning each individual separate1y.8 Questionnaires can be sent through
the mail; interviewers cannot.4 With a given amount of funds, it is
usually possible to cover a wider area and to obtain information from
more people by means of questionnaires than by personally interview-
ing each respondent.
The impersonal nature of a questionnaire-its standardized word-
ing, its standardized order of questions, its standardized instructions
for recording responses-ensures some uniformity from one measure-
ment situation to another.5 From a psychological point of view, how-
ever, this uniformity may be more apparent than real; a question with
standard wording may have diverse meanings for different people, may
be comprehensible to some and incomprehensible to others. Neverthe-
less, much can be done to ensure meaningful uniformity of questions
by careful pretesting and by helping the subjects to understand the
questionnaire during its administration. The interviewing situation, on
the other hand, is rarely uniform from one interview to the next. Not
only do the personalities of different interviewers affect the measure-
ment situation differently; each interviewer is bound to vary somewhat
from interview to interview. Moreover, in some types of interview, the
/
3 This is not true, of course, of group interviews, where as many as eight to

ten people may discuss the subject matter of an investigation under the direction
of an interviewer. However, such interviews are more satisfactory as a source of
hypotheses or as a way of gathering information about the group; they do not
ordinarily yield systematic information from every individual in the group on each
point covered in the interview schedule.
4 Interviewing by telephone may, in certain circumstances, cost less per return
than the mail questionnaire. The telephone interview is particularly useful in
obtaining information about what an individual or a family is doing (e.g., what
television program he is watching) at the time of the call. Usually, telephone
in~ervie\Ving has to be brief and superficial to obtain the cooperation of the re-
spondent. Another serious limitation is that telephone surveys cannot. reach a
random sample of the total population, since not all people have telephones, and
people who work away from hO,me are hard to reach by telephone. For a further
discussion of telephone surveys s~e Blankenship (1946) and Parten (1950).
i) This is not true for mailed questionnaires, where there is likely to be considp!-
able variation from home to, home in the conditions under which the question.
naire is filled out. In one home:, for example, the questionnaire may be filled out
by the head of the family, in 3I':other by some other member; in one the question.
naire may be given time and attention, in another it may be competing with a
television broadcast or a crying baby.
interviewer has no standard set of questions to ask. As a result, inter-

views may be less easily comparable with one another than question-
naires.
Another advantage of questionnaires is that respondents may have
greater confidence in their anonymity, and thus feel freer to express
views they fear might be disapproved of or might get them into trouble.
Although an interviewer may assure the respondent that he will not be
identified in any way, the respondent may doubt his good faith; since,
in most interviewing situations, the interviewer knows either the re-
spondent's name or his address or other identifying information, it is
always possible that he may include this information in the completed
interview. If a questionnaire is presented as anonymous and there is no
apparent identifying information, the respondent may feel greater con-
fidence that his replies will not (or cannot) be identified as coming
from him. Studies that have used both methods have sometimes found
rather marked differences behveen the replies to the interview and
those to the questionnaire. Edwards (1957a), for example, in a survey
of attitudes of residents of Seattle toward a proposed state bill pro-
viding a cash bonus to war veterans, had interviewers question half
of the respondents; the other half were given a sheet marked "Secret
Ballot," which they checked, folded, and inserted into a box labeled
"Secret Ballot Box." (This was, of course, an unusually vivid way of
emphasizing anonymity.) The interviews showed many more "don't
know" replies and fewer unfavorable responses than the ballots. A
referendum vote in an election held a few weeks later provided a check;
the proportions obtained from the secret ballot were much closer to
those of the actual vote than the results of the direct interview. It
should be noted, however, that anonymity is not always the best
method of inducing frank answers. On more complex questions, where
there may be strong emotional involvement-as, for example, questions
of marital adjustment-an understanding and permissive manner on
the part of an interviewer is likely to be more successful than the
anonymity of a questionnaire in eliciting frank responses.
Another characteristic of the questionnaire that is sometimes,
though not always" desirable is that it may place less pressure on the
subject for immediate response. When the subje~t is given ample time
for filling out the questionnaire, he can consider ~ch point carefully
rather than replying with the first thought that comes to mind, as often
happens under the social pressure of long silences in an interview.
ADVANTAGES OF INTERVIEWS
It has been estimated that, for purposes of filling out even simple
written questionnaires, at least 10 per cent of the adult population of
the United States is illiterate. For complex questionnaires, the per-
centage would undoubtedly be considerably higher. 6 Thus, one of the
major drawbacks of the usual questionnaire is that it is appropriate only
for subjects with a considerable amount of education. Complicated
questionnaires requiring extended written responses can be used with
only a very small percentage of the population. Even many college
graduates have little facility for writing, and of those who do, few have
the patience or motivation to write as funy as they might speak. Hence,
" questionnaires are not an appropriate method for large segments of
the population; for those for whom they are appropriate, the burden
of writing or of maintaining interest is great enough to limit the
number of questions that may be asked and the ful1ness of the
responses. On the other hand, interviews can be used with almost aU
segments of the population; in fact, in contrast with ;he questionnaire,
a frequent problem in interviewing is that of limiting the responses of
the verbose individua1.
Surveys conducted by personal interviews have an additional
advantage over surveys conducted by mailed questionnaires in that
they' usually yield a much better sample of the general population.
Many people are willing and able to cooperate in a study when all they
have to do is talk. When questionnaires are mailed to a random sample
of the population, the p,roportion of returns is usually low, varying from
about 10 to 50 per cent There are many factors that-influence the per-
centage of returns to' a mailed questionnaire: Among the most
important are: (1) the sponsorship of the questionnaire; (2) the attrac-
tiveness of the questionnaire format; (3) the length of the question-
naire; (4) the nature qf the accompanying letter requesting coopera-
6 Subjects with limited education may be able to fill out questionnaires with
the help of questionnaire a,dministrators. However, in such cases the questionnaire
loses much of its advantage over the interview with respect to economy.
tion; (5) the ease of filling out the questionna.ire anci mailing it back;
(6) the inducements offered to reply; (7) the nature of the people to
whom the questionnaire is sent. Attractively designed questionnaires
that are short, easy to fill out, simple to return, sponsored by a group
with prestige, and presented in a context that motivates the respondent
to cooperate are most likely to be returned. However, even under the
best of circumstances a sizable proportion do not return questionnaires.
The people who do return them are usually the less mobile (and thus
the more likely actually to receive the questionnaire), the more in-
terested, the more literate, and the more partisan section of the
population.7
Another advantage of the interview is its greater flexibility. In a
questionnaire, if the subject misinterprets a question or records his
responses in a bafHing manner, there is usually little that can be done
to remedy the situation. In an interview there' is the possibility of
repeating or rephrasing questions to make sure that they are under-
stood or of asking further questions in order to clarify the meaning of
a response. Its flexibility makes the interview a _f~r superior technique
for the exploration of areas where there is little basis for knowing either
what questions to ask or how to formulate them.
In addition, the interviewing situation offers a better opportunity
than the questionnaire to appraise validity of reports. The interviewer
is in a position to observe not only what the respondent says but also
how he,. ~ays it. He can, if he wishes, follow. up contradictory state-
ments. If heed be, the interviewer can directly challenge-the subject's
report-in order to see how consistent his, answers will be.
The interview is the more appropriate technique for revealing in-
formation about complex, emotionally laden subjects or for probing
the sentiments that may underlie an expressed, opinion. If a verbal
:r~port is to be accepted at face value,' it must be elicited in circum-
stances that encourage the greatest possible freedom and honesty of
expression. Although, as already noted, an anonymous questionnaire
may sometimes be the most effective way of producing such a permis-
sive atmosphere, its usefulness is limited to issues on which respondents
have rather clearly formulated views that can be simply expressed. The
, more or less rigid structure of questionnaires, the fnability to explain
7 For a fuller discussion, see Parten (1950, Chapter 11).
fully in writing one's asocial or antisocial feelings and behavior, and
the solemnity and permanent nature of a response that is put on paper
in one's own handwriting or (if the questionnaire is not anonymous)
under one's own name-all work against frank discussions of socially
taboo or socially controversial issues in response to a questionnaire.
With respect to manY""l.uestions, an interview is likely to be more
successful in creating an atmosphere that allows the respondent to
express feelings or to report behaviors that are customarily disap-
proved. 8
In the interview situation, the "social atmosphere" can be varied
in other ways. Behavior il1 real life occurs in situations that are seldom
free from social pressures. The interview, more than the questionnaire,
allows one to approximate in the measurement situation these varying
social pressures, since the interviewer can, within limits, vary the nature
of the atmosphere as he questions the respondent. He can, for example,
point out objections to the position of the person being interviewed,
and observe how the latter responds. 9 This is a very useful flexibility,
especially if the ultimate objective of the measurement is to predict
behavior in varied situations.
Question Content
In both questionnaires and interviews, information is obtained

by asking questions. Questioning is particularly suited to obtaining
information "about what a person knows, believes or expects, feels or
wants, intends or does or has done, and about his explanations or
reasons for any of the preceding. These major types of question content
are discussed below. It should be recognized, however, that questions
I
do not always fall n~atly into one or another "content type." The
distinctions among types are a matter of custom and convenience
rather than theoretical rigor.
I
8 For an interesting discussion of the flexible use of the interview method to
obtain information aboutJ a socially taboo subject, see Kinsey et al. (1948).
9 As a rule, this w01/ld constitute very bad interviewing practice. However, in
interviews designed to measure responses under varying degrees of social pressure,
such procedures on the part of the interviewer are an essential part of the measure-
ment process.
CONTENT AIMED MAINLY AT ASCERTAINING "FACTS"
Often the simplest and most economical method of obtaining

"facts" is to go directly to the people who are in a position to know
them and to ask for the desired information. It is reasonable to assume
that people who have access to information, who are sufficiently intelli-
gent to absorb it, and who are motivated to acquire and retain it are
able, if they are willing, to provide the investigator with reports of
many interesting and valuable "facts." We may expect, for example,
that the people who are responsible for the execution of a policy know
what it is. Similarly, we should expect every individual to know a
variety of facts about himself and his environment. A sizeable pro-
portion of most questionnaires and interviews is directed toward obtain-
ing such facts. Questions about the person's age, education, religion,
income, nationality, marital status, occupation, etc., are of this type.
So, too, are questions about the characteristics (behavior, beliefs,
feelings, desires, intentions, etc.) of persons who are known to the
respondent, such as family, friends, and colleagues. Questions about
events, circumstances, policies, etc., known to the respondent are also
of this nature.
Reported "facts" must, of course, always be evaluated in terms of
credibility. The rules of evidence that have developed through the
centuries in judicial procedure are a good source of insight into factors
affecting credibility. Thus, it is always pertinent to raise such questions
as: How did the respondent obtain knowledge of the "fact"-
through direct observation, through inference, through !learsay, etc.?
What motives may the respondent have in reporting the "fact"? HoW
accurate is the respondent's memory of the "fact" likely to be? Any
one of these factors may affect, for example, even such an apparently
clear-cut piece of information as a hpusewife's report of her husband's
income. Her ~nowledge may be based only on her husband's state-
ment, and he may be deceiving her; or she may wish to impress the
investigator and therefore claim a higher income; or she may not be
interested enough in her husband's income to remember it pre~isely.
The
, fallibility of memory
, for nonrecurring events, for events in the
distant past, for events of little interest. and oor events difficult to
comprehend; the ephemeral quality of memory a,nd its dependence
QUESTION CONTENT 245
on situational factors; the corruptibility of memory in relation to events
of significance to the self-all of these factors require caution in accept-
ing as true the remembrance of things past (see McGeoch, 1942).
When the focus is on description or understanding of an event, a
situation, or a community, rather than on information about the in-
dividual respondent, it is frequently possible to check accuracy through
comparison of the reports of several respondents. If respondents
occupying widely different positions in the community agree on a
statement, there is much better ground for accepting it as true than if
only one of these respondents makes the statement. On the other
hand, contradictions between the reports of apparently reliable in-
formants provide important leads for further investigation. Rapkin,
Grier, and Grier (1957) used this method of checking in a survey of
race relations in an urban community. The study was exploratory in
nature, the purpose being to identify problems on which research was
needed as a basis for policy decisions. The investigators interviewed a
variety of individuals in the community: city officials, representatives
of Negro organizations, white persons known to be concerned with
problems of interracial relations, other community leaders. The
respondents were asked about such matters as the extent of employ-
ment of Negroes, police treatment of minority groups, etc. On some
points the reports were almost unanimous; on others there was wide
discrepancy. Almost all informants, for example, agreed in their
accounts of the policy and practices of the Public Housing Authority
with respect to racial integration, though there was some disagreement
about the wisdom of that policy. This impressive agreement led the
investigators to conclude that there was no need to study further the
nature of the Housing Authority's policy and practices, though there
might well be need for research directed toward finding ways of solv-
ing the problems the agency was encountering in carrying out its
policy. At the other ~xtreme, reports about police treatment of Negroes
were contradictory; they ranged from the statement that Negroes were
arrested for even the most minor infractions, which would go unno-
ticed if committed py whites, to the view that law enforcement in
Negro neighborhoo~s was so lax that only the most serious offenses
brought police action. On the basis of this striking lack of agreement
about the actual state of affairs, investigation of the practices of the
Police Department with respect to Negroes was included in the

recommendation of research priorities.
Often, however, contradictions can be cleared up within a given
study. Whenever possible, statements should be checked against
statistical records. Comparison of contradictory statements may pro-
vide a clue to the accuracy of one of them. For example, Rapkin and
his associates were told by a representative of a minority group organi-
zation that the local breweries did not employ Negroes. Checking
with the appropriate union brought the explanation that no new
brewery personnel of any kind had been hired within the past ten or
twelve years; employment had been at a peak during World War II,
and when additional workers were needed in the subsequent period,
former employees were called back. Thus, if there was discrimination
against Negroes, it was a reflection of a situation that had existed some
years before, not necessarily of present policies. Several other in-
formants supported the union's version; the person who had originally
made the statement about discrimination,. when presented with the
union's explanation,'agreed that it was correct~
CONTENT AIMED MAINLY AT ASCERTAINING BELIEFS

ABOUT WHAT THE FACTS ARE
Instead of asking questions to find out the objective facts from

people in a position to know, the investigator may wish to learn what
people believe Ito be the "facts." This is the purpose, for example, in
asking a respondent to indicate w~ether the followint-statement is
true or false: "No Negro has ever made a worth-while contribution to
the arts." The respondent's answer is not used to establish what is
objectively true but rather to provide a picture of his beliefs. Fre-
quently, before asking questions about the nature of a person's beliefs,
it is desirable to find out whether he has- any beliefs or information
relevant to the topic under investigation. For example, it makes little
sense to inquire, except "projectively," abo'ut a person's beliefs about
the United Nations if he does not know that the organization exists.
The distinction between an inquiry into "facts" and an inquiry
into beliefs must be kept clearly in mind by tHe investigator. If, for
example, he wishes to know objectively how much delinquency there
is in a community, he will interview people who have been carefully
selected for their knowledge about the topic-members of the police
force, for example, workers in settlement houses or recreation centers,
group workers assigned to street gangs (if there is such a program in
the community)-and will consult court, police, and school records,
etc. On the other hand, if he wishes to know what people believe about
the extent of delinquency in the community, he will interview a general
sample of the population rather than selected "experts," and he may
simply ask about their beliefs rather than pressing for objective evi-
dence. To use ~ simple analogy, one does not measure the temperature
of a room by asking the people in it how hot they believe it is. But if
one is interested in the subjective experience of temperature under
varying conditions, one may follow precisely this procedure of asking
people how hot they believe the room is. In the field of social attitudes,
the relationship between objective reality and a person's beliefs is
frequently of considerable interest. Distortions in perception and
beliefs, as well as gaps in knowledge, are very often clues to a person's
desires or fears.lO
In addition to discovering the content ota person's beliefs, a ques-
tionnaire or interview may provide information about the character-
istics and interrelationships or structure of the beliefs. In other words,
the purpose of questioning may be to investigate such/
aspects of belief
as: What are the person's most important beliefs? Which beliefs are
relatively private, and which are accessible to the public? How specific
is a given belief? How clear is it? How strongly held?l1
CONTENT AIMED MAINLY AT ASCERTAINING FEELINGS
A person's beliefs about what the facts are will often give very
clear indications of :his feelings and his desires. The converse is also
I
true; an emotional reaction will sometimes reveal beliefs that a subject
is unable to
.
verbalize.
I
To understand a person's behavior, knowledge
of his f~elings may ~be at least as fruitful as knowledge of his beliefs.
In questionnaires, perhaps the most common method of investi-
10 For a discussiop of techniques that employ distortions in perception and
memory a~ a method oflmeasuring social attitudes, see Chapter 8.
U For a discussiori of the measurement of the various characteristics of beliefs,
see Krech and Crutchfield (1948). .
gating feelings is to include items that bear directly on various possible

emotional reactions-fear, distrust, disgust, contempt, hate, envy,
sympathy, admiration, etc. Some examples follow: 12
When prize fights are held bern.:een Negroes and white men,
I want the Negro to win.
The sight of a Negro almost always frightens me.
Feelings and motives are probably investigated better, however,
by questions that allow the subject considerable freedom in response.
Emotional reactions are frequently too complex to report in a single
phrase. Moreover, the words used to identify an emotional reaction
may not have the same meaning for the investigator and fOf the
respondent.
Questions that call for simple unitary responses assume that a
subject reacts with the same emotion toward members of a given group
regardless of the specific situation or of his relationship with specific
members of the group. Clearly, this is a dubious assumption. For
example, a highly prejudiced plantation owner in the United States is
likely to feel somewhat benign toward Negroes who are self-abasing and
"Uncle Tom-ish" but hostile toward Negroes who refuse to consider
themselves his inferiors. An investigation of emotional reactions, if it
is to provide a full picture, must uncover not only the individual's feel-
ings but also the circumstances in which the feelings are likely to be
aroused. Both can be studied most concretely by linking them to
specific events in the subject's past. Thus, instead of asking, "How do
you feel about walking through Negro neighborhoods alone at night?,"
one would ask, "Did you ever walk through a Negro -neighborhood
alone at night? (If yes) How did you feel?"13 Of course, if the investi-
gator is interested in the subject's response to an abstract "Negro
neighborhood," the former phrasing may be preferable.
CONTENT AIMED MAINLY AT DISCOVERING

STANDARDS OF ACTION
An individual's definitions of appropriate behavior in v!lrious

social situations are of interest both as a reflection of the prevailing
I
12 For further illustrations, see Kramer (1949).
13 For further illustrations. see Kramer (1949).
climate of opinion and as a basis for predicting his probable behavior
in such situations. Definitions of appropriate action frequently have
two components: ethical sta~dards of what should be done, and prac-
tical considerations of what it is feasible to do. Questions may be
directed toward either of these components.
The following are examples of questions focused on what should
be done (Kramer, 1949):
For years College, a privately endowed
college in New England, had followed the policy of admitting
Jewish students in n-qmbers no greater than eight percent of the
Freshman class. When' a new president took office' he dropped
this policy and proposed to admit Jews on the same basis as other
students. Do you approve or disapprove of the action of the new
president? Why?
In your opinion should Negro and white children go to the

same public schools, or to separate ones? Why do you think this?
The "should" or "ought to" question-whether phrased in terms
of "I," "he," "we," or "they"-provides an indication of the idealized
policies of the individual, of the actions he would favor in a situation
free from all but moral imperatives. These policies are, for the most
part, the product of the idealistic social expectations to which the
individual has been exposed in the teachings of societal surrogates-
parents" clergy, teachers, gove~m?ent spokesmen. When there, is a
large discrepancy b,etween social ideals and practices, the ideals, for
most adults, tend to function as a guide to what to sayan formal
occasions rather th,\n what to do in everyday behavior (see Lee, 1949);
they may also be th,e source of "twinges of conscience" with respect to
everyday behavior. t
A person's behavior toward any person or group is determined not
only by his beliefs, Jeelings, and social conscieIice but also by what he
considers to be feas'ible behavior in the existing social situation. Thus,
in parts of the United States or of South Africa, an unprejudiced person
iIi the face of existing social pressures against being friendly with
Negroes may act iIi, a manner resembling that of a prejudiced person.
Conformity needs,' as well as beliefs and feelings about the individual
or group that is the ostensible "object" of the behavior, are likely to
be important determinants of action. Thus it is useful to inquire not

only into beliefs and feelings about what is "right" but also into the
"realistic" policies that serve to guide the individual's actions ~n
specific situations.
For example, Stouffer (1949), in a study of conflicting social
norms, asked students to imagine that they were proctoring an ex-
amination and saw a fellow student cheating. A questionnaire listed
possible steps the proctor might take, ranging from "Take away his
notes and exam book, dismiss him and report him for cheating," to
"Act as if nothing had happened and not report him for cheating."
Each student was asked to indicate which of these actions he, as
proctor, would be most likely to take, next most likely, etc., under a
number of specific conditions: If he did not know the student who
was cheating; if the student were a good friend of his; if the authorities
but not his student friends were likely to hear about his action; if his
student friends but not the authoritit;s were likely to hear about his
action.
The "would" question involves a persona1 prediction with respect
.to one's behavior in a given situation. Unless the respondent has been
in a comparable situation at some time, he may have little basis for
making such a' prediction. His response in this case may be no more
than, an expression of his desires or of his moral standards. Behavior
in alr~a]clife;situation is influenced by many momentary social pressures
thaJ are dIfficult to imagine unless they have been previously experi-
enced'. p'ersonal prediction by a respondent may thus be~a hazardous
affair unless it is solidly based on past behavior.
, As a matter of fact, even when a respondent has in the past been
faced with a situation of the type described, his reply about what he
would do may be at variance with, his actual past behavior or his
probable future behavior. The classic investigation of the discrepancy
between actual behavior and response to a question of the "what would
you do" type is that of LaPiere (1934). LaPiere traveled throughout
the United States with a young Chinese couple, stopping at 66 hotels,
auto camps, and tourist homes, and eating in 184 restaurants and cafes.
Only once were they refused service. Six month~ later, LaPiere sent a
questionnaire to all of the places at which they had eaten or slept,
asking, "Will you accept members of the Chinese race as guests in
your establishment?" Responses were received from half of the estab
lishments; of these, over 90 per cent said "No."
On the other hand, Pace (1939) found a high correspondence
between answers to a series of "what would you do" questions on
social, economic, and political issues, and actual behavior as indicated
by voting and group membership. He presented college students with
37 questions, of which an example is given here:
Your state needs an additional tax so that its budget can be
balanced. If you could vote, and a tax bill which provided for a
widely distributed general retail sales tax was submitted to the
people of the State for approval, what would you do?
_ _ notvoteatall
_ _ vote for the bill
_ _ vote against the bill
_ _ vote' for the bill and try to persuade others to vote for
it too
_ _ vote against the bill and try to persuade others to vote
against it too
In a pretest, Pace administered this questionnaire to 25 known
"radicals" and 25 known "conservatives." The "radicals" included
members of the Young Communist League, Trotskyites, Farmer-
Laborites, liberal New Dealers; they had all been seen frequently at
partisan meetings and at addresses by such speal{ers as Earl Browder
and Norman Thomas; they had all voted Communist, Socialist, or
Farmer-Labor in 1936. None of the "conservatives" had ever been
seen by the investigator at such meetings; 23 of them had voted for
Landon in 1936. Scores on the questionnaire distinguished between
the two groups with no overlap. Moreover, on all but three of the
37 items, the mean scores of the two groups differed significantly.
Interesting differences have been found between answers to
"should" and "would" questions. In a study by Blankenship (1946),
at a time when an ajnendment to permit horse racing and parimutuel
betting was under 90nsideration, alternative question wordings were
used with two equivalent samples. One set of respondents was asked:
"Is it desirable to permit or to prohibit horse racing and parimutuel
betting in New Jer~ey?" The other was asked: "Would you vote for
or against the amendment to permit horse racing and parimutuel
betting in New Jersey?" The first received more "don't know" and
more opposed answers. This suggests that the "is it desirable" form was
answered from a social or moral point of view, whereas the "would you
vote" form was answered in terms of personal preference.
In a study of reactions to prejudiced remarks (Selltiz et al., 1950),
subjects were shown a skit representing an informal situation in which
an anti-semitic remark was made before a group of people. In inter-
views following the skit, the subjects were asked a series of three
questions: "What do you think is the right thing to do or say? What
do you think you yourself really would have done in a situation like
this? What do you think most people would do or say in a situation
like this?" More than half (56 per cent) of the respondents replied
that the right thing would be to answer the anti-semitic remark in
some way (that is, verbally express disagreement with it); only 35 per
cent said they themseTves would have answered the remark; and only
15 per cent said most people would answer the remark.14
CONTENT AIMED MAINLY AT PRESENT OR PAST BEHAVIOR
The present or past behavior of any person is a type of "fact" that

he himself is in a uniquely favorable position to observe. We single
out this type of "fact" for special notice because of the value of knowl-
edge of past and present behavior in predicting future behavior. How
a person has behaved in the past in a certain type of situation is, in the
absence of contradictory evidence, an indication of what his future
behavior will be in similar situations.' "
'In asking about present or past behavior, experience has demon-
'strated that the most valid answers are obtained by specific rather
'than general questions. For exaJ;Ilple, it is preferable, in a study of
consumer behavior, to ask, "Which brand(s) of coffee do you have in
the house at the moment? May I see it? Do you usually buy this
brand?" than to ask, "Which brand of coffee do you usually use?"
Specifying a concrete instance and then asking whether this instance
is typical or atypical provides the subject with more cues for recall
14 For further discussion of questions that ask direhl y about the respondent's
own reactions compared with those asking for estimates of ,othrrr people's reactions,
see Chapter 8, pages 290-292.
and, in a sense, binds him to a reality that acts as an obstacle to distO'i-
tion in response. Similarly, in the study of prejudice, questions about
past behavior in specific situations are likely to elicit a more accurate
report than general inquiries about previous behavior. Thus it is
preferable to ask, "For whom did you vote in the last mayoralty elec-
tion? What made you vote for him? Did you know the religion of the
candidates? Were you influenced for or against any candidate by
knowledge of his religion? Do you usually tend to consider a candidate's
religion in deciding for whom to vote?" than to ask only the general
question, "Do you usually tend to consider a candidate's religion in
deciding for whom to vote?"
~MEn MAINLY AT CONSCIOUS REASONS FOR BELIEFS,

I3:&Il1:lNGS, POLICIES, OR BEHAVIOR
---.-~
Finally, questions may be designed to obtain the reasons the

respondent is able to offer for his beliefs, feelings, policies, or behavior.
In effect, the investigator is interested in finding out why. "Why?" may
seem like a simple question, but, as Lazarsfeld (1935) has pointed out,
the answer to it is seldom simple. Consider merely the problem of
determining why one student selects one college, another a different
one. A full answer to the question would require knowledge of their
information about various colleges, of their needs and interests, of their
ability to meet the entrance requirements or the financial cost of dif-
ferent colleges, and perhaps of still other factors.
,In order to secure a full answer to the question "Why?" it is well
for the investigator to consider the various possible factors that may
influence the belief, behavior,
I
etc., in which he is interested, and to
provide for consideration of each factor by asking a number of specific
questions rather than alsingle "Why?" Although the specific influences
that are relevant depeQd on the question being studied, it is possible
to identify certain broad classes of considerations that are likely to enter
into the determination: of why. They include: (1) The history of the
act or feeling: e.g., ':What were the circumstances when you started
to . . .?" (2) The cha~acteristics in a given entity that provoke a given
reaction: e.g., "What is there about that leads you to (feel,
believe, act, etc., in a 'given way)?" (3) The supports for the beliefs,
feelings, etc., about a given entity: e.g., "What do your friends, rela-
tives, clubs, etc. (feel, believe, etc.) about ?" "What evidence
is there to support your beliefs, feelings, etc., about ___ ?" (4) The
personal desires, motives, values, or interests involved in a given reac-
tion: e.g., "Is there anything about yourself that makes you want to
(believe, feel, or act in a given way)?" (5) The specific situations and
circumstances in which a given reaction occurs: e.g., "In what types of
situation are you most likely to (feel or act in a given way)?"
In addition to these reasons tor a given belief, feeling, action, etc.,
it may be relevant to inquire into the reasons against alternative beliefs,
actions, etc. It may also be important to distinguish between past and
present influences; for exampl~, between reasons for starting on a given
course of behavior and reasons for continuing it.
Once the investigator has decided which kinds of influences are
likely to be relevant to his particular question, he sets up an "account-
ing scheme" (see Zeisel, 1957), mapping out, in preliminary fashion,
the various kinds of reasons in which he is interested and providing
questions to tap each of them. The following illustration of a set of
questions to serve as a guide in an interview aimed at learning why an
individual selected a particular college is ad~pted from ZeiseI. Note
that it starts with the general question, "Why?," permitting the investi-
gator to find out what is salient in the mind of the respondent, and then
provides specific questions to cover the history of the choice and the
kinds of influences in which the investigator is interested but which
the respondent may not have mentioned in his spontaneous reply:
1. Why, when planning your college years, did you decide to go

to college?
2. (Supplementary questions clarifying the history of the deci~
sion. Ask only those that were 'not answered in 1.)
a. When did you first seriously consider going to _ __
college?
(Probe for answer in terms of drcumstances rather than
dates.) .
b. Did you seriously consider any other colleges you might
go to? I
c. How did you come to your decision?

3. (Supplementary questions about reasons for the choice. Ask
only those that were not answered in 1 or 2.)
a. Was there anything about the college itself (and/or the
alternatives) that influenced your decision? What? How
did you learn about these particular qualities of the college?
b. Did you feel that any of your own particular needs would
be particularly well taken care of by this choice? (Needs
while at college? Postgraduate needs?)
c. Did your parents, friends, teachers or any other persons
help you to come to your decision? In what way? Did your
own decision in any way depend on the decisions made by
your frie~ds? Did you work it out together?
4. In the light of your actual experience would you now make a
different choice? Would you make the same choice on a differ-
entbasis?
Types of Interviews and Questionnaires

The form of interviews and questionnaires may vary widely. Inter-
views may range all the way from the rigidly standardized, in which
both the questions and the alternative responses permitted the subject
are predetermined, to the completely un5tructu~ed, in which neither
the questions to be asked nor the responses permitted the subject arC'
determined before the interview. Although the possible range of ques-
tionnaire structure is more limited, there too some variation is possible
STANDARDIZED INTERVIEWS AND QU~TIONNAIRES
tervie or questionnaire, questions are

presented with Ixac~ same wording, and in the same order, to
all respondents.'ifB'freason for standardization, of course, is to ensure
that all respondents are replying to the same question. If one inter-
viewer asks, "Would you like to see taxes reduced next year?" and
another asks, "Do you think a tax reduction next year would be
desirable?," the answers
t.
may not be comparable. Differences in ques-
tion order can also ~nfluence the meaning and implications of a given
question. The ques'tion about the desirability of a tax cut might well
be answered differently if it followed a question about the need for

developing intercontinental missiles than if it followed a question
about the respondent's budget.
Standardized interviews and questionnaires may differ, however,
in the amount of structuring of the questions used. They may present
fixed alternative answers, or they may leave the respondent free to
answer in his own words.
"FIXED-ALTERNATIVE" QUESTIONS. A "fixed-alternative" (or
"closed") question is one in which the responses of the subject are
limited to stated alternatives. These alternatives may be simply Yes
or No, Dr they may provide for indicating various degrees of approval
or agreement, or they may consist of a series of replies of which the
respondent picks one as being closest to' his position. The following
are examples of fixed-alternative questions:
To. what social class would you say you belong-middle class,
lower class, working class, Dr upper class?
FDr the purpose oIour survey, we need to' have a rough indication
Df the income Df your family. WDuld you mind telling me in
which Df these classes it falls: .
Below $1,000 a year From $3,000 to $4,000
From $1,000 to $2,000 From $4,000 to $5,000
From $2,000 to $3,000 Over $5,000 a year
Put a 1 in front of the thing that is most important to have or do
in order to get ahead in the world. Put a 2 before the ~ext most
important, etc.
_ _ pull brains
_ _ good luck _ _ hard work
Do you own an automobile? _ _ Yes _ _ No
As you probably know, the Governor of Arkansas has called out
the National Guard to prevent Negro children from attending
a fDrmerly all-white high school in Little Rock. How do. you
feel about this actiDn? Would you say you:
_ _ strongly approve
_ _ mildly approve
_ _ are undecided
_ _ mildly disapprove
_-strongly disapprove
TYPES OF INTERVIEWS AND QUESTIONNAIRES 257
Questions of this type are essentially the same whether they are
used in interviews or in questionnaires. The only reason for using
interviews rather than written questionnaires with this type of material
is to reach subjects who are either not willing or not able to fill out
questionnaires.
"OPEN-ENDED" QUESTIONS. The "open-ended" question is designed
to permit a free response from the subject rather than one limited to
stated alternatives. The distinguishing characteristic of open-ended
questions is that they merely raise an issue but do not provide or
suggest any structure for the respondent's reply; the respondent is
given the opportunity to answer in his own terms and in his own
frame of reference.
Examples of open-ended interview questions follow:
Now that you have been living in for years,
I wonder if you'll tell me how you feel about it?
a. What do you like most about it?
b, What do you dislike most about it?
c. How about the neighborhood? What do you think of it?
d. How about living in a public housing project?
When used in questionnaires, both the questions and the order
in which they are presented are predetermined; it is impossible to ask
any supplementary questions. When open-ende'd questions are used
in standardized interviews, the questions and their order are pre-
determined, but the interviewer is given freedom to repeat the ques-
tion if the reply is not to the point and to use at his discretion such
nondirective probes as, "Won't you tell me more? What makes you
think ... ? Why? ~n what way ...?," etc. The task of the inter-
viewer is to encourage the respondent to talk freely and fully in response
to the questions il1cluded in the interview schedule and to make a
verbatim record of h~ replies. Generally he has no freedom to raise new
questions except to' c~~. eaning of the subject's responses, and
these must be nondirective.
(mVANTAGES AND DISADVANTA OF OPEN-ENDED AND FIXED-ALTER-
NATI~ QUESTIONS. Fixed-alternativ uestions have the advantages of
being "standardizable," simple to administer, quick and relatively in
expensive to analyze. The analysis of responses to open-ended questions
is often difficult and expensive. Categories for analysis must be built

up, coders must be trained, and the responses must be coded into one
of the categories before they can be tabulated and statistically analyzed.
Compared to the simple process of tabulating the pretoded responses
to closed questions, the analysis of open-ended questions is complex
and often troublesomeY;
From the point of view of obtaining the information needed for a
given investigation, however, each type of question has certain advan-
tages and certain disadvantages. A closed question may help to ensure
that the answers are given in a frame of reference that is relevant to
the purpose of the inquiry and in a form that is usable in the analysis.
For example, to the question, "About how often do you go to the
movies, on the average?," if no alternative answ<rt~ are supplied, one
respondent might answer, "Not very often'"; another, "When I ha~
a.date"; still another, "Only when there's something} especially-want'
to see." If the investigator is interested in frequency of movie attend-
ance, such answers are not usable. Provision of a check list with
specific estimates ("more than once a week," "about once a week,"
"about three times a month," etc.) requires the respondent to frame
his reply in terms that will be usable.
Sometimes the provision of alternative replies helps to make clear
the meaning of the question. Respondents are more likely to under-
stand the question, "Are you married, single, widowed, or divorced?"
than the question, "What is your marital status?" This function of
clarification may be important not only in relation to words whose
meaning may not be generally known but in relation to concepts that
may not be familiar to the respondent.'A study by Gross, Mason, and
McEachern (1958) of the school superintendency role provides an
example. One aspect of this study concerned superintendents' percep-
! tion of role conflicts (defined by the investigators as exposure to in-
compatible expectations on the part of different groups). Open-ended
questions failed to bring replies relevant to the investigators' concept
of "role conflict"; experimentation with different wording brought J;l0
success. Finally, they changed to a procedure of opening the interview
with descriptions of situations involving problems that all superintend-
ents face (criteria for' the hiring and promotion 4lf teachers, for ex-
15 For a further discussion of the analvsi~ of interview material, see Chapter 11.
ample), including a number of fixed alternative actions that might be
taken. The respondent was asked which of these actions various
specified people or groups (e.g., his school board) would expect him
to take. This procedure seemed to make the concept clear; the inter-
viewers could then get meaningful answers to open-ended questions
about other role-conflict situations that the superintendent had
encountered.
A similar function of alternative responses is to make clear the
dimension along which answers are sought. Consider the question,
"Are you satisfied with your present wages?" One subject may answer,
"No; I'd like to earn $100,000 a year." Another may say, "Yes, I think
the wage scale at our plant is fair; I'm earning as much as other fellows
who do the same kind of work in other places." The question involves
neither words nor concepts that are difficult. But one subject has
answered in terms of his level of aspiration (or fantasy), while the
other has answered in terms of a judgment about the equity of his pay.
To code the first respondent as dissatisfied and the second as satisfied
might be misleading; if both of them had replied along the two dimen-
sions, both might have said "No" in terms of level of aspiration but
"Yes" in terms of fairness of wage scale. More precise wording of open
ended question might eliminate this difficulty by indicating mort'
clearly which dimension was intended or by asking separately about'
both. However, the dimension along which answers are sought can
frequently be indicated more clearly by a series of alternative responses
than by the wording of the que~tion itself.
F~nally, the closed question may require the respondent himself
to make a judgment about his' attitude, rather than leaving this up to
the interviewer or coder. I This mayor may not be desirable, depend-
ing on the nature of the question. On some issues the respondent may
be in a better position to make the judgment. Suppose, in answer to the
question, "How well sat~sfied are you with your job?," a respondent
says: "Well, some things about it I like, some I don't. My boss is a nice
guy; he doesn't chew usl put for every little thing. And they're fair
about promotions and things like that. But it's awfully dirty work,
and it's a crummy old building; no decent place to eat, either." Let us
say the plan of analysts calls for categorizing attitudes in terms of the
following scale: defini~~~ more dissatisfied than satisfied,
about half and half, more satisfied than dissatisfied, definitely satisfied.
The coder may find it difficult to decide in which of the three middle
categories to place this man. The man himself, however, might have
little difficulty in making the judgment if he were presented with the
alternative positions.
Most of these advantages of fixed-alternative questions have, how-
ever, corresponding disadvantages. One of the major drawbacks of the
closed question is that it may force a statement of opinion on an issue
about which the respondent does not have any opinion. Many individ-
uals have no clearly formulated or crystallized opinions about many"
issues; this important characteristic is not likely to be revealed by a
closed question. Inclusion of a "Don't know" alternative may help to
provide an indication of a lack of crystallized opinion, but the tendency
in much interviewing with questions of this sort is to press for a def-
inite response and to accept a "Don't know" only as a last resort. Under
such pressure, the answer chosen by a respondent may be an artifact of
the specific wording or phrasing of the question or of the stated alter-
native responses. Suppose one were to ask, "Do you approve or dis-
approve of the Eisenhower Doctrine for aid to Middle Eastern coun-
tries threatened by Communist aggression?" It is easy to say "Approve"
or "Disapprove," and many respondents may find this less embarras-'
sing than admitting that they don't know what,the Eisenhower Doc-
trine is, much less have an opinion about it. In the closed question, the
reply is taken at face value. Open-ended questions, especially when
they are used in an interview and can be followed by, probes, provide
a much better indication of whether the re~pondent has any informa-
tion about the issue, whether he has a clearly formulated opinion about
it, and how strongly he feels aboutit.
Even when a respondent has a clear opinion, a, fixed-alternative
question may not give an adequate representation o,f it because none
of the choices corresponds exactly to his position, or because they
do not allow for qualification. Ta~e such a question as, "Which of the
following considerations are mos~.important to you in choosing a job?
Interesting work; opportunity to assume responsibility; pleasant sur~
roundings; congenial associates; opportunity for advancement; high
salary; s~curity. Place a 1 next to the one most important to you, 2 next
.to the one that is next most important, etc." Let us suppose that .the
items cover the range of relevant considerations for a given respondent,
and that he has a fairly clear view. But his view may involve interrela-
tions among the factors. In general, interesting work may be more im-
portant to him than a high salary: However, given a choice between
two jobs, one of which pays twice'as much as the other but is slightly
less interesting, he may choose the higher-paying. Or there may be
some lower limit of salary beyond which he feels he cannot afford to
go, no matter how interesting the work. Such qualifications can be ex-
pressed in reply to an open-ended question; a closed question not only
makes no provision for them, but even discourages the respondent
from thinking about them.
Omission of possible alternative responses may lead to bias. Even
when a space is provided for "other" replies, most respondents limit
their answers to the alternatives provided. Omission of an alternative
may seriously change the replies to even a factual question such as
what magazines people read. In a study of applicants who were accepted
by a certain college but did not actually enter, the subjects were pre-
sented with a check list of reasons for not attending. These reasons
included such factors as the location of the college, its cost, the fact
that it was not coeducational, the fact that specific desired courses were
not offered, etc. However, the possibility that the applicant had in the
end entered another college because it had a generally high,er academic
reputation was not included. Although a few respondents added this
in the space provided for "other reasons," there was no way of esti-
mating how many would have checked it had it been included among
the suggested alternatives. Unless one can be reasonably certain, on
the basis of either the logical possibilities or prior investigation, that the
alternatives presented adequately cover the complete range of prob-
able responses, it is safer to ~se an open-ended question, which does
not bias the responses by suggesting some but not others.
The fact that the wording of questions is the same for all respond-
ents may conceal the fact that different respondents make different in-
terpretatio~s, some of which may be quite different from those in-
tended by the interviewer, This possibility exists, of course, in both
closed and open questions, but I,
it is much more likely to go undetected
in the former. An instance of interpretations made from such varying
frames of reference as to make the meaning of the obtained replies
unclear has been reported by Crutchfield and Gordon (1947). A na-

tional survey used the following question: "After the war, would you
like to see many changes or reforms made in the United States, or
would you rather have the country remain pretty much the way it was
before the war?" Most of the respondents replied that they wanted
the country to remain "pretty much as it was." In a follow-up study,
the same question was asked, but then followed by probes to ascertain
what the respondents had in mind when. they answered the question.
The investigators identified seven different frames of reference: do-
mestic issues (employment conditions, standards of living, etc.); tech-
nical improvements (better transportation, communications, etc.);
political affairs; and so on. It seemed clear that no single interpretation
of the responses to the closed question was justified.
From this discussion of the relative advantages and disadvantages
of open and closed questions, it is apparent that the two differ in !he
purposes for which they are appropriate. Closed questions are more
efficient where the possible alternative replies are known, limited in
number, and clear-cut. Thus they are appr?priate for securing factual
information (age, education, home ownership, amount of rent, etc.)
and for eliciting expressions of opinion about issues on which people
hold clear opinions. Open-ended questions are called for when the
issue is complex, when the relevant dimensions are not known, or when
the interest of the research lies in the exploration of a p1;ocess or of the
individual's formulation of an issue. The closed question has the ad-
vantage of focusing the respondent's attent~on on the dimension of
the problem in which the investigator is interested; by the same token,
I
it does not provide information about the respondent's own form!Jla-
tion of the issue, the frame of reference in which he perceives it, the
factors that are salient for him, the motiyations that underlie his
opinions. \\Then these matters are the foc:us of interest, open-ended
questions are essential.
Lazarsfeld (1944) has proposed that the development of a ~losed
question interview schedule be preceded by more intensive, fre(!r inter-
views with a subsample of the population in order to discover the range
! of probable responses, the dimensions that are seen as reIe:vant, and
the various interpretations that may be made M the question wording.

On the basis of such preliminary exploration, more meaningful closed
questions can be formulated. He has also suggested another method
of using the two types of question to supplement each other: after a
survey using closed questions, more intensive interviews might be held
with a subsample in order to delve more deeply into areas that appear
significant. For many purposes, a combination of open and closed
questions is most efficient; an interview or questionnaire need not con-
sist entirely of one type or the other.
LESS STRUCTURED INTERVIEWS
For some research problems, a still more flexible approach than

that provided by a standardized interview with open-ended questions
is appropriate. Largely as a result of the influence of clinical inter-
viewing and anthropological field work, a vdried assortment of inter-
views has been developed in which neither the exact questions the
interviewer asks nor the responses the subject is permitted to make are
predetermined. Such interviews take various forms and go under var-
ious names-the "focused" interview, the "clinical" interview, the
"depth" interview, the "nondirective" interview, etc. They are com-
monly used for a more intensive study of perceptions, attitudes, moti-
vations, etc., than a staridardized interview, whether with closed or
open questions, permits. This type of interview, is inherently more
flexible, and of course it requires more skill on the part of the inter-
viewer than do the standardized types. Obviously, this approach is
impossible in a questionnaire.
The flexibility of the unstructured or partially structured inter-
view, if properly used, helps to bring out the affective and value-
laden aspects of the ,subject's responses and to determine the personal
significance of his attitudes. Not only does it permit the subject's defi-
nition of the inter~iewing situation to receive full and detailed ex-
pression; it should a1so elicit the personal and social context of beliefs
and feelings. This type of interview achieves its purpose to the extent
that the subject's r~sponses are spontaneous rather than forced, are
highly specific anc;l concrete rather than diffuse and general, are self-
reveal{ng and personal rather than superficial.
The freedom which the interviewer is permitted is, at once, both
the major advantage and major disadvantage of interviews of this type.
The flexibility frequently results in a lack of comparability of one inter-

view with another. Moreover, their analysis is more difficult and time- .
consuming than that of standardized interviews. There can be little
doubt of their usefulness, in the hands of a skilled investigator, as a
source of hypotheses that can later be submitted to a systematic test.
Partially structured interviews are also used, on occasion, in studies
testing hypotheses. However, the lack of comparability from interview
to interview and the complexity of analysis usually make them less
efficient for this purpose than standardized interviews.
Let us briefly discuss several of the major types of partially struc-
tured and unstructured interview.
In the focused interview, as described by Merton, Fiske, and Ken-
dall (1956), the main function of the interviewer is to focus attention
upon a given experience and its effects. He knows in advance what
topics, or what aspects of a question, he wishes to cover. This list 9f
topics or aspects is derived from his formulation of the research prob-
lem, from his analysis of the situation or experience in which the re-
spondent has participated, and from hypoth~ses based on psycholog-
ical or sociological theory. This list constitutes a framework of topics
to be covered, but the manner in which questions are asked and their
timing are left largely to the interviewer's discretion. He has freedom
to explore reasons and motives, to probe ~urther in directions that were
unanticipated. Although the respondent is free to express completely
his own line of thought, the direction of the interview is clearly in
the hands of the interviewer. He wants definite types of information,
and part of his task is to confine the respondent to discussion of the
issues about which he wants knowledge.
Merton, Fiske, and Kendall (1956) have described this type of
interview:
First of all, the persons interviewed are known to have been
involved in a particular situation: they have seen a film, heard 'a
ra,dio program, read a pamphlet, article or book, taken part in
a psychological experiment or in an uncontrolled, but observed,
social situation (for example, a political rally, a ritual or a riot),
I Secondly, the hypothetically significant elements, patterns, proc-
esses and total structure of this situation h~ve been provision-
ally analyzed by the social scientist. Through this c0I?-tent or sit-
uational analysis, he has arrived at a set of hypotheses concerning
the consequences of determinate aspects of the situation for
those involved in it. On the basis of this analysis, he takes the
third step of developing an interview guide, setting forth the
major areas of inquiry and the hypotheses which provide criteria
of relevance for the data to be obtained in the interview. Fourth
and finally, the interview is focused on the subjective experiences
of persons exposed to the pre-analyzed situation in an effort to
ascertain their definitions ot the situation. The array of reported
respons~s to the situation helps test hypotheses and, to the ex-
tent that it includes unanticipated responses, gives rise to fresh
hypotheses for more system~tic and rigorous investigation.
The focused interview has been used effectively in the develop-
ment of hypotheses about which aspects of a specific experience (a
radio broadcast, a moving picture, a lecture, etc.) lead to changes in
attitude on the part of those exposed to it. The interviewer, being
equipped in advance with a content analysis of the stimulus experience,
can usually distinguish the objective facts of the case from the subjec-
tive definitions of the situation. Thus, he is alerted to the possibility
of "selective perception" and prepared to explore its implications. Sup-
pose, for example, that one is concerned with reactions to a series of
newspaper pictures portraying housing conditions in a slum neighbor-
hood, intended for use in connection with a campaign for more strin-
gent housing laws or for slum clearance and urban redevelopment. The
pictures show broken stairs, wallpaper peeling off, holes in walls through
which rats are reported to enter-in general, conditions that may
reasonably be attributed to inadequate maintenance On the part of the
landlord rather than slovenliness on the part of tenants. A respondent,
in discussing the pictures, may say, "They show how these low-class
people don't take care: of their places; there's no use trying to give
them decent housing, they just knock it to pieces anyway; you know,
like they always say, if IYoU give them bathtubs they just put coal in
them." The interviewer, knowing that the content of the pictures is
not intended to give this impression, can follow up the respondent's
interpretation, trying' to discover whether there are unconsidered as-
pects of the pictures that form a basis for this impression, whether
it stems from the subjec;:t's stereotyped views, etc.
The definition of a focused interview may be broadened to in-

clude any interview in which the interviewer knows in advance what
specific aspects of an experience he wishes to have the respondent
cover in his discussion, whether or not the investigator has observed
and analyzed the specific situation in which the respondent partici-
pated. For example, in a study of the functioning of a program of part-
time work for high school students, one may prepare a set of questions
to be covered even though he is not familiar with the specific job set-
ting of each of the students. Such a list might include questions such
as the following: "Does the student feel that he was given an adequate
picture of the job before he started? Does he feel that his job is at a
level appropriate to his skills?," etc.
Obviously, the more detailed the investigator's knowledge of the
situation in which the person being interviewed has participated, and
the more specific his hypotheses, the more precisely he can outline in
advance the questions to be covered in the interview. ..
Somewhat similar to the focused interview is the clinical inter-
vie~, the primary difference being that the clinical interview is con-
cerned with broad underlying feelings o~ motivations, or with the
course of the individual's life experiences, rather than with the effects
of a specific experience. In this type of interview, too, the interviewer
knows what aspects of feeling or experience he wants the respondent to
talk about, but again the method of eliciting the information is left to
his discretion. The "personal history" interview, used in social case
work, prison administration, psychiatric clilfics, and in social research
using individual life histories, is I perhaps
.
tae most common type ,of
clinical interview. The specific aspects of the individual's life hi?tory
which the interview is to cover are determined, as in all data.-collec-
tion instruments, by the purpose for which the information is gathered.
For example, Lee (1957) was interested in the possibility that
adolescents who become heroin addicts may be predisposed to addic-
tion by family experiences that lead to certain personal characteristics.
On the basis of earlier work with juvenile addicts, he and colleagues
conducting related studies hypothesized that among boys living in the
same neighborhood and thus exposed to roughly the same opportuni-
ties for using heroin, addicts are likely to diff~r from non-addicts in the
following ways: they have relatively weak ego. functioning, defective
soperego functioning, inadequate masculine identification, lack of real-
istic middle-class orientation, and distrust of major social institutions.
Next, the investigators asked themselves what types of family envi-
ronment might be expected to stimulate or enhance such character-
istics. On the basis of theoretical considerations, largely drawn from
psychoanalytrc thinking, they constructed a list of circumstances or
events of family life that might be expected to contribute to each of
the five characteristics. For example, it was considered that factors
such as the following might be conducive to weak ego functioning:
inappropriate handling of cIiildhood illnesses, discordant relationship
between parents, the mothe/figure either passionate or hostile toward
the boy, either parent's having unrealistically high or low aspirations
for the boy, etc.
It seemed clear that relatively unstructured interviews would be
a more appropriate method of getting the needed information than
would a standardized series of questions. Accordingly, the interviewers
visited the parents of the boys included in the study-a sample of
addicts and a control group of non-addicts-and encouraged them to
talk freely about their sons. The interviewers had no set questions to
ask. They were instructed to cover the following major topics: the
physical characteristics of the neighborhood and the /
house, the com-
position of the family and the household, the health history of the
family, the present and early adolescent life situation of the subject,
childhood training and socialization, relationships within the family,
and relationships between the family and the "outside world." The
interview guide indicated a number of subtopics to be covered under
each of these major ones; for example, under "childhood training and
socialization," the in~erviewer was to get information about early de-
velopment, disciplineiand patterns of handling by parents, early social-
ization experiences, early school experiences, etc. Under each of these
subtopics, the intervi~w guide listed more specific points to be covered.
In the nondirective interview, the initiative is even more com-
pletely in the hands, of the respondent. The term nondirective received
its currency from a type of psychotherapy in which the patient is en-
couraged to express his feelings without directive suggestions or ques-
tions from the therapist. In a more limited sense, nondirection is im-
plicit in most interviewing; that is, although the interviewer is expected

to ask questions about a given topic, he is instructed not to bias or
direct the respondent to one rather than another'response. 16 In non-
directive interviewing, however, the interviewer's function is simply
to encourage the respondent to talk about a given topic with a mini-
mum of direct questioning or guidance. He encourages the respondent
to talk fully and freely by being alert to the feelings expressed in the
statements of the respondent and by warm, but noncommittal, recog-
nition of the subject's feelings. Perhaps the most typical remarks
made by the interviewer in a nondirective interview are: "You feel
that ..." or "Tell me more" or "Why?" or "Isn't that interesting?"
or, simply, "Uh huh."
The nondirective interviewer's function is .primarily to serve as a
catalyst to a comprehensive expression of the subject's feelings and
beliefs and of the frame of reference within which his feelings and
beliefs take on personal significance. To achieve this result, the inter-
viewer must create a completely permissiye atmosphere, in which the
aubject is free to express himself without fear of disapproval, admoni-
tion, or dispute, and without advice from the interviewer.17
The Sociometric Method I8

Sociometry is concerned with the soc;ial interactions among any
group of people. The data col1c:;ction is geared to obtaining informa-
tion about the interaction or lack of interaction among the members
of any group (or among subgroups, or among groups, or among sub-
16 In "stress interviews," this statement does not hold. Here, the interviewer
tries to see how well the respondent can function under the stress of baiting, dis
paragement, expressed hostility, etc. The term 'may also be applied to interviews
in which the interviewer tries to see how much pressure or stress is required to
change the respondent's expression of views.
17 For a more detailed discussion of nondirective interviewing see Roethlis-
berger and Dickson (1939) and Rogers (1945).
18 Sociometry is most closely identified with the work of J. L. Moreno. For a
discussion of its o'rigin, see his Who Shall Survive? p 953). For numerous articles,
consult the journal Sociometry. For a survey of research using sociometric tech-
niques, and a discussion of the values and shortcomings of this approach, see
Lindzey and Borgatta (1954).
THE SOCIOMETRIC METHOD 269
groupS and individual members, etc.). The interaction that is inves-
tigated may be behavioral, or it may only be desired, or anticipated, or
fantasied. The content or type of interaction studied may be anyone
of a variety of social behaviors-sitting next to, eating with, buying
from, lending to, visiting, playing with, having as a friend, talking t'?,
living next to, etc.
Sociometric studies most commonly use questionnaires, although
observational data or other kinds of records may provide the grist for
a sociometric analysis. Essentially,_ sociometry is not so much a data-
collection procedure as a focus on a certain type of subject matter and
a related method of analysis.
When interest is in actual behavior, one may observe the partic-
ipants in an action, the nature of the action, and the relationship of
the participants to one another; both the initiation and the reception
of social behavior are of interest. When interest is in desired or fan-
tasied interaction, or in feelings about interaction with specified in-
dividuals, verbal reports are called for. The sociometric questionnaire
or interview, as most commonly used, involves simply asking each
member of a group to indicate which other members he would like
to have as a companion in some activity (e.g., "eat lunch with") and
which ones he would not like to have as a companion. Sometimes the
individual is allowed to name as many members ~s he wishes; more
frequently he is limited to naming a specific number. It is assumed
that preferences are more likely to be stated honestly when the subject
believes that they will really determine subsequent social arrangements;
that is, that he will be assigned to sit next to (work with, etc.) the
individuals he has named. Therefore, when it is feasible, sociometric
questionnaires are usually given with a statement that the investigator
will arrange circumstances
I
to permit the fulfillment of the individual's
preferences if possible;
Sociometric questionnaires are easy to administer and are adapt
able to many differel1t types of setting. Studies of the reliability of
sociometric data, on the basis of repeated tests, indicate that although
there may be considerable variation in specific choices, patterns of
group interaction and'various scores or indices derived from the data
are quite stable. Despite the ease of administration, however, the
analysis of sociometric data is frequently more complex than one

anticipates.19
Sociometric data can provide information about an individual's
position in the group, the social subgroupings within the group, the
relationships among the subgroups, the group's cohesiveness, etc. Data
of this type have been useCl in studies of leadership, of relations among
ethnic groups, of the-effect of experimental treatments on group struc-
ture, of the effect of variations in group structure on the behavior of
group members, of characteristics of individuals who are frequently
chosen and those who are seldom chosen.
One of the first studies of leadership that made use of sociometric
techniques was that of Jennings (1943). Girls in a state training school
were asked to indicate which girls in the school they would want to
live with and which ones they would want to work with; for each of
these activibes, they were asked to list also which girls they would
not want as companions. A "choice score" was computed for each girl
on the basis of the selections and rejections she received; in terms of
these scores, each girl could be described as "over-chosen," "average-
chosen," or "under-chosen." It was found that choice score was closely
related to leadership in the community; of the twenty members of a
Community Council selected a few months before the sociometric
measures were taken, eighteen were "over-chosen," and the other two
were just below the "over-chosen" point. Study of the reasons given
for the sociometric choices and rejections led to the conclusion that
leadership is not explainable by any particular personality characteris~ic
or constellation of traits, but rather on the basis of the interpersonal
contribution an individual makes in a specific group. '
Festinger, Schachter, and Back (1950)" in a study of the effect of
the location of dwellings on friendship formation, used reports of
actual social behavior rather than statements of preference. Studying
a relatively self-contained community of married veteran students, they
asked each of the wives, "What three people in Westgate or Westgate
West do you see most of socially?" (Westgate consisted of 100 single-
family houses arranged in nine courts; We~tgate West consisted of
19 For a discussion of methodological problems in the analysis of sociometric
data, see Proctor and Loomis (1951).
THE SOCIOMETRIC METHOD 271
seventeen two-story buildings with five apartments on each floor.) They
found that 65 per cent of the people named by Westgate West resi-
dents lived in the same building as the chooser, 44 per cent of them on
the same floor. Moreover, people living in the immediately adjacent
apartment were named more often than those living two doors a~ay,
who in turn were named more often than those living three doors away,
etc. A similar pattern was found in the Westgate community. Thus it
was clearly demonstrated that in this community, consisting of new
residents homogeneous in age, interests, socioeconomic status, etc.,
and relatively isolated from the larger community, ecological faci:ors
were an important determinant of friendships. ,'
Among the earliest applications of sociometric techniques to the
study of relations between racial or ethnic groups were the studies of
Criswell (1937, 1939). She asked children in mixed Negro-white
classes, from kindergarten through the sixth grade in a public school,
to choose two classmates beside whom they would like to sit. She
found that, in this school and within this age range, cleavage between
the sexes was far more marked than cleavage between Negroes and
~hites. The white children did not begin to withdraw from the Negro
children until the fourth grade and did not form a "racial" group until
the fifth grade-a finding previously reported by Moreno (1934).
Although such studies are extremely usef~l in revealing)nter-
relationships among members of a given group, one must be ca~tious
about interpreting cleavages as evidence of prejudice. Preference for
members of one's racial or ethnic group may simply indicate g~ater
familiarity. For example, the white students-in a racially mixed-school
class may live in the same neighborhood and know one another well,
but t}1ey may have! little acquaintance with the Negro students, who
live in a different Ijleighborhood. If this is true, a nonrandom choice
(more white students choosing other white students than would be
~xpected by chan~e) does not necessarily indicate an avoidance
of the Negro students. This is not to deny that prejudice may be in-
volved somewherel historically, in the chain of causation-as in the
creation of the segregated neighborhoods. But this is different from
asserting that the nonrandom choice patterns 'of the children are
necessarily indications of prejudice among the children.
Visual Aids in Interviewing

Occasionally visual aids-photographs, line drawings, dolls, etc.
-are introduced into the interviewing situation. We shall discuss here
the use of such techniques in the course of direct interviewing; that is,
in situations where the visual material is used simply as a substitute
for a verbal statement or to make clear what is being' asked about, and
the respondent is asked directly for his reaction to it. Such use has been
rather limited; more frequently, visual materials are used in indirect, or
"projective," questioning. This latter use will be discussed in the follow-
ing chapter.
Perhaps the most common use of visual materials has been in
studies of racial awareness and racial attitudes of young children. The
device was first used in this way by E. L. Horowitz (1936). He pre-
sented children with pictures of groups and of individuals, and asked
questions similar to those used in many of the sociometric procedures.
One' of his tests uses a set of paired photographs depicting identical
activities and settings. One photograph in -each pair shows five white
children participating, 'whereas the matched photograph shows the
same group with a Negro boy substituted for one of the whites (for a
portion of the study, three of the five boys were Negroes). The sub-
jects are shown each picture and are asked to indicate whether they'
would care to join in the activity with the children depicted in the
picture. From their responses ("Yes," "No," "Undecided"), it is pos-
sible to' compute a "willingness to join in" ~core for the series of aU-
I .
white pictures and for the series of racially mixed pictures; the ,dif-
ference between the scores is taken as a measure of bias.
Another approach developed by Horowitz is the "faces test," .con-
sisting of twelve photographs of boys-four white, four "light" Negro,
and four "dark" Negro. In one use of these photographs (the "ranks"
test), the children are asked to indicate which one of the twelve they
like best, next best, and so on until the twelve pictures are ranked in
order. A second use is the "show me" test, in which children are
~ asked: "Show me all those you want to come to your party," etc., for
twelve different activities. A later revision of ~he "show me" test also
attempts to get at stereotypes by including such, items as: "Show me
VISUAL AIDS IN INTERVIEWING 273
which children live in a dirty house," "Show me which are most
stupid," etc. For each test, the individual is scored in terms of the
extent to which his pattern of responses differs from that which would
be expected if the racial identification of the pictures were not a factor
in his choice. For example, if choices for companions in the twelve
activities were made without regard to race, one would expect that the
total choices would include roughly equal numbers of the white boys,
the "light" Negroes, and the "dark" Negroes. A child who consistently
chose only the pictures of white boys as those he wanted to share in
activities with him would deviate markedly from the random pattern
to be expected if race were not a factor.
A number of other investigators have used the Horowitz pictures.
In one study (Mussen, 1950) an attempt was made to estimate their
validity by comparing scores on the "ranks" and "show me" tests with
actual sociometric choices of cabin-mdies in an interracial camp. Scores
on the tests correlated significantly with the proportion of choices of
white and Negro boys as cabin-mates, and changes in test scores from
beginning to end of the camp period were correlated with changes in
proportion of whites and Negroes in the sociometric choices. "-
Picture tests have been used by Helgerson (1943), by Horowitz
and Horowitz (1938), and others to discover whether race or other
characteristics, such as sex, age, socioeconomic s~atus, facial expression,
etc., are more important in determining an individual's preferences.
The technique consists of pairing pictures in which several variables
are contrasted. In a study contrasting the cffects of sex and race on
preference, for example, the individual would be presented with the
following pairs of pictures: a Negro boy and a white girl, a white girl
and a white boy, a Negro girl and a white boy, a Negro boy and a Negro
girl. With respect to, each pair of pictures, he would be asked, "Which
one would you rat~er play with?"
Picture tests qave also been employed by-R. E. Horowitz (1939)
and by Clark and lark (1939) to study young children's awareness of
their own race. Th~ technique consists mainly in showing the children
a series of pictures (varying in race, sex, age, etc.) and asking the child,
"Which one is you?" or "Which one is most like you?" or "Is this you?"
I,
White and brown dolls have been used in a similar fashion by
Clark and Clark (1950) and by Goodman (1952). In both of these
Itudies, the dolls were used not only in connection with questions
::lirected toward the child's awareness of his own racial identification
but toward his attitudes: "Which doll do you like best?," "Which doll
is prettier?," etc.
Clark and Clark (1950) used an interesting variation of a pictorial
technique to get at children's awareness of their own racial identifica-
tion and their feelings about it. They presented Negro children with
a box of crayons, including a range of shades of brown as well as the
colors usually included in children's crayon sets, and two line drawings.
The child was asked first to color one figure "the color that you are,"
then to make the other one "the color that you like little girls (boys)
to be."
The use of pictorial techniques has not been limited, however, to
studies of children or of intergroup attitudes. Murphy and Likert
(1938) made use of both photographs and motion pictures in a study
of the attitudes of college students. The photographs, borrowed from
news services, all showed conflict situations-strikes, war, race riots. In
connection with each picture, the subjects were asked to answer such
questions as: "Describe 1:?riefly in outline form your reaction to this
photograph. . . . In this situation, with whom do you sym-
pathize? . . . What do you like or dislike in. this photograph? Why?"
Three motion pictures were shown: one portraying the aftermath ofa
race riot; another; the attempt of a mob to storm a courthouse with the
intent of lynching a Negro who was in custody there; and the third,
fleet maneuvers. After seeing each film, the student was asked to write
briefly what he thought about it, and then to express his agreement or
disagreement with a number of st~tements related to it (e.g., "Riots of
this kind are a tragedy for both the white and black races") . I
Visual aids are sometimes helpful in inducing people to discuss

matters about which they may feel awkward or embarrassed when
questioned directly. Thus, Whyte (1957) reported that in a study of
the meaning of work, he had difficulty in getting respondents (skilled
glass workers) to talk about certain aspects of their work. They talked
freely about working conditions, the union, fellow workers, foremen,
, and higher management people. But wh~n he tried to get at their
reactions to mental and physical processes in\rolved in the work itself
-for example, esthetic satisfactions, feelings of creativity-they became
VISU AL AIDS IN INTERVIEWING 275
embarrassed and inarticulate. Therefore he made line drawings of a
number of different products on which the men had worked, and
asked each respondent to arrange the cards in the order of his prefer-
ence for the different jobs. Then the respondent was asked to explain
why he had ranked the cards as he did. The resulting comments re-
vealed much more about feelings toward the work process itself than
had been elicited in the earlier, purely verbal interviews.
Collier (1957) carried out a small-scale experiment to discover
the relative effectiveness of a purely verbal interview and one using
photographs. In a study of migration and acculturation of French-
Acadians to an English industrial town in one of the Canadian mari-
time provinces, he was especially interested in the areas of work, home,
and community relationships. Two respondents were each interviewed
twice by purely verbal techniques in partially structured interviews, a
third time with the addition of photographs. In interviews with two
other respondents, photographs were used throughout. The photo-
graphs were of industrial plants in the town in which the respondents
worked, houses and street scenes in the communities in which they
lived, and activities in and around the home of the particular respond-
ent; these latter pictures were taken with the cooperation of each
respondent and his family. Collier found much greater interest in the
interviews with the photographs, much more specific information
given in the interview, greater ease of keeping the interview on the
topics in which the investigator was interested, and greater ease of
judging the extent of a respondent's information about certain matters.
Whether or not visual techniques are likely to be an effective aid
in interviewing depends on the circumstances and the research purpose
of a given study. Pictorial methods are particularly useful in the study
of attitudes of children or of those with limited literacy. Verbal com-
prehension is need:ed only to the extent of understanding the general
test instructions; s'ince these can frequently be given orally, reading
ability is not required. The usefulness of pictorial methods, however,
is not restricted to respondents with limited reading ability. Pictures
are of value in depicting many types of situation that are difficult to
describe; thus they permit the uncovering of reactions that are difficult
to obtain by other methods. Another advantage is that a pictorial test,
because it usually has more inherent interest than a written question-_

naire, is likely to meet with less resistance.
Against these assets must be counterposed certain liabilities. A
picture presents a concrete situation; if one is interested in investigat-
ing general attitudes, he may find that responses are influenced by
specific details of the pictures he presents. Thus, Murphy and Likert
(1938), in the study described above, found that responses to the
pictures dealing with unions and strikes seemed to depend on the
respondent's interpretation of which side had started the violence
shown in a given picture. A related limitation is the difficulty of using
pictures to study attitudes toward groups that have no distinguishing
visual characteristics (for example, Catholics, social~sts, psychologists) .
Though it is possible to use symbols or names to identify members of
different religions, nationalities, or other groupings, the symbols may
not have a clear-cut meaning for the respondents; or the respondents
may react to the symbols rather than the groups they represent. There
is a further difficulty in studies whose design calls for comparison of
responses, to pictures or photographs that .supposedly differ only in
the raciall (or other) identification of the individuals portrayed. A
score based on deviations from random choice assumes that the
pictures, for example, of Negro and white children are of equal attra~
tiveness apart from their racial characteristics. To be sure that this
condition has been met is no easy matter if unretouched photographs
are used; on the other hand, dubbing in color on equated "white" pic-
tures may create an artificial effect.
A Concluding Note
From this survey of questionnaire a~d interview procedures, it is
apparent that'the investigator interested in individuals' self-reports has
a choice of many different ways of eliciting them. In making such
decisions as whether to use a questionnaire or an interview, whether to
use a standardized or a less structured form, and whether to supple-
mcnt the verbal material by visual aids, he need~ to consider the advan-
tages and disadvantages of each approach in the light of the, purpose
of his study.
A CONCLUDING NOTE 277
The investigator should be concerned, of course, with the reli-
ability and validity of his measures. Although it seems likely that most
investigators are concerned with these matters in the sense of hoping
that their measures are reliable and valid, this concern has not often
been expressed in attempts to determine the reliability or validity of
the instruments used. This is at least as true of interviews and question-
naires as of other types of measuring instruments-perhaps even more
so, since most interviews and questionnaires are specifically designed
for the purpose of a single study, and thus do not have the benefit of
testing by repeated use.
An occasional investigator has tested the reliability of his instru-
ments by having two different interviewers interview the same individ-
uals, or by repeating a questionnaire or interview with the same in-
dividuals after a lapse of time. To be sure, this procedure is time-con
suming, and it is not always easy to secure the cooperation of subjects
for a repetition of the same interview or questionnaire to which they
have already responded. This latter difficulty can be lessened somewhat
by changing parts of the instrument, repeating only certain selected
questions for the test of reliability.
In connection with a few of the measurement techniques de-
scribed in this chapter, the investigators reported evidence about
validation; for example, Pace's testing of his /measure of political-
economic-social attitudes on .known "liberal" and "conservative"
groups, and Mussen's comparison of scores on the "faces test" with
sociometric choices. Many other investigators have attempted to find
ways of assessing the validity of their measures. Nevertheless, it remains
true that many-probably most-questionnaires and interviews have
been used without evidence of their validity. Again, the reasons are
not hard to find. It is not always easy to determine what would be
appropriate criteri~ of validity. Even if one can identify what would
I
constitute appropriate evidence, it may not be feasible to gather the
necessary data. But without such evidence, one can only hope that his
instruments are actually measuring what he believes they are measur
ing. Thus it would seem desirable to devote more time than is usually
given to investigation of the reliability and validity of questionnaire~
and interviews, as irell as of other instruments for collecting data.
It has been found in many public opinion surveys that even slight
differences in question wording may lead to considerable differences in

the proportion of people who answer in one or another way. Obviously,
too, the behavior of an interviewer may influence responses. A good
deal is known about common sources of bias in question wording and
in interviewer behavior. Even if an investigator decides that he cannot
invest the time, effort, and expense that would be needed to test the
reliability and validity of his questionnaire or interview, he can and
should examine carefully both his questions and the instructions' to
his interviewers to make sure that he has avoided obvious sources of
bias. A more detailed discussion of the construction of questionnaires
and interviews and of interviewing procedure, from this point of view,
is given in Appendix C.
8
DATA COLLECTION
III. Projective and Other Indirect Methods
Projective Methods
.Structured Disguised Tests of Social Attitudes
I
Substitute Measures
A Note on Validation
T ECHNIQUES that rely on the individual's own report of his be-
havior, beliefs, feelings, etc., presuppose, as has already been
pointed out, that the person is willing and able to give such informa-
tion about himself. But this is not alwayspue. People may be unwilling
to discuss controversial topics or to reveal intimate information about
themselves. They may'be reluctant to express their true attitudes if they
believe that such attitudes are generally disapproved. Or they may be
unable to give the desired information, either because they cannot
easily put their feelings into words or because they are unaware of their
feelings about the matter in question.
To get around these limitations, techniques have been devised that
are largely independent of the subject's self-insight and of his willing-
ness to reveal himself. These indirect techniques may be grouped in. two
broad c1assesl _differing in their degree o! structure. The less structured
ones are commonly referred to as projective methods; among the more
structured techniques we may identify disguised methods and sub
stitute measures.
Projective Method~
-
Frank (1939), the originator of the'term, has -given the following
of
definition a projective technique:
A projectiv~ method . . . involves the presentation of a

stimulus situation, designed or chosen because it will mean to the
subject not what the experimenter has arbitrarily decided 'it
should mean ... but rather whatever it must mean to the
personality who gives it, or imposes upon it, his private, idiosyn-
cratic meaning and organization.
The as'sumption is made that thej individual's organization of

the relatively unstructured stimulus situation is indicative of basic
trends in his perception of the world and in his response to it.
280
PROJECTIVE METHODS 281
PROJECTIVE METHODS IN THE STUDY OF PERSONALITY
Projective. methods were first devised by psychologists and

psychiatrists concerned with the diagnosis and treatment of patients
suffering from emotional disorders, and this has continued to be their
major use. Tests for this purpose attempt to give a comprehensive
picture of the individual's personality structure, his emotional needs,
his conflicts, etc. To discuss either the theory underlying these tests or
the methods of interpreting them is beyond the scope of this book.
They are primarily tools of the clinical psychologist rather than the
social psychologist or sociologist; moreover, their use requires intensive
specialized training. However, in view of their great usefulness in the
investigation of certain types of problems in social psychology, soci-
ology, and anthropology, we shall briefly describe their general char-
acteristics, mention some of the more frequently used tests, and in-
dicate the kinds of social relations research in which they have been
used.!
It wil1 be sufficient, for our purposes, to note the following clli!r=.
acteristj_cs of these techniques: The stimuli are capable of arousing
many different kinds of reaction: for example, an ink blot, which can
be perceived in different ways; a picture, which can elicit a variety of
stories; a set of dolls, which can be made to behave in many ways. There
are no "right" or "wrong" answers; nor is the respondent faced with a
set of limited alternatives. The emphasis is on his perception of the
material, the ll.leaning he gives to it, the way in which he organizes or
manipu~ates it. The nature of the stimuli and the way in which they are
preseri'(~d do not clearly indicate the purpose of the test or the way in
which the responses will be interpreted. The individual is not asked to
talk directly abou~ himself. The ostensible subject matter is the ink
blot, the picture, Hie dolls, or whatever, not the individual's own experi-
I
ences or feelings. However, the responses are interpreted as indicating
the individual's own view of the world, his personality structure, his
needs and feelings,! his ways of interacting with people.
In a projectiv~ test, the individual's r~p0l!:ses are not. taken at face
1 The student w~o is interested in pursuing the subject of projective techniques
in the study of personality should consult one of the volumes that give descrip-
tions of the various methods-for example, Abt and Bellak (1950), Anderson and
Anderson (1951 ) , and BeII (1948).
282 DATA COLLECTION: III
value-that is, with the meaning that the subject presumably would
expect them to have-but are interpreted in terms of some pre-estab-
lished psychological conceptualization of what his responses to the
specific test situation mean. 2 This underlying conceptualization pro-
vides the framework for interpreting the responses. Usually the system
of interpretation provides for considering responses not in isolation
but in terms of patterns. In effect, the clinician attempts to arrive at a
psychologically coherent picture of the individual by deriving the full
meaning of any particular response tendency from the total record of
his replies. _
One of the most frequently used projective techniques in the \
clinical setting is the Rorschach Test, consisting of ten cards, on each
of which is a copy of an ink blot. The subject is asked, "What might \
this be?" Another commonly used technique is the Thematic Apper-
ception Test, or T.A.T. This test consists of a series of pictures about
which the subject is asked to tell stories. In some of the pictures the
persons or objects are quite clearly represented, in others they are not;
some. of the pictures deal with ordinary or ~sual events, some with
situations that are unusual or bizarre.
Techniques such as these are designed to elicit a rich sample of
behavior, from which a great variety of inferences can be drawn. Some
inferences may have to do with adaptive aspects of the person's be-
havior; that is, how well he carries out the task posed by the test (to
tell. what an ink blot looks like, to make up a story, etc.). Others have
to do with expressive aspects. The way the person deals with the
materials of the test is taken as reflect~ng the "style" of his personality;
for example, his approach to the test may show constriction or expan"
siveness, intellectual control or impulsiveness, etc. Inferences about
adaptive and expressive aspects are generally considered relevant' to a
description of the individual's personality struc;ture. In addition, from
the content of what the individual says, inferences may be drawn about
2 Some authors (e.-g., Deri et aI., 1948, Proshapsky, 1950) consider this the
essential characteristic of projective techniques. We do not share this view because
we believe that all data-collection methods which attempt to measure -the
characteristics of a person-for example, his attitudes-except those which assume
In' a priori validity in self-report, involve an interpretation of responses in terms
of some psychological formulation. What is distinctive about projective tests'in this
respect, perhaps, is the extent of inference involved-that is, the lack of apparent
relevance of the responses to the characteristics about which inferences are drawn.
his needs, attitudes, values, conflicts, ideologies, and conception of
himself. Such inferences rest on the assumption that what the individ-
ual perceives in the test materials represents in some way (though not
necessarily in terms of direct correspondence) an externalization or
projection of processes within himself.
Other projective tests have somewhat more specific focus. A rela-
tively new technique, the Tomkins-Horn Picture Arrangement Test, is
designed for group administration and machine scoring. It consists of
twenty-five plates, each containing three sketches that may be arranged
in various ways to portray a sequence of events; the subject is asked to
arrange them in the most reasonable sequence. The responses are
interpreted as providing evidence concerning the following dimensions
of personality: conformity; social orientation (sociophilia, sociophobia,
aggression, dependence, etc.); optimism-pessimism; level of function-
ing (relative emphasis on thinking, fantasy, affect, overt behavior);
and work orientation.
Other commonly used tests are: word association, sentence com-
pletion, doYl play, and figure drawing. In the word-association test, the
subject is presented with a list of words; after each one, he is to respond
with the first word that comes to his mind. Both the rate and the
content of his responses may indicate areas of emotional disturbance.
In the sentence-completion test, the first few words of a possible sen-
tence are given, and the individual is asked to complete it. Like the
word-association test, the sentence-completion method may provide
clues to areas of emotional disturbance; any given area may be investi~
gated by presenting the respondent with relevant sentence beginnings.
In doll-play procedure, the subject is given a set of dolls, usually repre~
senting adults and cHildren of both sexes, and is either encouraged to
play freely with themjor to show how they would act in various circum"
stances. This pw~ed).lre is, of course, especially appropriate for use
with children; it is weI! suited for eliciting feelings about family rela-
tionships. In the figure-drawing test, the subject is asked to "draw a
person," to "draw a, man," or to "draw a woman," etc. The assumption
is that the drawing rf-presents the person's image of himself, and that
unusual features represent areas of confli,:t. strain, etc. In addition to
the relatively specific function ascribed here' to' each of these tests,
each of them may be used as a basis for broader interpretations about

the individual's personality.
Many of these techniques have been subjected to much investiga-
tion; standardized methods of administration, scoring, and interpreta-
tion have been published. Nevertheless, questions have been repeatedly
raised about their validity, and the research evidence on this point is
far from conclusive. This matter will be discussed in more detail in the
summary section of this chapter. \
Despite these questions about validity, projective tests from which
inferences about personality structure may be drawn have been used
fruitfully not only in clinical research and practice but in studies con-
cerned with the relation of individual personality to various social and
cultural factors. A number of studies concerned with the relation of
social attitudes to the individual's broader personality structure have
made use of projective tests. For example, Adorno et a1. (1950), in
their study of The Authoritarian Personality, used T.A.T. pictures as
well as partially structured interviews in assessing the personalities Qf
individuals who scor~~ high and those who scored low on question-
naires designed to measure anti-semitism and eth~ocentrism. In com-
paring the T.A.T stories of these prejudiced and unprejudiced people,
they found that ':~he unprejudiced identified more closely with the
heroes in thei.~:stories, expressed aggressiop in more sublimated forms,
tended to:,..efllphasize autonomous behavior based on inner rational
decision rather than on external forces, and described status relation
ships between men and women, parents and children, or Negro~ and
whites as more nearly equal than di~ the prejudiced individuals.
Another area in which projective tests of personality 'have been
widely and fruitfully used is the study of the relationship between
culture and personality. An example is DuBois' study (1944) of The
People of Alar. DuBois, an anthropologist, spt;nt some time in a srnal1
village in what was then the Netherlands East Indies, making obseJ;Va-
tions about the behavior and beliefs of the people and their cultural
practices, and reconstructing the biographies of several individuals.
From these materials a psychoanalyst, Kardiner, drew a picture of 'the
"modal personalitY': of the people of Alar. However, in order to. study
rdationships between culture and personality, it is important to have
evidence about personality patterns which is secured independently of
PROJECTIVE METHODS 28',
the description of cultural behavior. Accordingly, Rorschachs were
administered to a number of the villagers; the records were analyzed
by a specialist in this technique, Oberholzer, who had no knowledge
of the cultural descriptive materials. The similarity between the per-
sonality sketches obtained in the two ways was striking.
Some investigators (Henry and Guetzkow, 1951, and Horwitz
and Cartwright, 1953) have experimented with adaptation of the
Thematic Apperception Test for the diagnosis of characteristics of
smaIl groups. Pictures are used which are believed especially likely to
elicit responses indicative of group functioning; the group as a whole,
rather thari each individual member, is asked to make up a story about
each picture. The assumption is that a small group develops a "group
personality" which is revealed in the response to the pictures in a way
analogous to the revelation of an individual's personality through his
response to a projective technique.
PROJECTIVE METHODS IN THE STUDY OF SOCIAL ATTITUDES
When the focus of an investigation is on the specific content of

an individual's attitudes toward some social object, rather than on his
general personality structure, the projective techniques designed for
the study of personality are not very helpful. Yet f projective approach
freguently seems desirable as a way of encouraging freedom and
spontaneity of expression. This is especially likely to be the case. when
there is rca-son to believe that respondents may hesitate to express their
opinions directly for fear of disapproval by the investigator, or when
respondents are likely to consider direct questions as an unwarranted
invasion of privacy or to find them threatening for some other reason.
A considerable, variety of ingenious techniques have becn devel-
oped for use in such situations. These techniques share some of the
characteristics of the projective methods already described: they have
the capacity of arousing many different reactions, they encourage a
free response on th~ part of the individual, they do not ask him to talk
directly about himself or his own views or feelings, yet his responses,
are t'1}en as reflecting his own attitudes. In line with the purpose of'
tapping specific athtudes, the test materials or instructions usually
provide a more specific subject-matter focus than do those used in tests
designed to yield information about broad personality patterns; for

example, the individual may be presented with a picture in which
some of the characters are white and others Negro, or with a photo-
graph which he is told represents a union meeting, or with a cartoon
in which a leader is giving orders to a group.
Of course, if no precautions are taken, the general purpose of the
test, or at least the topic with which it is concerned, may be quite ap-
parent. However, two procedures may be used in the construction of
a projective attitude test in order to make its objective or the topic
involved less patent: a fictitious purpose may be ascribed, which ap-
pears plausible in the light of what is demanded of the subject; and,
in addition to the stimuli relevant to the investigator's objective,
"neutral" or unrelated stimuli may be included, so that the subject's
suspicions are not aroused by having the same "theme" or "issue" occur
in every stimulus situation presented to him. For example, suppose
that we were interested in measuring attitudes toward Negroes by
means of a series of pictures showing Negroes and whites in a variety
of social settings. By having each picture exp~sed for a brief period of
time (which incidentally makes for greater ambiguity and hence in-
creases the influence of inner determinants) and having subjects de-
scribe what was seen, it is not at all difficult to get subjects to accept
the idea that the purpose of the test is to see how much people can
recall of stimuli presented very briefly. If the subject is allowed to keep
the picture in front of him as he makes up a story or gives his opinion
of what is happening, then the purpose of getting at "imaginative
ability" or "h~w different people ,see different situations" c~n be
ascribed to the'test procedure. If, in addition to the pictures showing
Negroes and whites together, we also employ pictures involving only
whites, country scenery, animals, etc., the likelihood that the subject
will correctly perceive the purpose of'the test is reduced.
When limited time prevents the inclusion of "neutral" stimuli,
or with highly sophisticated respondents, it may not be possible ,to
conceal the purpose of a projective attitude instruplent. Yet Ahe
transparency of the purpose is not necessarily a serious disadvantage.
While the subject,may be motivated to distort his responses in the
direction of creating a favorable impression, this is not as easily done
as one might suppose. In the process of telling a story about a Negro-
~
PRO JECTIVE METHODS 287
white social situation with the intention of appearing unbiased, the
subject may still reveal his beliefs or feelings by automatically assign-
ing Negroes to inferior roles or by unwittingly attributing behavior to
Negroes along stereotyped lines. 3 The subject may not even realize
that some of the beliefs and views he has about Negroes are indicative
of bias; he may assume that most people see Negroes as "lazy" or
"unintelligent" because that is the way they are. Of course, it is still
true that conscious attempts to distort may reduce the. exte~t of vali~ ,
info~:tion obtained. h~ vSo..b~'o.\:- ~~t)\U.t\'\t. 1e.. V-~m
~ems--like~tha t-the--purpose-o'f-a- projec~ive- -test will
be-apparent,one-may.-ask Why bother with such a test, in view of the
time and effort involved in its construction and application? Why not
simply use a direct interview or questionnaire? There are several pos-
sible reasons for preferring the projective test, even though theJatter
may be transparent, which must be evaluated for each particular situa-
tion:
1. The subject may find it easier to express himself if he is not
explicitly talking about his own feelings and attitudes, even though he
knows that what he says will be so interpreted. He may have a certain
measure of distance from the topic if he can ostensibly talk imperson-
ally, and a certain measure of security in not having to say in so many
words that this is how he himself feels. /
2. With the best intentions in the world, the subject may be
unable to describe his feelings and attitudes as accurately as they may
be discerned in the projective test situation. Suppose one asks a sub-
ject, "When fou meet a person for the first time, what do you notice,
first, the person's sex or race?" The subject may well have been totally
insensitive to this aspect of his behavior. He may literally not know.
But suppose one asks him to describe the persons represented in a series
of pictures, mentioning whatever occurs to him as he looks at the
pictures. One may readily determine what is remarked upon first, and
one may even determ'ine this analytically by presenting pictures of
individuals representing the various possible combinations of sex and
race along with age, m;Inner of dress, and situational context. That is,
llWe..do not wish to .suggest th.a~ is a simple correspon ce between a
person's belicts-o~elingsland the characte:rs--he...d$~ts in a story. A s 'ect may
depict a Negro as bemg~ an inferior social role in a-story without feeling bat
this is the only appropriate role-for~~oes.
288 DATA COLLECTION: 1II
from the responses to the entire series of pictures, one may be able to
make such statements as, "In describing white females, the subject
remarks upon their sex before he remarks upon their race, but in
describing white males the reverse is true; in describing Negroes, he
consistently remarks upon race first, regardless of sex." And one can
compare the patterns with those available from normative data. Can
one generalize from such a test? This is a problem of validation.
3. Sometimes access to certain populations of potential subjects
(e.g., school children, workers in a factory, etc.) may be withheld if
the topic under investigation is made explicit to the subjects, but
granted if it remains tacit even though obvious.
4. Even though the purpose of a projective attitude test is appar-
ent, it may produce more extensive information than a questionnaire
or 'even an interview with open-ended questions would. In the descrip-
tions of the pictu~ed situations, for example, we may see how attitudes
cQlor perception, or what aspects of attitude (feelings, beliefs, etc.) are
significant for thc individual, and so on.
The projective techniques that have been.. devised for the study
of social attitudes vary in the effectiveness with which they mask'their
purpose, in the richness of personality material they reveal, in the
ambiguity of the stimulus presented to the subject, and in the expendi-
ture of $}<iJI.-{nd effort ~ecessary to, the collection an~ analysis of re-
spons_3Few of the specIfic adaptatIOns oUhese techmques have been
employed in more than one investigation; none is supported by a wide
body of experience. Nevertheless, we shaH briefly describe the major
techniques, since they are directly relevant to the study of social rela-
tions and since they do not usually require the complex skills demanded
by the projective methods for the study of personality. Caution in their
use and interpretation, however, is indicated.
We shall group the many specific adaptations of projective tests
for the study of' social attitudes in terms of certain general materi"lls
and methods of approach: verbal, pictorial, play, and psychodrama tic
techniques. '
VERBAL TECHNIQUES. Perhaps the simplest method is based on the
classic technique of word association employed many years ago by Carl
Jung in the study of abnormal behavior. As employed in the study of
social attitudes, the technique is essentially the same: a number of
PROJECTIVE METHODS 23=)
words are presented to the subject, one by one, and he is asked to in-
dicate the first thought that he associates with each word. Some of
the words used as stimuli are neutral; some relate tb the social attitudes
being investigated. The speed of response and its emotional concomi-
tants, as well as its content, may constitute valuable indicators of atti-
tude. This technique has been used frequently in market research, to
discover, for example, associations to a given brand name or to a
proposed name for a new product.~ It has also been used in several
studies of the relation of specific attitudes to broader personality pat-
terns. Murray and Morgan (1945), for example, used a modification of
the word-association technique, asking the subject to respond to such
words as communism, religion, Negro, by giving the most descriptive
adjective he could think of.
Somewhat similar to the word-association method is the sentence-
completion technigg~ The individual is presented with a series of in-
~lete sentences which he is asked to complete, usually under some
time pressure to ensure spontaneity of response. The content. of
responses, if the items are carefully selected, may provide considerable
insight into the person's attitudes. However, for sophisticated subjects
it is unlikely that the purpose of the task is effectively masked, even
though neutral or irrelevant items are included. On the other hand,
time p-lessure may do much to prevent concealment of attitudes.'
This technique ~,as been used for studying attitudes of many kinds.
fi, For example, Kerr (19:4-3) used it in a study of national stereotypes held
by English people. Some of the sentence beginnings were:
The thing I do admire America for is . . .

The trouble with America is . . .
When I think of the Russians I think of . .
If the British and Soviet armies fight side by side they
If you invite an An;erican to your home he may . . .
The technique

has I been used frequently in market research; for
example, a study of attitudes, motives, and behavior of subscribers to
4 Smith (19 54J gives a comprehensive review of the use of projective tech
niques in marketing researcH. For a discussion from a different point of view, raising
<Juestions about the ethics bf using projective techniques, see Packard (1957).
5 For detailed discussion of the method, see Rohde (1946), Rotter and Willer
man (1947), Rotter (1951), and Stein (1947).
Better Homes and Gardens (Smith, 1954) used the following sentenc~
beginnings, among others:
To own your own home

The ads in Better Homes and Gardens
Most fiction magazines . . .
Sometimes the items are phrased in the first person ("When I am
asked to be in charge, I generally tend to . . ."), sometimes in the
third person ("When tjley asked him to be in charge, he . . ." or
"When they asked Bob to be in charge, he ..."). As yet, there is no
clear evidence about which phrasing is more successful in eliciting the
respondent's own attitudes .. Two studies directed to this point have
reached contradictory conclusions. Sacks (1949) found that inter-
pretations based on the first-person form agreed better with psychia-
trists' diagnoses, and that respondents, comparing the two forms; re-
ported that their responses to the first-person items corresp,onded more U
frequently to their own feelings. However, Getzels (1951) found that
third-person items discriminated better between a group previously
classified as "maladjusted" and one classified as "adjusted." He also
found that the third-person items yielded more expres~ons of hostility
to Negroes, more expressions of cO!lflict with parents, and more
"neurotic" answers to personality-relevant questions than did the
first-person items. He found little difference between the two forIl}s
on topics permitting a variety of acceptable responses-that is, on topics
not generally subject to social pr7ssure. His subjects reported that in
responding to the third-person items, they usually thought of. the
person as "Someone like myself." However, 'when the item -refers to
"most people" rather than some mythical third person, it appears that
respondents frequently make a: serious attempt to estimate other
people's views rather than expressing their own.
Similar to the sentence-completion technique are the story- and
argument-completion techniques. The subject is given enough of the
story or argument to focus his attention on a given issue but not enough
I to indicate how it wi11 turn out. He is then asked to supply a conclusion.
For example, a subject might He asked to complete the following
argument (Murray and Morgan, 1945):
X, who has grown up in the belief that war never settles
anything, is contemplating with some misgivings the probability
of being drafted. He meets his friend Y, and they stop to have a
beer together.
Y has been accepted, and is on his way to camp. He tells X
that he expects to be sent to the Pacific. He is buoyant at the
prospect, and says that he will be glad to take his shirt off and
get in to it all.
"How is it possible," asks X, "that . . ."
Similar in some respects to the open-ended question (discussed
jn the preceding chapter) is the so-called projective question. The sub-
ject is asked to respond to a vague question (e.g., "What are you?"),
and the response is taken not at face value but rather from the point
of view of the perceptions, beliefs, sentiments, and motives that pre-
dispose the individual to answer in the way he does. Such questions
frequently take the form of asking about a possible event in the future
("What would you do if you inherited a large fortune?"), or about an
imaginary event ("Suppose a man from Mars came down to this planet
and you were the first person he saw and he asked what kinds of people
there were in the world; what would you tell him?"), or about some-
thing_the subject clearly is not in a position to know.
Sometimes projective questions take the for111 of asking about
other people's views. The assumption here is that a respondent may
hesitate to express critical or unpopular views as his own, but will put
them into the mouths of other people. Smith (1954) gives the follow-
ing illustration from the field of market research. Instead of asking
housewives, "What do you think are the objectionable features of this
cleanser?," the interviewer said: "Some women who use this cleanser
find a lot of faults, with it. I wonder if you can guess what they are
f/ objecting to." I
Again, as i~ the case of sentence completion, it must be not~d that

one cannot always safely assume that a person's answer to a question of
this sort really refl~cts his own feelings, especially if the question is of
the "most people,,1 form. Maccoby and Maccoby (1954) give the fol-
lowing striking example:
I,
. . . a young woman being interviewed about her attitudes to-
ward her job situation was asked: "How do most of the girls in
the office feel about the supervisor?" She answered: "They think
he's wonderful. They'll do anything for him." At which point the
interviewer followed with: "And how about you-how do you
feel about him?" with the reply: "I really detest him. I'm trying
to transfer out of the unit."
Weitz and Nuckols (1953), in a study of the relationship between
job satisfaction of insurance agents and continuance in the job, used
both direct questions and those asking for estimates of the reactions
of others. The indirect questions were introduced with the explana-
tion:
We want to get your opinion about the attitudes of other agents
toward their job. Below are a number of questions which can be
answered by a percentage. Circle the per cent figure you believe
best answers the question. If you don't know, guess.
The questions were of the form:
Approximately what per cent of the agents in your company think
that: The training they received was good.
0%,10%,20%,30%,40%,50%,60%-, 70%, 80%,90%,100%_
The direct questions were introduced with the explanation:
Now we'd like to get your attitude about your own job. Check the
word or phrase which you feel best completes the statement for
you.
The questions were of the form:
The training I received for my present job was
__ poor
--adequate
__ excellent
Although neither the score lfased on the direct questions no~ that
based on the indirect questions correlated yery highly with job survival
at the end of a year, these investigators found that the direct questions
about the person's own reactions provided a better basis for prediction
than did those asking for estimates of other people's reactions.
Another verbal technique is that of asking the respondent to
describe the kind at person who \~ould behave in a specified way. This
:J.pproach has been used most frequently in mar,ket research, to elicit
respondents' "images" of a given product, but it would seem easily
adaptable to the investigation of attitudes of other kinds. Smith (1954)
gives an example of this approach in a study of attitudes toward small
cars. The person being interviewed is asked to imagine that a new
family has moved into his block. Before he sees any members of the
family, he se~s their car parked outside the house; it is a Burton (a
small car). He is asked, "What kind of people would you guess they
are?"
One use of this approach in market research (Haire, 1950) has be-
come almost a classic.' In a conventional survey of attitudes ,toward
Nescafe, an instant coffee, women were asked, "Do you use instant
coffee?" If the answer was "No," they were asked, "What do you dislike
about it?" Most of the replies were along the line, "I don't like the
flavor." The investigators, suspecting that this was a stereotype that did
not express the underlying reasons for rejection of instant coffee,
switched to an indirect approach. Half of the sample of housewives
interviewed were presented with the following shopping list made out
by a hypothetical woman:
pound and a half of hamburger
2 loaves Wonder bread
bunch of carrots
1 can Rumford's Baking Powder
Nescafe instant coffee
2 cans Dd Monte peaches
SIbs. potatoes
The other half of the sample were presented with the same list,
except that "1 lb. Maxwell House Coffee (Drip Ground)" was sub-
stituted for the Iiescaf~. Each respondent was asked to read the
shopping list and then to write a brief'description of the personality
and character of the woman who had made it out. The differences
between the descriptions of the woman who bought Nescafe and the
one who bought Maxwell House coffee were striking. Almost half of
the women who read the list containing instant coffee described its
writer as lazy an9 failing to plan her household purchases well; the
woman who boug4~ the drip-ground coffee was hardly ever described
in these terms. In addition, the woman who bought instant coffee was
more often descri~ed as a spendthrift and a poor wife. A check of the
pantries of the respondents showed that most of the women who de-
scribed the buyer of instant coffee in these unfavorable terms did not
have instant coffee on their shelves; those who did not describe her
unfavorably were much more likely to have instant coffee. In other
words, it seemed clear that the decision to buy or not to buy instant
coffee was influenced at least as much by attitudes about what COn-
stitutes good housekeeping as by reaction to the flavor of instant coffee,
but these attitudes could not easily have been elicited by direct ques-
tioning.
A variation of this approach is the matching technique, in which
the respondent is given a list of various brands of the same kind of
product and another list of different kinds of people (e.g., doctors'
wives, electricians' wives, stepmothers, career women) and is asked
to match each kind of person with the brand she would be likely to buy.
Gardner and Levy (1955) report striking differences in the "image" of
different brands~elicited by this technique. f
PICTORIAL. TECHNIQUES.6 Pictorial techniques, many of them bor-
rowed from well-established clinical proced.ures, have long been popular
in the projective study of social attitudes. The Thematic Apperception
Test (T.A.T.) has been the stimulus for several ventures.
Proshansky (1943) was one of the ~rst to employ the T.A.T. type
of picture in the study of social attitudes. Ambiguous pictures of situa-
tions involving labor were intermingled with regular T.A.T. pictures
and exposed to a group for five seconds each. The subjects were asked
to write briefly what they thought the pictures represented. On the
basis of these stories, three judges rated the subjects' attitudes toward
labor on a five-point scale. The pooled ratings of the" three judges cor-
related .87 for one group of subjects and .67 for another group wi\h a
standard scale for measuring attitlfde toward labor.7
Sayles (1954) used pictures somewhat differently in a study' of
attitudes of union members toward grievance procedures. Finding that
many questions seemed threaten_jng to his respondents, or too personal,
6 The pictorial techniques discussed here differ from those discussed in Chapter
7 in that they use pictures to stimulate indirect expression of. the respondent's
attitudes, whereas those discussed in the preceding chapter ask direct questions
about the subject's response.
7 The construction and use of non-disguised attitude scales are discussed in
Chapter 10.
he decided to use a projective approach. On the basis of exploratory
interviews, he identified seven stages in the grievance process; for
example, informal discussion with fellow workers on what to do about
a complaint, informal meeting of the foreman and union official and
'the worker involved, etc. He took photographs of such situations in
other plants, with personnel unknown to his respondents. These
photographs were then shown to the respondents, with an accompany-
ing explanation-e.g., "This person has a grievance, something bother-
i~g him, but before taking it to the union he's discussing it with his
fellow workers." The respondent was then asked such questions as,
"How do you think this guy is feeling right now? What has happened
just before this picture was taken? What do you think is going to
happen next?" Sayles reports that the responses elicited by this tech-
nique were very similar to those obtained in the intensive interviews
conducted in the exploratory stage of the research. The pictorial tech-
nique had a number of advantages: It took only about ten minutes to
administer, compared witn an average of two hours for the intensive
interviews; it could be administered in the factory, whereas the inter-
views had to be conducted at home in order to assure free responses; it
could be administered by someone unknown to the respondents,
whereas the interviews had to be preceded by a six-week period of
developing rapport before respondents were willi1)g to talk freely.
Pictures of the T.A.T. type have been used in several studies of
attitudes toward minority groups. In the Authoritarian Personality
study (Adorno et al., 1950), ten pictures were presented to each sub-
ject: six pictures from the T.A.T. and four especially aimed at uncover-
tig attitudes toward minority groups. The latter were of "Jewish-look-
ing people in a poor district," "an older Negro woman with a younger
Negro boy," "a young I
couple in zoot suits," and "a lower-class man
accosted by a policeplan wielding a nightstick." Subjects were asked to
construct a complete story about each picture, and their stories were
recorded verbatim ~y the examiner. Stories were analyzed quantita-
tively, in terms of t~e strength and frequency of expression of various
needs, and qualitatively, in terms of the theme expressed in the various
stories. Interestingly' enough, the pictures designed to distinguish be-
iween those with high and low ethnocentrism were less effective for
this purpose than the pictures from the regular T.A.T. series.
The "Human Relations Test" developed by the staff of the

College Study in Intergroup Relations (Cook, 1950) offers another
example. It consists of ten drawings, each depicting an ambiguous
situation of intergroup contact. One picture shows a basketball scene
with a white player lying on the floor and a Negro player standing
above him; another shows a couple applying for a room in a hotel, with
a sign on the wall indicating that the hotel has a restricted clientele; a
third portrays a scene in a restaurant in which three Negro men are at
the entrance talking to the headwaiter, who has his hand upraised; etc.
Each scene can be interpreted as one of conflict or of amity. The sub-
jects are asked to construct a short story about each scene. Their stories
are evaluated in terms of whether or not conflict is portrayed, of how
the subject allocates the blame for conflict, of how the conflict turns
out, e~c.
T.A.T.-typ~ pictures have also been used in market research (see
Sinith, 1954). In a study of magazine readership, for example, respond-
ents were shown a picture of a family sitting in a living room reading,
and were asked to tell a story about it. In a study of factors influencing
purchases of consumer items, respondents were asked to ten stories
about such pictures as one of a woman walking down the street with
money in her hand; there is a store on one side of the street, a bank on
theother;;
Fromme (19:U}presented to subjects political cartoons, each with
'four alternative captions judged by the author to represent a-full range
of favorable and unfavorable attitudes, and requested the subjects to
choose the one that best fitted each cartoon. Although Fromme' used
the cartoons and their captioning primarily to stimulate discussion and
to obtain qualitative insights into political attitudes, it seems prbbable
that the use of cartoons has fruitful possibilities for the quantitative
study of social attitudes.
J. F. Brown (1947) modified the Rosenzweig Picture-Frustration
Test for use in the study of ethnic attitudes. The Rosenzweig Test uses ,;
a cartoon format, in which one character is represented as saying some-
thing; there is ~ blank balloon for the other character, and the respond-
ent is asked what this second character would probably say. In the
clinical form of the test, all the pictures portray-situations of frustration
for the second character; the responses given for him are scored in terms
PRO JECTIVE METHODS 297
of whether they indicate a tendency to place blame on others, or on
oneself, or not to react in terms of blame. The Brown adaptation of this
test includes a number of pictures in which (a) Negroes or Jews
frustrate whites or non-Jews and (b) whites or non-Jews frustrate
Negroes or Jews. In one picture, for example, a "Jewish-looking"
moneylender refuses a loan to a "non-Jewish-Iooking" man, saying:
"Sorry, sir, but we cannot see our way clear to make the loan." The
subject has to fill out a blank space indicating what the frustrated
person replies. The type of reply is considered by Brown to be indica-
tive of the subject's attitude toward the minority group in question.
As with many of the pictorial techniques, the visual cues necessary to
elicit the attitude toward the different ethnic groups make the pur-
pose of the test manifest to any but the most naive subjects.
A similar format has been used in other studies, with different
content. In a study of reactions to leadership and authority (Sanford,
1950; Sanford and Rosenstock, 1952), several drawings were used. One
showed two men sitting and talking; the first is saying, "You were
telling me about something that's been bothering you. What was the
problem?" Another showed a man saying to a group of people, "Since
I'm head of this group you'd better do as I say." This study, unlike
many of those in which projective devices have b~en used, was a large-
scale survey, in which interviewers questioned a cross-section of the
population of a large city. The investigators reported that respondents
were interested and cooperative; the use of the pictures seemed to
) ,reak the monotony of the interview, and made readily comprehensible
situations and problems that it might have been difficult to present
verbally. The reliaqility of coding was reported to be equal to that
generally achieved in coding interview questions, and repetition of
I
the interview with ;part of the sample after a _month's lapse showed
repeat reliability c6mparable with that in conventional surveys. As
evidence of validitx, it was reported that the pictures discriminated
between groups rat~d as authoritarian or equalitarian on an authoritar-
ianism-equalitariani,sm scale included in the interview.
fictures of this sort, too, have been used in market research. For
example, one used ~n a study of attitudes toward cigar srnoking showed
a man saying to a woman (presumably his wife), "I've decided to take
up smoking cigars, dear" (Smith, 1954). A picture used in a study
carried out for a grocery store whose sales were declining showed two
women sitting at a table drinking coffee: One woman said, "Well, I feel
I have to buy food where the price is lower-that's the main thing
as far as I'm concerned." The other woman was shown as saying, "Art
and I agree that I should shop where . . ." The respondent was asked
to fill in the rest of the answer (Zober, 1956).
The device of asking the respondent to describe the kind of person
who would behave in a certain way, described in the preceding section
on verbal techniques, has also been used in conjunction with pictorial
material. Smith (1954) gives an example from a study testing proposed
ads for a new perfume. One of the ads featured a Gauguin picture of
South Sea girls; the other, a picture of a young American girl clasping
a bouquet of flowers. The interviewer said: "A good many women
prefer this picture; others prefer this one. I wonder if you could say
anything about these two types of women-the sort of people they
are." The responses made it clear that the Gauguin picture would not
appeal to the market for whom the perfume was intended.
PLAY TECHNIQUES. Techniques involving the manipulation of dolls
have been used in investigating the attitudes of young children.
Hogrefe, Evans, and Chein (unpublished study) gave their subjects a
number of "white" and "colored" dolls 'and asked them to play out
specific scenes, such as "going to school" or "arranging a party," as
~hough they were producing a movie. The inclusion or exclusion of
the colored dolls, as well as the role assigned to them, provided a simple,
objective score which was taken as a measure of the child's 'attitude
toward Negro children, The majority of the children they tested
showed a striking avoidance of segregated patterns; that is, they created
mixed situations far more often than would be expected by chance.
This was in keeping with the children's reports of their own play
activity; in answer to the question, "Do you ever play with Negro
children?," four fifths of the white children said "Yes." But observa-
tion of their actual behavior in an interraCial recreation center showed
a striking contrast; on repeated occasions, when children were asked
to pick partners fQr some activity, the number of segregated pairs was
far greater than would be expected by chance.
Hartley and Schwartz (1948) combined doll play with pictorial
material in the investigation of intergroup attitudes of children. Pic-
torial backgrounds carried characteristic symbols of the Catholic reli-
gion in one set, the Jewish religion in another, and middle-class sur-
roundiJags (without religious identification) in a third. Identical
family sets of dolls were placed on the three backgrounds and the
child was allowed to use them in playing out situations such as a
birthday party, school bus, etc.
PSYCHODRAMATIC AND SOCIODRAMATIC TECHNIQUES. Although psy-
chodrama and sociodrama have not been used systematically in the
study of social attitudes, the fact that they are methods of considerable
flexibility makes it reasonable to examine the possibility that they
might be .used in this way. The methods require that the subject act
out a role, either as himself (psychodrama) or as somebody else (socio-
drama), as he would in a real-life situation. For example, a white sub-
ject may be presented with the problem of acting out the role of a
Negro factory worker who has been absent from work several times and
who has just been called into the foreman's office to explain his absen-
teeism. The manner in which he plays his role, the history that he
creates for the role, etc., may provide considerable insight into his
attitudes. The investigator, in much the same way as an observer, can
record the behavior for later analysis, can categorize it on the spot, or
can rate"it in terms of various scales, etc. Psychodr~ma and sociodrama,
it should be noted, are among the few tools available for the systematic
investigation of social skills. They enable one to place a person in sit-
uations in which one can observe how skillfully he behaves in relation
to other people. s
Structure? Disguised Tests ot Social Attitudes

Investigators of! s<?cial attitudes have been interested in develop-
ing tests that would have some of the advantages of the projective
I
techniques-notably that of not making apparent to the subject the
investigator's purpose-but that would be simpler to administer, score,
and interpret than riiost projective tests are. Accordingly, a number of
S For a discussion .of the research uses of psychodrama, see Franz (1940).
structured disguised tests have been devised. Campbell (1950) de-

scribes such measures as "approximating the objective testing of
attitudes."9
Most tests of this kind are based On the fact that a person's atti-
tudes are likely to influence his perceptions, beliefs, judgments, mem-
ory, etc. That this is so has been demonstrated in a number of studies.
For example, Bartlett's classic experiment (1932) on the recall of
pictorial materials and stories revealed the influence of social attitudes
on memory. In one of these experiments, students at Cambridge Uni-
versity read twice a story drawn from the folklore of a culture foreign
to them and dealing with matters with which they had little familiarity.
After varying periods of time, the subjects were asked to repeat the
story as accurately as possible. Bartlett found that the reproductions
all showed systematic changes, increasing with time, which could
partly be accounted for by the cultural frame of reference of the sub-
jects.
Similarly, Newcomb (1946), in- his study of the effects of social
dl~nrate on attitudes and information, showed the nonrandom character
Of right and wrong answers to factual questions. On a questionnaire
about the Spanish Civil War, pro-Franco students tended to know
those items of information that favored Franco, but not those that
were anti-Franco; the reverse was true for anti-Franco students. New- I
comb commented that on difficult items "the direction of guessing;
is altogether likely to be weighted toward the subject'_s attitude. If this i
reasoning is correct, the '. . . test tends to become an attitude test." I
Findings of this nature have given rise to the development of aP-1
parently objective tests difficult or complex enough to allqw the sub-/
ject's attitudes to influence his performance. To the respondent, the
situation of.measurement is similar to that of an achievement or ability,/
test. In contrast with most projective tests, in which the: individual iS
encouraged to believe that there are no objectively "correct" answers,
I
the subject is led to believe that there are "right" and "wrong" re-I
sponses, and the attempt is made to motivate him. to do as well al.
I
possible on the test. .
9 In this section we have drawn heavily on Campbell's review (1950) of sud

tests. I
STRUCTURED DISGUISED TESTS 301
However, the rationale underlying the use of such tests is essen-
tially the same as that underlying the use of projective techniques:
When there is no clear, objectively verifiable basis for choosing among
alternative responses to a situation, an individual's response tends to
reflect his predispositions. Of course, the stronger the individual's pre-
dispositions, the more likely they are to be reflected in his responses.
Campbell (1950) provides a formula for constructing such tests:
Find a task which all your respondentswill take as objective,
and in which all will strive to do well. Make the task sufficiently
difficult, or use a content area in which respondents have had
little experience or opportunity for reality testing. Load the test
with content relative to the attitude you study. Look in the re-
sponses for systematic error, or for any persistent selectivity of
performance. 'If such be found, it seems an adequate basis for the
inference of an attitude.
Let us briefly indicate some of the disguised "objective" tech-
niques that have been used to measure social attitudes.
INFORMATION TESTS
Hammond (1948) employed an "information test" to measure

/
attitudes toward labor-management relations and toward Russia. His
questionnaire included three types of question. The first type, of which
there were eight items, forced the respondent to choose between two
alternative answers, both of which were, by intent, equally wrong, but
in opposite directions from the correct answer-e.g., "Average weekly
wage of the war worker in 1945 was (1) $37.00, (2) $57.00." The
second type, of wh~ch there were twelve items, required the subject
to respond to a question about which the facts could not be deter-
I
mined-e.g., "Russi,a's removal of heavy industry from Austria was
(1) legal, (2) illegal." The third series of questions consisted of twenty
genuine informatiori items, which were interspersed among the twenty
items mentioned above. These items, all relating to the same general
fields as the "test"j,items, were introduced in order to increase the
apparent authenticity of the test as one of information. Despite the
limited number of i~ems, Hammond's scales differentiated a labor-
302, DATA COLLECTION: III
union group from two business clubs, with practically no overlap, on

attitudes toward both Russia and labor-management relations.
Weschler (1950) also used this technique in a test of attitudes
toward labor and management. He found that, within a group of col-
lege students, scores on this indirect test correlated highly with self-
ratings of attitude. In addition, representatives of unions and of man-
agement attending a course in labor relations differed markedly in their
scores on the test.
REASONING TESTS
In laboratory investigations of the process of reasoning it has been

apparent for many years that syllogistic inference do~ not always pro-
ceed logically; frequently it is affected by the content of the syllogism.
G. B. Watson (1925) included as part of his "Measurement of Filir-
Mindedness" an inference test composed of statements of fact fol-
lowed by several conclusions that might be drawn. For example, one
statement was:
Statistics show that in the United States, of one hundred
men starting out at an age of 25, at the end'of forty years, one will
be wealthy, while fifty-four will be dependent upon relatives or
charity for support. lO
This statement was followed by several possible conclusions:
The prese~t social order cheats the many for the be_nefit of the
few.
The average young man, under present conditions, cannot count
on being wealthy atthe age of 65.
Most men are shiftless, lazy, or extravagant, otherwise they would
not need to be dependent.
The one man is living upon luxuries ground out of the bones ,of
the masses of common people.
Some day the workers will rise in revolt.
No such conclusion can fairly be drawn.
10 To the student of today, this may not appear to be a factual statement. But
at the time the test was devised-before the introduction of social security programs
-statistics did show that 54 percent of 65-year-old men were dependent on rela-
tives or charity for support.
STRUCTURED
, DISGUISED TESTS 303
The subjects were asked to check only conclusions established by
the facts given in the statement, drawing upon no other evidence.
They were cautioned to check only those inferences that were certain,
none that were merely probable. Checking anything other than "No
such conclusion can fairly be drawn" or the alternative that simply
rephrased the original statement (in this case, item # 2) was taken
as evidence of bias.
In a study of attitudes toward the Japanese, Morgan (1945) used
a test made up of pairs of syllogisms. One syllogism in each pair was
stated either in abstract form or with neutral content; the other,
roughly parallel, used statements about the Japanese. For example,
one pair read as follows:
No A's are B's. Some e's are B's. From these statements it is
logicalto conclude:
1. All C's are A's.
2. Some e's are A's.
3. Only a few e's are A's.
4. Some e's are not A's.
5. Most e's are not A's.
6. No e's are A's.
7. No logical conclusion can be drawn from the given statements
/
A trustworthy man does not engage in deceitful acts. The bomb-

ing of Pearl Harbor by the Japanese was a deceitful act. From'
these statements it is logical to conclude:
1. All of the Japanese are trustworthy.
2. Some of the Japanese are trustworthy.
3. Only a few of the Japanese are trustworthy.
4. Some of the Japanese are not trustworthy.
5. Most of the Japanese are not trustworthy.
6. None of tne Japanese are trustworthy.
7. No logical:conclusion can be drawn from the given statements.
The difference between the response to the abstract form and the
form referring to ,the Japanese was used as an indicator of attitude
toward the Japanese.
I,
The two syllogisms forming a pair were not, of
course, placed next to each other in the questionnaire as it was pre-
sented to the subj~cts.
TESTS OF PERCEPTION, l\1EMORY, AND JUDGMENT
It has long been recognized that perception, memory, and judg-

ment can be markedly influenced by one's predispositions and past
experiences. E. L. and R. E. Horowitz (1938), in their studies of the
development of race attitudes, designed a number of techniques based
on this fact. In one, the Aussage test, a complicated picture was exposed
for two or three seconds; then a series of standardized questions testing
perception and memory were aSKed. For example, after exposure of a
picture that did not include a Negro, the question was asked. "What
is the colored man on the corner doing?" With increasing age, the
children were more likely to "perceive" or "recall" the Negro in a
menial role. The Perception-Span Test consisted of a series of posters,
on each of which were mounted pictures of ten items. Each poster
was exposed for ten seconds; after it had been removed, the children
were asked to "tell all the pictures you can remember." Among younger
children, Negroes were less well remembered than would be expected
by chance; among older children there was a selective awareness of
them. The Pictorial Recognition Test involved exposing sets of Negro
and white faces and testing for recall by asking that the previously .
exposed pictures be picked from a larger group. Using this test, Seele-
man (1940-41) found that the discrepancy between memory for white
and for Negro faces correlated about .70 with a questionnaire measur-
ing prejudice. An unfavorable attitude tended to obliterate recognition
or recall of individual differences among Negro pictures.
Following the lead of the Horowitz studies, there-has been in-
creasing interest in the relation between attitudes, on the one han~,
and perception and memory, on the other. Illustrative of this develop-
ment is a series of techniques developed by C(lttell and his co-workers
(1949, 1950). One test, Immediate: Memory, required the subject to
recall statements from sets of twelve presented at one-second intervals.
Selective recall of statements relevant to a given attitude was taken as a
measure of the strength of the attitude. Another technique, the Misper-
ception method, iI1volved a one-second tachistoscopic exposure '?f at-
titude statements with misspellings. The subject was required to recall
the statement and note the misspellings; overlooking of the latter was
regarded as a sign of a strong attitude. Although Cattell's results, in
terms of correlation with other tests designed to measure attItude
toward the same objects, are not particularly encouraging, his tests
are cited as examples of the range of possibilities in the objective but
disguised measurement of attitudeY
Levine and Murphy (1943) found that both the learning and the-
forgetting of controversial material were related to attitude. Strongly
pro-Soviet and anti-Soviet groups of American college students wen:
asked to learn two passages about the Soviet Union, one of which wa~
very favorable, one yery un:favorable. Each group learned more rapidly,
and remembered longer, the passage in harmony with its own attitude.
Jones and Kohler (1958) introduced a qualification of the finding
that people learn best those statements which are in favor of theiI
position by suggesting that this is true only if the statement is suffi-
ciently plausible and effective to permit the individual comfortably
to endorse both the position implied by the statement and the sup-
porting reasons given. They hypothesized that people would be most
likely to' remember plausible statements that agree with their own
position and implausible statements opposing their position; and con-
versely, that they would tend to forget implausible statements in favor
of their own position and plausible ones opposing it. In testing this
hypothesis, they presented subjects with six statements favoring and
six opposing segregation; within each type of }tatement, three were
"plausible," three were "implausible." For example, a plausible pro-
segregation statement was, "Southerners will have to pay the price of
lowered scholastic standards if they yield to the pressures to integrate
their schools"; an implausible pro-segregation statement was, "If Ne-
groes and whites were meant to live together, they never would have
been separated at the beginning of history." As predicted, subjects who
were rated as pn~-segregation (on the basis of other attitude scales
administered at a jdifferent time) were most successful in learning the
plausible pro-segregation arguments and the implausible anti-segrega-
tion ones; those :who were anti-segregation were most successful in
learning the plausible anti-segregation arguments and the implausible
11 It is possible that the Misperception procedure gets at a dimension of atti-
tude not tapped by lother tests-something like "mental alertness or flexibility in
the presence of attitude-entangling stimuli." Note the possibility of getting com-
l'arative data by uSIng favorable, unfavorable, neutral, and irrelevant statements
of equal difficu;'~' '
pro-segregation ones; for each type of argument, the neutral group

made scores intermediate between the two extreme groups. Using an
index representing the difference between learning of plausible pro-
segregation and implausible anti-segregation arguments, on the one
hand, and implausible pro- and plausible anti- arguments, on the other,
they found no overlap between subjects who scored as pro- or anti-
segregation on their criterion attitude scale. On the basis of this find-
ing, these investigators suggested the possibility of constructing an in-
direct test of attitude in terms of the learning of relevant material in
which both direction and plausibility were varied.
A number of investigators have employed tests of "judgment" in
the study of social attitudes. Murphy and Likert (1938) showed re-
spondents pictures labeled as a union president, a railroad magnate, a
pacifist, a Negro champion of Negro rights, etc., and asked them to
judge the character of the pictured person in terms of courage, selfish-
ness, intelligence, conceit, etc. They found no relationship between
these ratings and attitudes as measured in a variety of paper-and-pencil
tests. However, since photographs do not usually provide significant
clues to character, it is reasonable to suppose, that such judgments
would be influenced by reaction to labels (e.g., union president) . Thus
this approach seems worth further exploration.
Several investigators have used discrepancies in judgments of com-
parable situations as measures of social attitudes. Watson (1925), in
a test of "moral judgment," asked for judgments of approval or dis-
approval about a variety of situations, sets of these situations being
identical except for the specific persons or groups involved. For ex-
ample, unwarranted search is made of a suspected "radical" head-
quarters in one item, and of a business corporation suspected of dis-
honesty in a parallel item. The discrepancy in judgment between the
two parallel situf!tions formed the basis for scoring.
Seeman (1947) employed a similar technique in the study of
attitudes of white persons toward Negroes. He presented college stu-
dents with brief descriptions of relations between men and women,
,involving such iss~es as extramarital sexual relationships, divorce, etc.,
'and asked for judgments about the behavior of the characters. For
example:
Bob and Helen want to get married soon. They have been
engaged for a year. So far as they can foresee, it will be impossible
for the marriage to take place for another two years at least. Bob
and Helen have already had complete sexual relations upon a
number of occasions. Helen says she can see nothing wrong with
this "as long as people marry eventually" and "do not feel guilty
about it."
a) Is this wrong for Helen? Yes __ No __ Uncertain __
b) Is this wrong for Bob? Yes - - No - - Uncertain--
Any remarks?
In half the copies of the test, each story was accompanied by a
picture of a white couple; in the other half, by a picture of a Negro
couple.
Since it would have been immediately apparent that the test was
concerned with attitude toward Negroes if each student received both
forms, each was given only one; the Negro and white forms were dis-
tributed randomly among the sample. The results were analyzed by
comparing the replies of the group of students who received the Negro
form with those of the group who received the white form. As expected,
the responses to the two forms of the test differed. Subjects who had
the form with the white illustrations more often judged the unconven-
tional behavior as being wrong; there were more "No" and "Uncertain"
answers to the form with the Negro illustration. /
The more detailed results of this study provide a vivid illustration
of the problems involved in the use of indirect techniques. Seeman had
hypothesized that prejudiced subjects would be more likely than un-
prejudiced ones to make different judgments for Negroes and for
whites. In order to test this hypothesis, he administered the Likert
scale of attitude toward Negroes 12 and divided the sample into a "more
prejudiced" and a "less prejudiced" group On this basis. To test his
hypothesis, he coni pared the replies of the "more prejudiced" subjects
I
who had received the Negro form with those of the "more prejudiced"
subjects who had received the white form, and the replies of the "less
prejudiced" subjects who had received the Negro form with those of
the "less prejudiced" subjects who had received the white form. Con-
trary to his hypoth~sis, he found that the replies of the "less prejudiced"
12 Attitude scales of this type are discussed in Chapter 10.
subjects who received the Negro form differed significantly from those
of the "less prejudiced" subjects who received the white form, whereqs
among the "more prejudiced" subjects the distributions of responses
to the two forms were essentially the same. This finding led Seeman
to question what was actually being measured by the indirect test and
by the Likert scale, and which one was more valid for what purposes.
He concluded by pointing to the "need for extreme care in interpreta-
tion of projective and semi-projective techniques for the study of
specific attitudes."
Other investigators have employed changes in judgment of literary
merit, changes in evaluation of the quality of mottoes, and changes
in level of aspiration as indicators of attitudes. The general technique
is always essentially the same. The item to be evaluated is presented
as a product or as a characteristic of a given person or group, and the
same or an equivalent item as originating with a different person Qr
group. The discrepancy in judgment is taken as a measure of the
attitude toward the given group. Note, however, that the discrepancies
in judgment are more complex than may appear on the surface, since
changes in meaning go along with changes in imputed origin. 13
Hovland and Sherif (1952) have suggested another way in which
judgments might be used as an indication of attitude. In a study dis-
cussed in more detail in Chapter 10, these investigators found that
people's judgment of the degree of favorableness or unfavorableness
of a given statement about Negroes was influenced by their own at-
titude toward Negroes. This finding led to the suggestion that at_titude
might be assessed indirectly through study of the wayan individual
I
places items in terms of his judgment of their favorableness or un-

favorableness toward the object. Individuals with strongly "pro" att'i-
tudes would be expected to perceive relatively many items as unfavor-
able; those with strongly "anti" attitudes to perceive relatively many
as favorable; and those with neutral attitudes 'to see the items as being
rather evenly spaced over the entire range. Hovland and Sherif point
out that such a test should include a large number of ambiguous items,
since it is the judgment of these items that is most affe~ted by the
individual's attitude.
J~st as the perception of other types of stimuli may be affected
13 See, for example, Lewis (1941).
by attitudes, so may perception of social facts. Saenger and Birch,
for example (cited in Saenger and Proshansky, 1950), asked respond-
ents to estimate the number of Communists in the country; they found
that the average figure given by Republicans was five times as great
as that given by Democrats. Merton et a1. (forthcoming) found that,
in a mixed housing project, white residents who were prejudiced
against Negroes overestimated the proportion of Negroes in the proj-
ect and underestimated the average amount of education the Negro
residents of the project had had.
Many people tend to overestimate the size of the group that
agrees with their opinions. For example, in a study of attitudes toward
Negro sales personnel in department stores, Saenger and Gilbert
(1950) found a striking correspondence between subjects' own atti-
tudes, as rated on the basis of their replies to a number of interview
questions, and their estimates of the proportion of people who would
object to the hiring of Negro sales clerks. Of those who were rated as
showing no prejudice, only 13 per cent said "most people" would be
opposed to the employment of Negroes, while 52 per cent said "a few"
or "none." At the other extreme, of those who were placed at the most
prejudiced end of the rating scale, 92 per cent said "most people"
would object to the employment of Negroes, and none said "a few" or
"none" would object. Findings such as these h::t:ve led several investi-
gators to attempt to measure social attitudes indirectly by asking sub-
jects to estimate the percentage of a group that would agree with
specific opinions.
Although such "estimating" techniques have been used success-
fully in showing that groups with different attitudes differ in their
perception of social facts, as measures of attitude they must be used
with extreme caution. First of all, in order to determine the extent
and direction of dis~ortion, the investigator must have some independ-
ent measure of group composition or opinion. -Second, different indi-
viduals within the; group may have had different experiences that
provide objectively :different bases for their estimates. For example, if,
in Merton's mixed 'housing project, the proportions of Negroes were
different in different
I,
buildings, tenants in buildings with relatively
many Negroes might be expected to overestimate the proportion of
Negroes in the total project simply on the basis of their everyday ob-
servation of the people around them. Third, in the absence of firm

theoretical or empirical support, it is frequently not clear whether
distortion in a given direction reflects a favorable or an unfavorable
attitude. For example, Saenger and Birch, in addition to finding dif-
ferent estimates of the number of Communists, found that individuals
were likely to exaggerate the size of their own group. Thus, people
who described themselves as "liberal Republicans" gave higher esti-
mates of the size of this group than did people who described them-
selves as "conservative Republicans," and so on. In other words, dis-
tortion in estimation may reflect either wishes or fears; not enough
research has been done to know the conditions under which distortion
indicates one or the other.
Substitute Measures
Still another indirect approach to the measurement of a character-
istic involves measuring something else, or sQme combination of other
things, that is sufficiently highly correlated with the characteristic one
wants measured to enable it to serve as a satisfactory substitute. We
may call this approach the substitute measure.
The F-scal~ (Adorno et al., 1950) was constructed on this prin-
ciple. It was reasoned that authoritarian attitudes should correlate
highly with anti-semitism, so that a measure of the former should also
yield a satisfactory measure of the latter. To increase the substitut-
ability of the F-scale for a measure of anti-semitism, items were selected
for inclusion in the F-scale not only on the basis of the criterion ;of
internal consistency, but also on the basis of how well they correlated
with-a scale of anti-semitic attitudes.
Wilner, Walkley, and Cook (1955) used a different type of sub-
stitute meastire in their effort to establish post hoc whether two groups
of white tenants in housing projects (one living relatively close to,
Negroes in the project, the other relatively far from them) had held
similar attitudes qefore moving into the project. Working from the
fact that a number of socioeconomic characteristics (for example, re'
ligion and education) are known to be correlated with attitudes toward
SUBSTITUTE MEASURES 311
Negroes, they constructed an index of "probable initial attitude" on
the basis of such characteristics.
Ideally, the, method of substitute measurement calls for the com-
bination of a number of indices, each of which has a relatively high
correlation with the characteristic for which one needs a measure, and
a relatively low correlation with each of the other indices in the com-
bination. In principle, such a measure involves an application of the
logic of pragmatic validation. It should be noted, however, that such
a measure cannot be used as an "after" measure in an attitude-change
experiment. The experimental factor is designed to change the attitude,
not necessarily the variables which were initially correlated with it.
If successful, the experimental factor may change the basis of the rela-
tionship and, among other things, lower the pragmatic validity of the
substitute measure. 14
ANote on Validation
As we indicated earlier, many questions have been raised about

the validity of indirect techniques, and relatively little research evi-
dence is available to answer them. Actually, as noted in the preceding
chapter, there is not much evidence of the validity of direct techniques
depending on self-report, such as interviews and questionnaires. The
validity of such instruments is less often questioned, however, probably
because of the "obvious" relevance of the questions to the character-
istics they are intended to measure. It is the degree of inference in-
volved in indirect tests-the gap between the subject's response and
the characteristic it is presumed to indicate-that intensifies ques-
tions about validity.
In the case ot projective techniques, there are additional reasons
for questions about validity: the great variety of aspects about which
inferences may bb drawn and the heavy reliance on the interpretive
14 This conside}ation points to the grave weakness of pragmatic validation
without support in a: body of theory or knowledge of the conditions on which the
correlation between test and criterion depends. Under these circumstances, we never
know whether the underlying conditions may not have changed so that the test
no longer has the pragmatic validity we think it has. That is, we can only hope
that since the test has worked until now, it will continue to work in any giveD
application.
skill of the individual analyst. For most of the projective techniques,

the rules by which the data are to be transformed into scales of meas-
urement are not specified in detail; nor, for that matter, are the di-
mensions one is attempting to measure. As a consequence, each investi-
gator is forced to some extent to develop his own rules. Moreover,
the very flexibility of the tests means that they do not always cover
the same aspects in the same detail. In other words, at least in their
present form, projective tests sacrifice precision and reliability in the
interest of breadth and depth.
This variability both in the detail with which various aspects are
covered and in the interpretations made by different users introduces
a question of the extent to which one can speak of validating "the
Rorschach" or "the T.A.T." Rather, specific inferences made by specific
interpreters using these techniques can be validated. Cumulation of
evidence of validity in specific respects contributes, of course, to vali-
dation of the technique as a whole.
To a considerable extent, projective te'tllniques intended to meas-
ure broad aspects of personality must rely on_ construct validation (see
Chapter 5) rather than on evidence of high correlation with any single
criterion. All studies in which scores or qualitative descriptions based
on projective techniques are found to be related to some other variable
-and there are many such-contribute to the construct validation of
the given technique. Thus, the finding in the Authoritarian Personality
study that the T.A.T. stories of prejudiced and unprejudiced individ-
uals tended to differ in specified ways helps not only to confi~m the
basic hypothesis of the study (tha~ prejudice is a fundion of deep-
lying personality trends) but to validate the T.A.T. as an instrument
for uncovering such personality trends.
In estimating the validity of indirect measures of social attitudes,
both those of the projective type and the more structured tests, an
investigator may administer the test to two or more groups that may
be expected to differ in the characteristic the test attempts to measure,
and compare their scores. If the test differentiates between the groups,
this is contributory evidence of its validity. Thus, Hammond's fiI}ding
that there was almost no overlap between scores of union members and
6f members of business clubs on his "information test" provides some
assurance that the test is a valid measure of attitudes toward labor-man-
A NOTE ON VALIDATION 313
agement relations. A number of other investigators have used this
approach.
Or the user of an indirect test may attempt to validate it by com-
paring scares on it with those on a non-disguised test designed to meas-
ure the same attitude. Proshansky used this procedure with respect to
his projective test of attitude toward labor, with positive results. See-
man used it, with respect to his indirect "moral judgment" test of
attitudes toward Negroes, with negative results. It has been used by
a number of other investigators in estimating the validity of an in-
direct technique.
This use of an undisguised test as a criterion may seem a rather
odd procedure, considering the fact that the major reason for using
indirect techniques is to get at information that might not be revealed
in response to a direct technique. However, two points should be con-
sidered in this connection. One is that an indirect technique may be
used simply as a more effective way of getting information that the
individual could give in response to a direct question if he were willing
to. When this is the case, it is appropriate to use responses to a non-
disguised instrument as a criterion for the validity of the indirect tech-
nique, provided that the non-disguised test is administered under cir-
cumstances that make it reasonable to expect minimal concealment of
true attitudes. For example, it would be reason,able to use scores on a
non-disguised test of attitudes toward labor as a criterion against which
to compare the results of an indirect test only if the non-disguised test
were administered under circumstances that assured the subjects of
anonymity or if the rapport between investigator and subjects were
such that they would feel free to reply frankly; it would,not be reason-
able to use as a criterion scores on a test administered to workers by
their supervisor. ;
Another point in evaluating the use of a non-disguised test as a
I
criterion for a disguised one is that the purpose of the disguised test
may be to gain more extensive insight into the nature of an individual's
attitude. This mqre extensive knowledge may not mean a different
estimate of how favorable or unfavorable an individual is toward a
given object, but rather
I,
some understanding of how his attitude is re-
lated to other fac,tors. It is not unreasonable to assume that if a pro-
jective technique:gives evidence of measuring some aspect of an atti-
tude validly (as indicated by its consistency with the results of a vali-
dated undisguised technique), it may also measure other aspects of an
attitude validly; nevertheless, the consistency should not be taken as
conclusive evidence of validity in measuring these other aspects.
Another approach to validation is to compare the results obtained
by a measuring instrument with observations of actual behavior. That
this approach has not often been used is undoubtedly due in part to the
difficulty of determining what kinds ~ehavior in what situations
would provide an adequate criterion, in part to the difficulty of. securing
measures of such behavior. Nevertheless, a few of the studies mentioned
in this chapter have used this approach. Haire used it in the instant
coffee study, and found high correspondence between responses to his
indirect measure and actual behavior. Hogrefe, Evans, and Chein, in
their study of relations between white and Negro children, found high
correspondence between scores on their indirect measure and chil-
dren's reports of their own behavior, but little correspondence with
observation of actual behavior in test situations.
In summary: Much more investigation of the validity of indirect
tests is needed. Some of the studies carri~a out to date have given
encouraging evidence of correspondence between the results of an in-
direct test and those provided by an independent criterion; others,
however, have revealed discrepancies between different meaSj.lres that
raise questions about what the various tests are in fact measuring. In-
vestigation of the validity of indirect tests-and especially of projective
tests which attempt to measure more than one dimension-is hampered
by the difficulty of finding appropriate criteria. Nevertheless, more
attention needs to be paid to validation of tests of this type before
they can make their full contribution to social research.
9
THE USE OF AVAILABLE DATA
AS SOURCE MATERIAL
Statistical Records!'
Personal Documents
, Mass Communications
Summary
Statistical Records
A CONSIDERABLE AMOUNT of statistical data on the behavior of its

members is available in every literate community. Although
~hese data have been accumulated primarily for purposes of adminis-
tration and historical description, social science research can make
good use ofl them. To neglect their existence often involves either a
disregard of relevant information or, if the investigator laboriously col- .
lects data that already exist, a waste of effort.
The range of subject matter covered in available records, and the
treatment a subject receives in such records, varies with the adminis-
trative needs for which they were originally collected. Many available.
statistical data refer to socioeconomic attributes of individuals. Thus,
the census of a population contains information about age, sex, family
size, occupation, residence, etc. Health statistics give birth and death
rates and the like; federal, state, municipal, and private economic in-
stitutions collect and publish data on wages, hours of work, produc-
tivity: absenteeism, strikes, financial transactions, and so on. Many vol-
untary, organizations have records not only of their own membership
but of groups ot' people whom they serve. In addition, a small but
steadily increasing body of data is being collected by various institu-
tions on the psychological level proper. For example, school~ hospitals,
social service agencies, pers~nnel departments in industry, and similar
institutions nowadays frequently administer psychological tests of
various kinds to their entire populations.
Data collected in the course of such other activities have a number
of advantages in social research, in addition to that of economy. A
major one is the fact that much information of this sort is collected
periodically, thus making possible the establishment of trends oyer
time. Another is that the gathering of information from such sources
does not require the cooperation of the individuals about whom in-
formation is being sought, as does the use of questionnaires, interviews,
projective techniques, and, frequently, observation. ,Moreover, since
316
STATISTICAL RECORDS 317
such data are collected in the ordinary course of events, the measure-
ment procedure is less likely to reveal the inv,estigator's purpose or t~
change the behavior in which he is interested than are some of the
other data-collection techniques.
GUIDING PRINCIPLES FOR USING STATISTICAL RECORDS
The use of available statistical records requires that the social

scientist be familiar with the better known sources of such data and
that he display some ingenuity in discovering less well known material.
The study by Kenesaw M. Landis, Segregation in Washington
(1948), is based exclusively on the analysis and interpr(;tation of avail-
able data. Among other sources, Landis used publications by the Bu-
reau of the Census'to indicate the pressure on Negroes to live together
in great numbers in a small area and to illustrate the poor conditions
of housing available for Negroes; official health statistics to demon-
strate the consequences of these conditions in terms of a higher death
rate, especially from tuberculosis; official employment data and records
from a private industrial concern to point up discrimination in work;
and the city-wide figures collected by the Department, of Research of
the Washington Council of Social Agencies to illustrate the relation
between overcrowding and the arrests of juveniles llJade by the metro-
politan police. Landis' contribution in this study consisted in discover-
ing the sources of these data and in extracting from them information
relcvcrnt to his topic.
Other research questions demand a greater investment of energy
or ingenuiJ;y in obtaining relevant records. Leo Srole, while at work
.on the study of status ,and prestige in the Yankee City series of invest-
igations (Warner et al., 1941-1947), discovered an unusual source of
data relevant to his problem-cemetery records.! Apparently it had
become the practice iiI Yankee City for residents who had risen to a
social status higher than that of their dead relatives to have the remains
of their family memBers transferred from a lower-status cemetery to
one of higher social standing.
These examples',could easily be multiplied. They demonstrate
! Reported in a perspnal communication to the authors.
I
318 AVAILABLE DATA AS SOURCE MATERIAL
I
that systematic searching and ingenuity will uncover many more avail-
able data than are often assumed to exist.
Another requirement for the use of statistical records has to do
with the formulation of research hypotheses. In essence, the use of
such data demands a capacity to ask many different questions related to
the research problem. By definition, the purpose for which available
records have been collected is different from the purpose for which the
social scientist wishes to use them. (If that were not the case, resea~ch
would be superfluous, since the records would answer the research
questions.) A research problem can, as a rule, be translated into a
wide variety of questions; a hypothesis can be verified in many different
ways. The guiding principle, then, for the use of available statistics
consists in keeping oneself flexible with respect to the form in which
research questions are asked. If a research idea or hypothesis can be
formulated in such a manner that the available recorded material bears
on the question, the use of such material becomes possible.
Perhaps the outstanding example of how the superior flexibility
of a great intellect resulted in the testing'ofa social theory by available
statistics is Durkheim's Le Suicide. As Talcott Parsons (1937) said
about Durkheim, he "possessed to a remarkable degree the faculty of
persistence in thinking through the consequences of a few fundamental
assumptions . . . his empirical observation is of the nature of the
crucial experiment. . . ." Durkheim's basic concern was with the
relation of an individual to a group and to the prevailing norms and
values in the group. He started with the hypothesis that the causes
of suicide are to be found in social conditions. To test this theory, he
studied records of suicide rates in all European countries for which
they were available. Some of these statistics had already been collated
by other investigators; others w~re readily available in public doc-
uments; still others Durkheim compiled from official files. With, great
ingenuity, he examined these statistics for their bearing on a number
of different hypotheses. In his book, he examines first a number of
alternative hypotheses: that suicide is the result of psychopathic states,
of imitation, of racial or hereditary factors, of cosmic factors. He dem-
onstrates that the statistics are not in accord with any of these hypothe-
'les. For example, in considering the hypothesis that suicide is influ-
enced by climatic factors, he starts with the observation that in all
the countries for which statistics are available over a period of years,
the incidence of suicide increases regularly from January until about
June and then declines until the end of the year. This observation had
led other writers to conclude that temperature has a direct effect on
the tendency to suicide. Durkheim examines this possibility in great
detail and demonstrates that the data do not support it. He argues,
for example, that if temperature were the basic cause, suicide would
vary regularly with it; but this is not the case. There are more suicides
in spring than in autumn, although the temperature is slightly lower
in spring. Moreover, suicide reaches its height not in the hottest
months (July and August), but in June. By a series of such analyses, he
demonstrates that the seasonal regularities in suicide rates cannot be
accounted for by temperature, and suggests the alternative hypothesis
that social activity is seasonal and that the rate of suicide is related to
the extent of social activity.
Turning to the hypothesis that a basic cause of suicide is lack of
integration into a social group, he examines three major social influ-
en,ces: religion, family, and political atmosphere. He finds suicide
rates lower among Catholics than among Protestants, lower among
married than among single people, lower among those with children
than without, lower during periods of national/fervor. All of these
findings, he argues, support the hypothesis that belonging to a co-
hesive social group is a deterrent to suicide. As he analyzed the data
in more and more detail, he continuously modified and refined his
theory.
Some studies, such as Durkheim's research on suicide and Landis'
account of segregation in Washington, rely entirely on the analysis of
data collected for pu~poses other than those of the particular study. In
l
others, such data are used in conjunction with other procedures. Data
regularly collected f9r other purposes may be used to measure the
effects of an experimental treatment. For example, the effects of an
I
election appeal may be measured by a study of voting rcords, the effects
of various personnel procedures by records of productivity. Thus, in
the "Hawthorne stJdies," Roethlisberger and Dickson (1939) found
that changes in suh conditions as illumination, rest periods, and
hours of work could not account for a consistently rising rate of produc-
tivity in their experimental groups over a period as long as one year;
they concluded that changes in the social organization of the 'work
groups and in their relationship to management were responsible for
the rise in productivity.
Available data may be used at other points in a study. They are
frequently helpful in selecting cases with specified characteristics for
intensive study, or a random sample for interviewing in a survey. A
study of worker morale in war industries by Katz and Hyman (1947)
illustrates both these uses of available data. First, production records
were used as a basis for selecting five shipyards which differed in pro-
ductivity; within each of these yards, a sample of workers to be inter-
viewed was selected by taking every nth name from the payroll lists.
These investigations found a circular relation between morale and
production, with high production giving a feeling of accomplish!llent
which led to increased effort, while low production reduced motivation,
which in turn reduced productivity. They concluded further that fac-
tors directly associated with the job we{e more important determi-
nants of worker morale than mOre general community conditions such
as housing, transportation, and recreational facilities.
Available records may also be used to supplement or to check
information gathered specifically for the purposes of a given investi-
gation. For example, in a study of the psychological impact of long-term
unemployment in an Austrian village (Jahoda-Lazarsfeld and Zeisl,
1932), the ~ccounts of their experience given by sev~ral unemployed
men suggested that they felt much worse at the onset of unemploy-
ment than after three years, in spite of the gradual deterioration of
their economic condition. This "shock" effect of unemployment was
checked against such records as the accounts of the local grocer, which
showed a s,udden drop in sales in the months immediately after the
onset of unemployment, followed by a slight recovery and a, steady
decline thereafter.
In many of these examples, the investigator's interest ~as in be-
havior or characteristics of the sort directly reflected in the statistical
records-segreg~tion, suicide, voting, productivity. Like other types
of data, however, records of specific behavior n:ay be used as an indi-
\
STA TISTICAL RECORDS 321
cator of some more general concept. A series of studies by Tryon (1955)
illustrates both these uses of existing data. Tryon was concerned wifh
the problem of identifying subcultural groups in more meaningful and
reliable ways than the usual ratings of "social class." He was interested
in two related hypotheses: (I) that demographic social areas can be
identified on the basis of census data; and (2) that a demographic
social area is also a psychosocial area-that is, that residents of a com-
mon demographic social area will experience certain common socially
relevant situations and certain common psychological states elicited by
those situations, and will behave in certain common ways. In connec-
tion with the first hypothesis, he examined thirty-three items in the
1940 U. S. Census for the 243 census tracts in the San Francisco Bay
Area; the items included, for example, the percentage of detached
single-family homes, the percentage of women not in the labor force,
the percentage of managerial or professional workers, etc. By a statis..
tical technique known as cluster analysis,2 he found that these thirty.
three items fell into three main groupings, which could be described
as: socioeconomic independence, having to do with wealth and social
independence; assimilation, or the incorporation of persons into stand-
ard white-collar American culture; and orientation around the family.
Each of the census tracts could be described in terms of its position on
each of these three basic variables. Many tractsl of course, showed
the same pattern; for the entire area, eight basic patterns were found.
These could be given descriptive labels as well as index scores; for
example, "the exclusives"-above average in assimilation, in orientation
around family life, and especially in socioeconomic independence;
"the workers"-average in orientation around family life but some-
what below average in assimilation and socioeconomic independence;
"the segregated"-lqw in assimilation, socioeconomic independence,
and family life as defined in the study.
As evidence to tbt
I
his second hypothesis-that demographic areas
are also psychosocial areas-Tryon used voting records. His interest
was not in voting per se, but as an indicator of social attitudes. He
found a high correspondence between demographic pattern and voting
in the 1940 presideptial election. In most of the "exclusive" tracts,
about a quarter of the votes were cast for Roosevelt, and in none were
2 For procedures invo1ved ill duster ana1ysis, see Tyron (1957a).
more than a third for him; at the other extreme, in most of the "worker"
and "segregated" tracts, three quarters Or more of the votes went to
Roosevelt, and in none did he get less than half the votes. Analysis (as
yet unpublished). of election results fourteen years later showed a
continued relationship between demographic pattern (as determined
in 1940) and social attitudes as revealed in 1954 votes on such issues
as bonds for a hospital, tax exemption for welfare institutions, and pen-
sions for needy aged.
SOME PITFALLS IN USING STATISTICAL RECORDS
THE DEFINITION OF TERMS. The definitions of categories used in

available statistical material frequently do not coincide with those used
in social research, a fact often hidden by the use of similar terminology.
For example, the social scientist may be interested in family composi-
tion. When he consults census reports, he may find material under the
category "household composition." Alt~ough family composition in-
cludes blood relations only, household composition extends to lodgers,
servants, and other employees who may share the residence of their
employer.
An even more striking invitation to confusion is provided by some
statistics on criminal offenses. For example, in all the forty-eight states
of the United States, a distinction is made between a felony and a
misdemeanor; however, an act which is a felony in one state may bt1
classified as a misdemeanor in another.
In view of such differences, Ithe use of available records may be
more misleading than enlightening unless the precise definition on
which the statistics are based is known.
METHODS OF DATA COLLECTION. To know what the original col-
lector of available data set out to gather is not enough; one must also
inquire into the adequacy of his methods. Many records are collected
with the intention of covering an entire universe of events, and not
merely a sample. However, many obstacles stand in the way of realizing
this ideal. First, the informants from whom the original collecting
agency drew the information may not have been willing or able to
provide it. There is a well-grounded suspicion, for example, that in-
come statistics based on individual tax declarations tend to be under-
estimates, whereas expense accounts tend to be overestimates of ex-
penditures. For equally understandable reasons, statistics concerning
illegitimacy are somewhat less accurate than those on births in wed-
lock.
It is quite possible that the degree of inaccuracy in official records
due to these reasons is negligible from the point of view of the social
scientist's purposes. But there are other methodological errors that
may lead to serious inaccuracies. This should be kept in mind, espe-
cially when using data collected over many years. For example: The
Prussian suicide statistics (see Halbwachs, 1930) go back to 1816.
Until 1883 the duty of keeping suicide records was the responsibility
of the local police. Then, in 1883, the task was transferred to the civil
service. In view of this change, one would hesitate to interpret the fact
that official Prussian suicide figures show an increase of 20 per cent
between 1882 and 1883 as anything but a reflection of the change in
techniques.
, Occasionally, it is possible to correct available-records in the light
of what is known about the methods by which they have been gathered.
More often, this is not possible. In any case, the proper qualifications
of such data when used for research purposes can be made only if the
social scientist is aware of the possible errors inhe~ent in the particular
method employed.
Personal Documents
In speaking of personal documents for research purposes we refer

to autobiographies, letters, diaries, certain types of school essay, prize
essay, and the like.IOther authors 3 have used the term in a much
broader way to incl6de such additional material as questionnaire data,
interviews, art form~, projective productions, records of social agencies,
etc. In our limited usage we refer only to items that meet the following
criteria: They are' ~1) written documents; (2) documents that have
3 See especially t~e very illuminating publications by the Social Science Re
.earch Council: Allport (1942), and Gotbchalk, kiuckhuhn, and Angell (1945).
been produced on the writer's own initiative or, if not, in such a way
that their introspective content has been determined entirely by the
author; and (3) documents that focus on their author's personal ex-
periences. These criteria exclude interview material, however informal
the interview sittiation may have been. They exclude also those literary
efforts that can be used as personal documents only through projective
interpretation. This more limited definition of personal documents
has the advantage of bringing to the.fore their most distinctive charac-
teristic: they permit us to see other people as they see themselves.
Augustine, who produced one of the greatest personal documents
of all times, fully realized the uni~1 contribution they can make in
this respect. In Book X of his Cop{._essions he explains why he wrote
this personal document. He starts fro the assumption that "men
are a race curious to know of other men's lives," an assumption which
is as valid now as then; and he argues that othing but what a man says
about himself can fully satisfy this curiosity:
As to what I now am while I am writing my Confessions,
there are many who desire to know-both people who know me
personally, and people who do not, but have heard from me or
about me. Yet they have not their ear at my heart, where I ,am
what I am. They wish, therefore, to hear from my own confes-
sion what I am inwardly where they cannot pierce with eye or
ear or mind [italics suppliedJ.
It is true that the social sciences have developed modern tech-
niques that aim at piercing through outward appearance and behavior
to inner experiences. Depth interviews, projective techniques, and
psychoanalysis aim at just this. Successfully applied, they can often
penetrate even beyond what a man knows of himself.4 But these tech-
niques, although they are able to 'discover the nature of selected inner
experiences; can hardly ever reconstruct the entire structure of a per-
son's self-image, with its spontaneous emphases and complexities.
Gordon Allport (1942), in his defense of the value of personal
documents for psychology, stresses the importance of aiming for a view
4 Augustine realized the limitation of personal documents in this respect: " ...
yet there is something of man that the very spirit of man that is in him does not
know."

PERSONAL DOCUMENTS 325
of the whole before details of inner experience are subjected to mon
systematic scrutiny:
Acquaintance with particulars is the beginning of all knowl-
edge-scientific or otherwise. In psychology the font and origin
of our curiosity in, and knowledge of, human nature lies in our
acquaintance with concrete individuals. To know them in their
natural complexity is an essential first step. Starting too soon with
analysis and classification, we run the risk of tearing mental life
into fragments and beginning with false cleavages that misrepre-
sent the salient organizations and natural integrations in personal
life. In order to avoid such hasty preoccupations with unnatural
segments and false abstractions, psychology needs to concern
itself with life as it is lived, with significant total processes of the
sort revealed in consecutive and complete life documents.
By and large, the rationale for the use of personal documents is
similar to that for the use of observational techniques. What the latter
may achieve for overt behavior, the former can do for inner experiences:
to reveal to the social scientist life as it is lived without the interference
of research. However, although the number of situations that can be
observed is considerable, personal documents are relatively rare; hence
the scope of their usefulness for research is rather limited.
Even when they are available, personal documents have to be used
with some caution. Augustine pointed to one of the basic reservations
-the doubt about authenticity: "And when they hear me confessing
of myself, how do they know whether I speak the truth . . . ?" He
saw clearly that there was no completely satisfactory answer to such
skepticism, for "I cannot prove to them that my confession is true."
There are two possible kinds of falsification with which a social
scientist who uses personal documents has to be concerned. In its
crudest form, falsificat~on amounts to conscious, deliberate deceit. A
document can be prod~ced in the form and manner of a personal docu-
ment by someone else' and be presented to the world as the genuine
article. The motives fbr such falsification are various: material gain,
malice, practical joke, literary exercise. The most outstanding example
in the psychological literature was an extremely skillfully presented
falsification of the diary of an adolescent girl, which deceived, even
Freud, who called it a "gem." Hug-Helmuth, who originated this
Tagebuch eines halbwiichsigen Miidchens,5 asserted that it was the

diary of a young girl and that it had not been edited or altered. It took
several years for this falsification, which was gaining wide recognition
as a demonstration of the development of sexual consciousness, to be
unmasked for what it was.
The incident is worth noting because of the manner in which it
was finally proved to be a falsification. Critics had asserted that the
style of the document was too mature to be credited to a girl between
the ages of 11 and 14 years. Yet the content was not exaggerated, and
was similar enough to much that enters authentic diaries of early
adolescents. Finally, however, a check on various casual references to
the weath~r on given days, the mention of a visit to a place that was
nonexistent at the date on which the entry appeared, and similar small
inconsistencies convinced the psychological world that the diary was
not authentic. Historians, who are perhaps more often exposed to
deliberate falsifications than other social scientists, have developed
techniques of detection. similar to, those mentioned above which
might well be applied whenever the use of personal documents is
considered (see Gottschalk et al., 1945).
However, Augustine, when he raised the question as to the truth-
fulness of a personal document, obviously had in mind a less crude
form of falsification. He was concerned not with deception about
the authorship but with the possibility of the author's misrepresenting
what he knew about himself. If such misrepresentation is due to self-
deception, it does not impair the value of the document. If one is in-
terested in the author's self-image, the question of whether or not the
self-image agrees with the image other persons have of the author is of
secondary importance. To be sure, a confrontation of self-image and
images that others have of a person would be highly illuminating. Yet
even if a discrepancy appears between two such descriptions,. the deci-
sion as to which is right and which is wrong cannot easily be made.
There is still another possible reason for an author's mi~representa
tion of himself in a personal document. All personal documents are
produced for. some purpose. Letters are written to communicate with
someone else; school essays are submitted to teachers, prize essays to
5 Translated into English by E. and C. Paul as A Young Girl's Diary (1921).

judges; the writers of even the most intimate diaries probably keep an
eye on a potential future reader. Augustine wanted to "stir the hearts
of the sinners and the just"-the sinners to find courage from his
Confessions that a way of life can be changed, the just to rejoice in his
conversion. Many autobiographical statements are produced for pur-
poses such as this-in cruder form, for propaganda purposes. The social
scientist using personal documents will do well, whenever possible,
to inquire into the motives that induced their production.
The disadvantages of using personal documents in research have
been amply discussed in the literature (see especially Allport, 1942),
particularly in the prolonged debate about the scientific value of
Thomas and Znaniecki's The Polish Peasant in Europe and America
(1918), which relied heavily on letters and diaries for data. The use
of personal documents has been criticized on the grounds that they
are rarely suited for treatment by statistical techniques; that their
validity is hardly ever beyond doubt; that they can be the result of
deception or self-deception; that they are subject to errors of memory
and are at the mercy of passing moods. In addition, they are by no
means easy to come by.
By and large, these arguments carry weight. It should be pointed
out, however, that the applicability of statistical techniques is not
determined by the nature of the data but rather by the type of ques-
tion with which the social scientist approaches his data. Unique as
Augustine's Confessions and other autobiographical writings are, they
can be treated by quantitative content-analysis techniques if such pro-
cedure is in accord with the research problem.
Two types of personal document have added considerably to our
knowledge of inner experiences: (1) descriptions of rare and extraor-
dinary events in human life, and (2) diaries and letters dealing with
I
the inner aspect of m.ore frequent and ordinary events.
In 1907, the first edition of C. W. Beers' autobiography appeared.
This book, written soon after the author recovered from a manic-
I
depressive breakdown, was largely responsible for the foundation of

the mental hygiene movement in the United States. It stands as a
model for se1f-descri~tions of mental states, not easily obtained by any
other method. Helen Keller's famous book on her conquest of life as
l
a blind deaf-mute, and the Confessions, also belong to the first group.
Among the most frequently used documents of the second type
are diaries of adolescents. Charlotte Buhler (1934), in developing
her psychology of adolescence, conducted and sponsored a series of
interrelated studies based on the use of such diaries. Buhler's interest
led to the establishment of a collection of diaries of adolescents at the
Psychological Institute of the University of Vienna; in a relatively
short time almost one hundred specimens were assembled. Their avail-
ability in such number permitted a more systematic comparison than
is ordinarily possible with unique documents.
At least two of Buhler's studies involved comparison of individuals
at very different periods in time-something difficult to achieve by
other methods of data collection. One study was based on three diaries
of girls from three successive generations. Buhler demonstrated that in
spite of the considerable cultural change between 1873 and 1910 (the
years in which the oldest and the youngest of the three diary writers
were born), some basic desires of adolescence, such as the need for
intimate personal relationships, remained the same. Yet in other re-
spects, such as the girls' relations to their parents, cultural changes were
reflected in the diaries.
The second study was based on diaries of two girls of the same
generation who,. at _the time of the study, were about twenty years
older than at the time they had produced the diaries. A comparison be-
tween "then" and "now" revealed considerable similarities between
the two girls during adolescence and considerable _differences in later
life. t
Buhler's studie~ illustrate how a considerable gap in time can be

bridged by the use of personal docume.nts. Occasionally it becomes
possible to use personal documents, especially letters, to bridge the gap
in space ,which often separates the investigator from his subjects. Sev-
eral studies have used the mail received by political representatives as
a basis for gauging the climate of political opinion among the letter
writers; others have used the fan mail of film, radio and -television
stars in attef!1pts to analyze their attractiveness to the ,public. The
study of such mail, is, of course, not free from ambiguities. Letters
vary in length, in information about their a~thors, and in the degree
of their apparent spontaneity. Nevertheless, they often provide the
only feasible approach to an otherwise unreachable group of the pop-
ulation.
It must be kept in mind, however, that the study" of personal
documents permits generalization only to the universe of document-
producers, not to the population at large. The inner life of individuals
who keep diaries may be different in important respects from that of
individuals who do not; the opinions of constituents who write to their
Congressmen may be quite unrepresentative of the opinions of other
citizens. It is one thing to study personal documents for the sake of
understanding the particular individuals who produced them, or for
clues about psychological processes that may be common to other in-
dividuals. It is quite another matter to assume that such documents
provide a basis for conclusions about persons other than their writers.
A striking example of this point was provided in the New York Times
of March 17, 1957, under the headline, "5 Surveys Dispute Mail Oppos-
ing Aid." The story reported that analysis of Congressional mail had
led to the belief that public opinion in the United States was strongly
opposed to the continuation of foreign aid; "most" of the letters re-
ceived on this subject recommended outright elimination or heavy
cuts. However, five nation-wide surveys carried oyt by the National
Opinion Research Center, from January 1956 to June 1956, using a
variety of question wordings, all found substantial proportions of re-
spondents favoring the continuation of foreign aid. The proportions
e;xpressing approval varied with the specific question wording. On the
question of economic aid "to some countries like India, which have
not joined us as allies against the Communists," opinion was about
evenly divided; the proportions favoring aid to such nations ranged
from 43 per cent to p2 per cent at different times. However, on the
question of continuing "economic aid to countries which have agreed
I
to stand with us against Communist aggression," as many as 90 per
cent of the respondents expressed support. It should be noted that
such survey data do not discredit the study of Congressional mail; they
supplement it, as th~y are supplemented by the mail analysis.
These examples reveal both the values and the limitations of
personal documents as sources of data. They suggest that better use is
made of the peculiar nature of data contained in personal documents

when they are employed for exploratory and descriptive rather than
statistical purposes; for the development of insights and for illustra-
tions rather than the verification or refutation of hypotheses; in con-
junction with other methods rather than by themselves.
Mass Communications 6
In addition to statistical records and autobiographical documents,
every literate society produces a variety of material intended to inform,
entertain, or persuade the populace. Such material may appear in the
form of literary productions, newspapers, and magazines or, more re-
cently, motion pictures and radio or television broadcasts.
Mass communication documents are not produced for the bene-
fit of the investigator, and in this respect (although not in others) are
free from the influence of his theoreti~al or personal bias. Like avail-
able statistical records, they enable one to deal with the historical past
as w~llas with contemporary society, an advantage that can hardly be
overestimated in view of the considerable methodological difficulties
stari'ding in the way of a historical perspective in social science. Even
mote than statistical records, documents of mass communication re-
flect broad aspects of the social climate in which they are produced.
PURPOSES OF ANALYSIS
Mass communications provide a rich source of data for investigat-

ing a variety of research questions. They may be used to throw light
on some ~spect of the culture ~f a given group, to compare different
groups in terms of some aspect of culture, to trace cultural change.
For example, Lowenthal (1943), in a study of cultural changes in
American society in the course of the twentieth century, analyzed
biographies appearing in popular magazines from the beginning of the
6 In this section we have drawn heavily on McGranahan (1951); in fact, some
of the material is taken verbatim from that source, with Dr. McGranahan's permis-
sion. .
MASS COMMUNICATIONS 331
century until 1941 in terms of the profession of their subject. ThE.
following table summarizes some of his main findings:
PERCENTAGE DISTRIBUTION OF BIOGRAPHIES

ACCORDING TO PROFESSIONS IN "SATURDAY
EVENING POST" AND "COLLIER'S" FOR SELECTED
YEARS BETWEEN 1901 AND 1941
1901-1914 1922-1930 1930-1934 1940-1941
Political life 46% 28% 31% 25%
Business and
professional 28 18 14 20
Entertainment 26 54 55 55
Number of cases 177 395 306 125
Yearly average of
biographies 36 66 77 125
In analyzing further the sphere of entertainment, Lowenthal

points out that in the earliest period covered by his study (1901-)914)
77 per cent of the persons described were engaged in the fine arts; in
the latest period (1940-1941) the corresponding figure was 9 per cent.
In Lowenthal's study the communication content, taken at face
value, was analyzed for clues to cultural changes. Other communication
studies, especially those that aim to throw light! on the entire com-
munication process from sender to receiver, often deal with the ambi-
guity of meaning of communication. An example is the study by Wolf-
enstein and Leites (1950) of an American anti-prejudice motion pic-
ture,No Way Out:
[The film] revolves around the difficulties of a young Negro
interne in a county hospital of a large Northern city. On his
first assignment to the prison ward, he is called upon to treat
a hoodlum who has been wounded in the leg during a robbery.
Led by various Stgns to suspect a brain tumor, the doctor per-
forms a spinal puncture. The patient dies immediately, while
his brother, also wounded and handcuffed to the adjoining bed,
screams: "That nigger killed my brother!" The doctor wishes to
have an autopsy j~O prove that his diagnosis and treatment were
correct. But permission must be obtained from the surviving
brother, the Negro-hater, who has gruesome fantasies about what
they would do to his brother's body and refuses. He gets word

to his friends that his brother has been "murdered" by a Negro,
and they plan to avenge the "murder" by a race riot. The Negro
community is forewarned and by a well-organized strategy sur-
prises and beats its white enemies, but without the participation
of the Negro doctor, who disapproves of. resorting to violence.
The doctor forces the issue of the autopsy by "confessing" to
the "murder" of his patient. The autopsy exonerates him. The
hoodlum, to whom the.physician who has performed the autopsy
gives an explanation of his brother's death, remains unconvinced:
"That's medical double-talk-I tell you I saw him kill my
brother." The hoodlum escapes, lays a trap for the Negro doctor,
and is about to murder him when he is stopped at the last mo-
ment.
It seems"clear that the intention of the producers was to depon-
strate tpe. dangerous irrationality of color prejudice. The task which
the investigators set themselves was tt? view the correspondence be-
tween intent and probable reception. This involved an interpretation
of the major scenes in the light of what is known about irrational
elements in color prejudice and the functioning of the unconscious.
The al?~1]Zation of this frame of reference revealed certain "negative"
elements in the film. To quote from the study:
In the death of the patient under treatment by the Negro
doctor, we are shown one thing and told another. We are shown
that the doctor is unsure of,himself. He has come to the prison
ward for the first time; the guards do not know him, are uncer-
tain whether he should be there, and increasingly puzzled at/the
seemingly irrelevant instruments he requires for what everyone
supposes to be a simple leg wound. Moreover, the do_ctor is al-
ready-visibly shaken by the insults-o~'the patient's brother. What
we see is that an inexperienced, insecure Negro doctor, whose
procedure looks dubious to everyone present, treats a man with
a seemingly minor injury in such a way that he immediately dies.
Like'jVise, the sequence of the race riot contai~s images which
tend to confirm the fantasy of the Negro as a dangerous attacker.
We know that the Negroes in stealing a march on the white gang
who are preparing to attack them are acting in justified self-
defense. But at the moment when the Negroes rush in, we see
standing beside the brutal bully who has been organizing the
white gang a clean-cut young girl who does not appear before
or after in the film. And as the fight begins, there is a close-up
of another white girl, who also has no other part in the film,
screaming in anguish. These two images would seem very likely
to evoke the fantasy of "white womanhood" assaulted by the
"bestial Negroes."
Whether the hypotheses about audience perception of meaning
which are inherent in this interpretation are correct is a question to
be tested by audience research and is outside the scope of analysis of
communication content.
Another type of question that can be answered by analysis of mass
communications has to do with the type of information made available
to the public, or the light in which various issues are presented. For ex-
ample, Davison (1947) took samples of newspapers from the Soviet,
United States, and French zones of Berlin during December 1946. He
analyzed the front-page news items in terms of whether references, to
"United States," "Great Britain," "U.S.S.R.," "France," "United
Nations," "Communist Party," "Pro-Soviet States," "Greece," and
"Iran" were favorable, unfavorable, or neutral. If, for example, the
news item as a whole depicted the United States favorably, as in a
story of United States assistance to starving Europe, then a positive
tally was made for "United States"; if the news item described or
implied, say, United States military imperialism, then a negative tally
was made. Davison's results showed a wide divergence between the
content of the Soviet- and non-Soviet-controlled Berlin press, and
demonstrated the extent to which "news" conforms to the ideas of
those who issue it.
Mass communica~ions have been analyzed also for such purposes
as identifying propagadda techniques or describing the appeals of polit-
I
icalleaders to their followers. For example, R. K. White (1949), in
a study of the values tp which Hitler and Roosevelt appealed in their
public speeches prior 'to the outbreak of World War II, used two
books: My New Oider, English translation by De Sales of Hitler's
speeches, and Nothihg to Fear, selected addresses of Franklin D.
Roosevelt, edited by Zevin. Within each speech White classified every
value statement. His tables are based on 4,077 value statements by
Hitler and 1,249 by Roosevelt. The following table, a simplified version

of that published by White, summarizes his main findings:
PERCENTAGE DISTRIBUTION OF VALUES APPEALED TO

BY HITLER AND ROOSEVELT IN SPEECHES
BEFORE WORLD WAR II
Hitler Roosevelt
Strength values 34.8% 15.2%
Moral values 38.0 28.3
Economic values 10.8 27.7
Other values 16.4 28.8
Total 100.0 100.0
These are only d few examples of the kinds of research questions

that can be investigated by analysis of mass communications: A survey
of the field by Berelson (1952) enumerates the following specific pur-
poses for which communication content has been analyzed:
Questions concerned with characteristics of content
To describe trends in communication content.
To trace the development of scholarship.
To disclose international differences in communication con-
tent.
To compare media or "levels" of communication.
To audit communication content against objectives.
To construct and apply communication standards.
To :aid in technical research operations.
To expose propaganda techniques.
To measure the "readability" of communication materi!lls.
To discover stylistic features.
Questions concerned with producers or causes of content
To identify the intentions and other characteristics of the
communicators.
To determine the psychological state of persons and groups.
To detect the existence of propaganda (primarily for legal
purposes) .
To secure political and military intelligence.
Questions concerned with audience or effects of content
To reflect attitudes, interests, and values ("cultural patterns")
of population groups.
To reveal the focus of attention.
To describe attitudinal and behavioral responses to communi-
cations.
A given study may, of course, pursue several of these purposes.
TECHNIQUES OF ANALYSIS
Stimulated by the rapidly increasing volume of material produced

by the mass media, a special technique-content analysis-has been
developed within the last few decades for describing in systematic form
the content of communications.7
Many of the concepts and assumptions underlying this technique
are, however, much older than its name. Long before content analysis
became established as a technique with a name of its own, students of
society used records of communication for a variety of purposes. His-
torians examined them in order to reconstruct the period in which they
were produced. Literary critics studied the productions of writ~rs to
discover the message they wanted to convey, their peculiarities of style,
the values they propagated, and many other aspects of creative work.
The arguments about whether Shakespeare actually was the author
of all the works associated with his name, for example, were supported
largely by what today would be called content analy,sis.
It is true that modern content analysis has added a new feature
to the exploitation of communication content for research purposes-
namely, the development of elaborate techniques for quantification of
the material. Indeed, in recent methodological publications, quanti-
fication is, as a rule, regarded as a necessary element. Berelson (1952),
for instance, defines content analysis as "a research technique for the
objective, systematic, and quantitative description of the manifest
content of communitation."8
7 Although the tec~nique of content analysis has been worked out primarily
in relation to the mass media, it is applicable to other materials as well. For ex-
ample, personal documepts, unstructured interviews, protocols of responses to pro-
jective tests, records. of patient-therapist interactions, etc., may all be subjected to
content analysis. .
8 For other discu~sions of content analysis, see Bruner (1941), Goldsen
( 1947) , Janis (1943), Kaplan (194 3a ), Kaplan and Goldsen (1943), Lasswell
(1942a, 1942b, 1946)/ Lasswell, Leites, and associates (1949), and Sargent and
Saenger (1947).
This emphasis, due largely to the work of Lasswell and his as

sociates during the late 1930's, has advanced the study of communica-
tion content considerably by adding precision to insight. The analysis
proceeds under certain controls that render it systematic and objective
in comparison with a conventional review or critique of communication
content. (1) The categories of analysis used to classify the content are
clearly and explicitly defined so that other individuals can apply them
to the same content to verify the conclusions; (2) the analyst is not
free to select and report merely what strikes him as interesting but must
methodically classify all the relevant material in his sample; (3) some
quantitative procedure is used in order to provide a measure of the
importance and emphasis in the material of the various ideas found
and to permit comparison with other samples of material. If we take
a systematic sample of, say, newspaper editorials, and count the relative
number of editorials expressing favorable, unfavorable, and n_eutral
attitudes toward a given foreign nation, we are carrying out a simple
form of quantification that has proved feasible and reliable. We shall
come out with a more exact summary of. the situation than would be
possible if only general impressions and memory were relied upon,
since, without mathematical aid, there is a very clear limit to the
ainount of material that can be digested arid recalled in balance and in
detail by the human mind .
. Yet-and this is the other side of the coin-concern with quanti-
ficition has become so dominant that it often overshadows concern
WIth the unique content of communications. Defin~tions of content
analysis tend to emphasize the procedure ot analysis rather than the
character qt the data available in recorded communications. In addi-
tion, they imply a somewhat arbitrary limitation of the field by exclud-
ing from it all accounts of communications that are not in the fbrm of
the number of times various themes or other elements appear in the
material being analyzed. It is indeed difficult to see why quantIfication
should be regarded as a requirement in content analysis when it is not so
regarded in the analysis of data obtained by interviews or observation.
Whatever the nature of the data, quantification is always a mote precise
procedure than qualitative description or exploration. Yet it is not
always feasible; as we shall see in Chapter 1~, both quantified and
unquantified data have their legitimate place in contemporary social
science.
By and large, the technical problems of analysis arising in the
study of communication content are simply specific instances of
general problems of analysis and interpretation in the social sciences,
which will be discussed in Chapter 11. Here, as elsewhere, the execu-
tion of a study demands that the research problem be formulated; that
a study design be developed; that categories be established for the
classification of data; and that the data be systematically tabulated and
summarized in terms of these categories. However, the character of
communication material requires some modification of these customary
procedures, especially if a study aims at quantitative analysis.
DRAWING A SAMPLE OF THE MATERIAL. 9 Suppose that one wishes
to analyze the concern of the press of a given country or countries
with the question of disarmament. The first task of the analyst-one
of considerable difficulty-is to define his universe, the national press.
The techniques of sampling from the mass media are not well de-
veloped. We are concerned with newspapers as potential molders of
reader opinion; for our purposes it is not satisfactory to list all the
newspapers published in a given country and draw every tenth or
twentieth one, even if we also introduce controls to, ensure that news-
papers representing different geographical areas, political orientations,
economic groups, ethnic groups, etc., are included in the proportion
in which they are represented in the total population of newspapers.
The difficulty arises from the fact that newspapers vary tremendously
in size and influence, and a realistic sample should not weigh an obscure
journal equally with a giant metropolitan daily. The situation is not
the same as that of drawing a representative sample of a voting popula-
tion, each member of: which has equal influence at the polls-namely,
one vote. I
To take account pf the size problem, one might divide the news-
papers into a series o~ classes according to their circulation, each class
having a total circulation value equal to that of every other class, and
then draw from each such class a random sample covering a given
number of readers. Another way of approaching the problem would
be to view the population under study as the population of total news-
9 General principles and procedures of sampling are discussed in Appendix B.
paper copies issued (a paper with a circulation of 1,000 would have

1,000 units in this population) and to draw a random sample from
such a population rather than from the population of newspaper titles.
So far as items concerning disarmament are concerned, however, we
cannot assume that the circulation figure of a newspaper correctly re-
flects its influence upon the population. One paper may be concerned
almost exclusively with local news, whereas another may feature many
items dealing with international relations.
In view of the problems involved, we may use samples that do
not pretend to be fully representative of the mass medium in question
but that can be objectively defined and systematically drawn. One pos-
sibility is a "popularity sample," based only on circulation or audience
figures-e.g., a sample that would use the ten largest newspapers in
the country. Or, if we are interested in comparing the press of different
countries in their handling of items concerning disarmament,.. and if
resources pemlit the study of only one newspaper from each country,
it may be appropri:lte to select the paper that is commonly quoted
as the organ with the greatest authority_ or prestige in the nation con-
~erned-e.g., the New York Times in the United States, the London
Times in England, and Pravda in the Soviet Union.
There is another problem in sampling from the mass media that
must be considered-the time problem. It would be easy to get a
distorted impression of the general policy of newspapers if the editions
for only a single day were studied, or even a single month. The impact
of a particular current event may be such as to. obscure a paper's usual
policy. If the analyst elects to c9ver a period of several months, the task
will be unmanageable unless he draws a sample of issues of the news-
paper during that period. Before he knows how many issues he can
handle, he will have to decide whether he wishes to analyze the entire
issue of each paper on the sampling ~ay~, or to concentrate on specific
aspects, such as headlines, editorials, news reporting, position of prom-
inence within the paper (page-one articles, for example), or the like.
In other words, he will have to decide on the nature and size of the
units that are to compose his sample.
Frequently, then, the sampling procedure in communication
analysis consists of three stages: sampling of sources (which news-
papers, which radio stations, which films~ etc., are to be analyzed);
sampling of dates (which period is to be covered by the study); and
sampling of units (which aspects of the communication are to be
analyzed). With respect to the sampling of units, decisions are often
arbitrary and based on tacit assumptions about which feature of a
medium best characterizes it': For example, is it the headline, the
human-interest story, the editorial, or some other feature that best
indicates the policy of a newspaper?
To avoid such arbitrariness, content analysts frequently follow one
of two possibilities: They analyze on the basis of several different units
(for example, they take samplings of headlines, of human-interest
stories, of editorials, and count how many times a given subject is men-
tioned in each); or they disregard these "natural" units completely,
dividing the issue of a newspaper mechanically into lines or inches of
space from which they draw a sample. Much more work on readership
habits is necessary before it will be possible to decide whether the
apparent accuracy of a procedure that yields units of equal size compen-
sates for the neglect of context and inherent organization.
ESTABLISHING THE CATEGORIES OF ANALYSIS. 10 Suppose that our
analyst has decided in favor of a sample of two types of unit-editorials
and front-page stories. His next task is to establish the categories ill
terms of which every unit can be classified. As elsewhere in soci~~'re:
search, he has two sources for the establishment of relevant cat~gories:
the formulated research purpose, including any hypothe~~s he may
have put forward; and the material itself. It is, of cours<::- the material
that suggests categories peculiar to communication anal'sis.
~ newspaper's concern with ~isarmament ~n fr:d expression in
a v~nety of way~. The pap~r can eIth~r e~phasIze~ or ignore the issue
. or It~ contro~ersIaI nature; It can res~nc~ I:self to"straight reporting on
the ISSU~ or It can p~oduce much edI:onahzed comment on it, playing
up certam themes ~ore than .others; It can use ~~rtain key words, such
~s peace, frequently ?r rarely; It can treat the w,:atter lightly or seriously;
It can appeal to commonly accepted values 01:. omit the moral implica-
tions of the issue. EJch of these categories o~fanalysis, and many more,
have been used by content analysts. .
The most extetisive work in the contel tt analysis of newspapers
10 A more general ~iscussion of establishing categ 0 flo on b f d
in Chapter 1 1 . ; ones 0 ana YSIS WI e oun
has been carried out by Harold D. Lasswell and his associates (1949).
Lasswell developed a system of "symbol analysis," which was employed
during World War II in several branches of the United States govern-
ment. In this system, newspaper content is studied for the appearance
of certain symbols, such as "England," "Russia," "democracy," "Jews,"
"Stalin," etc. The frequency with which these symbols appear is noted,
as well as whether their presentation is favorable, unfavorable, or
neutral (or "indulgent," "deprivational," "neutraL") Favorable refer-
ences are sometimes further divided into those stressing "strength" and
those stressing "goodness" or "morality"; negative references into
"weakness" and "immorality" categories.
Davison's analysis of Berlin newspapers, described on page 333,
made use of this type of analysis. The symbols he considered were
I the names of the countries; the favorable, neutral, or unfavorable
quality of each reference was noted. Davison did not use the dimen-
sions of strength-weakness and morality-immorality, but he made an
additional analysis in terms of "themes," another widely used method
of content analysis. In this approach, the analyst immerses himself
in the material until its recurrent ideas or propositions become evideut
a ~nd then counts the frequency with which these propositions occur.
For ',example, some of the major themes Davison found in news items
in the h,S.oviet-controlled Berlin newspapers were: the United States
is torn by I.. ~conomic unrest and industrial strife; the United States is
in the grip 01 ( reactionaries; the United States is pursuing policies of
militarism, impt 7!ialism, and dollar diplomacy.
Wright and 1 ,Telson (1939) employed a more complicated method
of content analysis I,of newspapers. Using a sample of editorials con-
cerning Japan and Ch: ina in the New York Times, the Chicago Tribune
and the Chicago Dall)'T News for the period January 1937 to March
1938, they selected a '\epreserHative statement" from each editorial
and then asked judges to classify these statements in eleven piles, rang-
ing from pile 1 (most. hO,\)tile toward the country concerned) through
pile 6 (neutral) to plle ~ 1 (most favorable). Over-all scores for the
different newspapers and ~for different periods of time were obtained
by averaging the score valu~s of specific statements. The results showed,
, among other things, that the bombing of Nanking had more, influence
in promoting unfavorable references to Japan than did the Panay inci-
dent, and considerably more than the juridical act of the League of
Nations in branding Japan the aggressor.
RELIABILITY OF CLASSIFICATION. Reliability of responses and of
classifications is, of course, a universal problem in social science.
Methods of ascertaining, and of increasing, the reliability of measure-
ment were discussed in Chapter 5. More specific discussion of problems
of reliability of coding appears in Chapter 11. Ideally, our methods of
analysis and quantification should be so clearly defined that different
judges would arrive at exactly the same results when analyzing the
same material. Perfect reliability, however, is something that can be
achieved at the present time only when the more superficial kinds of
analysis are made, such as counting the number of times a particular
word turns up in a given amount of material. As soon as some degree
of interpretation enters the analysis, judges tend to differ to some
extent in their results.
White's study of the values stressed by Hitler and by Roosevelt,
described on pages 333-334, provides an example of difficulties that
may arise when complex judgments must be made. In the category
"other values" White included the value safety, which is illustrated by
the following statement from a Hitler speech: "It is quite unimportant
whether we ourselves live, but it is essential that our people live, that
Germany shall live." It is obvious that the assignment of the va!ue
safety to this statement involves an interpretation of the meaning of
the sentence in the context within which it was used. The primary
method of increasing reliability of classification is to specify clearly the
characteristics of statements that are to be placed in a given category,
and to use many examples drawn from the material being analyzed to
'illustrate what kinds, of statements are to be considered as belonging
in a given category. But it is obviously much more difficult to give a
definition of the value safety which will be both sufficiently compre-
hensive and sufficien'tly specific to serve as a guide for the coder, than
to give an adequate qefinition of a category such as "mentions racial
groups" in coding an interview question about "what kinds of people
live in this neighb~rhood."
L
The difficulty is increased by the variety of
material to be considered in content analysis, which is limited only by
the interest or intention of the communicator, without the restricting
influence of a specific interview question. To increase reliability of con-
tent analysis there is no other way but patient experimentation with

the refinement of definitions and careful training of the persons en-
trusted with their use in classifying the data.
Summary
In the last four chapters we have discussed various ways of gather-
ing data needed to answer research questions: observation, interviews
and questionnaires, projective and other indirect techniques, and the
use of available data in the form of statistical records, personal docu-
ments, and mass commurlications. We have pointed out that each
method has its advantages and limitations, and that each is more ap-
propriate for answering certain types of research questions than it is
for others. Moreover, we have noted that no matter what technique
an investigator uses, he must be alert to problems of reliability~ and
validity of his data.
In the next chapter we shall consider procedures for placing in-
dividuals on scales on the basis of data collected by any of the methods
discussed in these last four chapters.
10
PLACING INDIVIDUALS
ON SCALES
Rating Scales
Questionnaires that Form Scales

-Some Modifications of Scaling TeC'hniques
A Concluding Note
I N OUR DISCUSSION of measurement in Chapter 5, we noted that it is
frequently desirable to make distinctions of degree rather than of
quality. We wish to be able to assert, for example, that Mr. Brown is
more prejudiced than Mr. Smith rather than being limited to the
statement that their prejudices are different. Making these distinctions
of degree may be thought of as a function of analysis rather than data
collection. However, the desire to be able to make such distinctions
influences the form in which the data are collected. At the very least,
it means that the questions asked must be such as to give information
on which judgments of degree can be based. Frequently, the distinc-
tions of degree ale introduced into the measuring instruments them-
selves.
Techniques for registering differences in degree are of two broad
types. In one type, someone makes a judgment about some ch3.racter-
istic_of an inqividuaJ and placeS-him directly_on a scale defined in terms
of that characteristic. Let us say that the characteristic in question is
"attitude toward school desegregation"; the rater places the individual
on a rating scale set up in such a way that it indicates various degrees
of favorable or unfavorable attitude toward school desegregation. The
person making the judgment may be the individual himself, someone
who knows him, an observer, an interviewer, a coder, etc. He may
make the judgment on the basis of his entire experience with the in-
dividual, or after observation of the individual in a specified situation
or situations, or on the basis of the. individual's responses to interview
questions, projective techniques, etc. Different types of rating scale
are discussed in the following sectio~.
The other broad type of technique for' registering differences in
degree consists of questionnaires constructed in such a way that the
score of the individual's responses places him on a scale. If the investi-
gator is interested in an individual's attitude toward school desegrega-
tion, for example, he does not ask either the individual himself or
anyone else to make a direct judgment about his attitude. Rather, the
individual is asked to reply to a series of questions relevant to desegre-
344
PLACING INDIVIDUALS ON SCALES 345
gation or to indicate his agreement or disagreement with a series of
statements-for example, "I would be willing to sit next to a Negro
in class; I believe the academic standards of my school would be
lowered if Negroes were admitted," etc. From his replies to these
questions or statements, a score is computed that is taken as indicating
his position on a scale of favorable or unfavorable attitude toward de-
segregation. This technique has been used most often in the measure-
ment of attitudes. Various types of attitude scale are discussed in a
later section.
Both rating scales and attitude scales have the same purpose: to
assign individuals to numerical positions in order to make distinctions
of degree possible. The scale positions, generally, indicate only the
order of positions with respect to the characteristic being measured.
Although there have been attempts to construct interval scales, most
rating scales and attitude scales do not provide more than ordinal meas-
urement (see Chapter 5, pages 186-197).
Rating Scales
A number of types of rating scale have been employed; the number
of types distinguished and the names given to them tend to vary with
the measurement theorist. 1
One feature is common to all types: the rater places the person or
object being rated at some point along a continuum or in one of an
ordered series of categories; a numerical value is attached to the point
or the category. Scales differ in the fineness of the distinctions they
permit and in the procedures involved in assigning persons or objects
to positions. These differences will become apparent in our discussion
of several of the more common types of rating scale.
~APHIC RATING SCALES!

Perhaps the most widely used is the graphic rating scale. In this
type, the judge (the subject himself, the interviewer, observer, coder,
1 For more detailed discussions of rating scales, see Cronbach (1949), Guilford
(1 q 54 ) , Freeman (1955). and Krech and Crutchfield (1948).
346 PLACING INDIVIDUALS ON SCALES
etc.) indicates his rating by simply placing a check at the appropriate

point on a line that runs from one extreme of the attribute in question
to the other. Scale points, with brief descriptions, may be indicated
along the line; their function is to serve as a guide to the judge in
localizing his rating rather than to provide distinct categories.
The following example is a scale checked by interyiewers of
respondents living in interracial housing projects (Deutsch and
Collins, 1951) :
RESPECT FELT FOR NEGROES -IN THE PROJECT

(Place check on appropriate position on line, or circle X or Y.)
Thinks
highly of Generally Is ambivalent;
Negroes in respects partly respects, Generally Strongly
project Negroes partly feels feels they feels
without living in they are are in- they are
q ualifica tion project inferior ferior inferior
1 2 3 4 5 6 .7 8 9
X: Is indifferent to Negroes as a group; doesn't think about them.
Y: Doesn't think of Negroes as group; considers them as individuals.
In this example, the X and Y items were introduced to avoid

forcing the interviewer to make a rating he considered inappropriate
and to allow him to explain why a rating could not be made.
~One of the major advantages of graphic rating scales is that they
are relatively easy to use:-Cuilford (1954) has pointed out that "the
graphic scale provides opportunity for as fine discrimination as that
of which the rater is capable, and the fineness of scoring can be as
/ great as desired:)Yet:for the effective, use of graphic rating scales,
experience has shown that certain ~recautions-must be taken in their
design and use: end statements so extreme that they are unlikely to
be used should be avoided;. descriptive statements should be placed
to correspond as-closely as poss.ible with numerical points on the scale;'
etc. For a more detaile<;l discussion of practices to be followed in the
I construction and use of ~raphic rating scales, t?e reader is referred
to Guilford (1954).
RATING SCALES 347
ITEMIZED RATING SCALES
l Itemized rating scales have also been referred to as "specific

category scales" (Krech and Crutchfield, 1948) and "numerical scales"
(Guilford, 1954). In this type of scale, the rater selects one of a limited
number of categories that are ordered in terms of their scale position.
The number of scale positions or categories used has varied, depending
on the research problem and the kinds of judgments required. Scales
with five or seven categories are most frequently employed, but many
investigators have used as many as nine or eleven points.
Itemized rating scales, like graphic ones, can be used in connection
with data gathered by any of the methods discussed in the preceding
chapters. Barker, Dembo, and Lewin (see pages 115-116), for example,
in their study of the effect of frustration on constructiveness of play of
young children, set up a seven-point scale for rating "constructiveness"
in analyzing their detailed narrative records of the children's activities.
From a detailed analysis of a few of these records they drew specific
illustrations of activities that were judged to fall at various points on
the scale; a different set of illustrations was provided for each of the
toys available to the children. The descriptions of the scale points for
constructiveness of play with a truck are given below, with one concrete
example under each: /
1. The toys are examined superficially.
Ex: Sits on floor and takes truck and trailer in hand.
2. The truck is moved to a definite place or from one place to
another.
Ex: Bends over to truck and trailer, pushes back and forth.
3.. Somewhat more cOl;nplicated manipulation of truck.
Ex: Truck and trailer backed under chair.
4. Definitely more complicated and elaborate manipulation of
truck. :
Ex: Truck and trailer unloaded, detached; pulled in circles;
I
reattached, detached, reattached . . .
5. The truck is used as:a means to haul other things.
Ex: "This is a fire nnick.' Pushes truck to middle of room,
around in middle. ~'You can load things in it. Mr. Duck! I'll
haul Mr. Duck." .
6. The meaning of the play is an extensive "trip" or another

elaborated story in which the handling of the truck is merely
a part of a larger setting.
Ex: "Here's a car-truck, and it's going out fishing, so we have
to take the trailer off. First, we have to go to the gas sta-
tion. . .." Gets gas; now back for the trailer and the fish
pole; attaches motor boat to truck and trailer. . . .
7. Play showing more than usual originality.
Ex: Detaches truck, has it coast down trailer as an incline ..
When the characteristic being rated can be thought of as extend-

ing on either side of a neutral point, the scale usually provides a central
neutral point, with an equal number of categories on each side. For
example, in the analysis of an open-ended interview question in a study
of attitudes toward international relations, the Survey Research Chater
of the University of Michigan used a five-point itemized scale e~nd-
! ing from 'lEmphatic approval or agreement" (scale value 1) to
"Emphatic disapproval or disagreement'~ (scale value 5'). The coding
instructions included, for each category, a rather elaborate description
of the kinds of responses that were to be assigned to it. Thus, the
descriptions for the first two scale positions were as follows:
1. Emphatic approval or agreement. Included in this cate-

gory are the people expressing agreement or a favorable attitude
without equivocation or doubt and supporting it with a reason
("I think we should lend m~)lley to England. Their country is so
torn up and lots of people need, help:') and the people giving
an emphatic' answer, regardless of whether it is supported by a
reason ("Absolutely, it's the thing to do") . -
I 2. Approval or agreement, with qualifications or uncertainty.
Included here are people expressing a qualified opinion, generally
in such ways as "if . . ." "as long as . . ." ("It is a good thing
for us to do as long as we get it back some day"), and those ex-
pressing a generally favorable attitude contaIning one negative or
unfavorable argument ("It's a big load for us to take on with our '
country so in debt.,But we didn't have the destruction the way
England did, so I guess we have to do it"), and those expressing
a favorable attitude without certainty or conviction by using ex-
RATING SCALES 349
pressions like "maybe," "perhaps," "I guess," "I suppose," etc.,
or expressing a simple affirmative answer with no reason. 2
The verbal descriptions used to identify the scale positions may
vary from brief statements indicating only degrees of the attribute to
more elaborate descriptions including illustrations of behavior ap-
propriate to the category. The two examples just given provide rela-
tively detailed definitions and illustrations. At the other extreme, Pro-
shansky (see page 294), in the analysis of descriptions of pictures
in a projective test of attitude toward labor, instructed his judges to
rate the descriptions in terms of a five-point scale on which the two
end categories were defined simply as "very favorable" and "very
unfavorable," and the middle category as indicating that it was impos-
sible to classify the respondent as either favorable or unfavorable; the
second and fourth positions were given no specific definition other
than that they were to be used for subjects who fell between the
extreme and the middle position.
The more clearly defined the categories, in general, the more~
reliable the ratings are likely to be. How much definition is needed de-
pends on such considerations as the nature of the material, the familiar-
ity of the coders with the concepts involved, the fineness/ of distinctions
required for the study, etc.
, The problems involved in the construction and application of
itemized rating scales are for the most part similar to those that char-
acterize other types of rating scales. These problems are treated later
in this section.
I I
COMPARATIVE RATING SCALES

I
I
In using graphic and: itemized rating scales, the rater makes his
judgment of the individual without direct reference to the positions
of other individuals or grriups with which he might be compared. On
the other hand, comparative rating scales-as their name suggests-
t,
2 Adapted from Cottrell (1947). This rating scale was used by coders in the
analysis of interviews. Howev;er, similar ratings may be made directly by inter-
viewers or by observers of group discussion.
clearly imply such relative judgments. The positions on the rating

scale are expressly defined in terms of a given population or social group
or in terms of people of known characteristics. For example, a question-
naire used in selecting applicants for admission to a graduate school
may ask the rater for an estimate of the given applicant's ability to do
graduate work "as compared with the total group of graduate students
you have known"-is he more capable than 10 per cent of them, 20
per cent, 30 per cent, etc. Or the rater may be asked to indicate, for
example, whether an individual's leadership skill most closely resembles
that of person A, of person B, of person C, etc. (all of whom are known
to the rater and all of whom have been assessed in terms of their leader-
ship skill) . In the first example above, in order to make a valid rating
the judge must have a clear conception of the range and distribution
of the abilities of the total graduate student group. Scales of the
second type are often difficult to construct, since there may not bE
sufficient variation in leadership behavior (or whatever attribute is
being rated) among the people known to the judges to serve as ex-
amples for the various points on the scale. .
Another comparative or relative rating prQcedure is the rank-order
scale. Here the judge is required to rank individuals specifically in rela-
tion to one another; he indicates which person is highest in terms of
the characteristic being measured, which is next highest, etc., down
to the one who is lowest. Ranking in this fashion is used only wh..en the
investigator is concerned with a limited group of individuals. The
rating an individual receives indicates simply his relative rank or posi-
tion in the grOtlP being studied; it' would not necessarily be of any use-
fulness apart from the specific group whose members are being
compared.
SELF-RATINGS VERSUS RATINGS BY OTHERS
All these types of scale may be used to secure an individual's

rating of himself or someone else's rating of him. Self-ratings have the
same advantage~ and limitations as self-reports, discussed at the be-
ginning of Chapter 7. It seems reasonable to assume that the in-
dividual is often in a better position to observe and report his beliefs,
RATING SCALES 351
feelings, etc., than anyone else is. This assumption is valid, however,
only if the individual is aware of his own beliefs and feelings and is
willing to reveal them to others. If the individual is unaware, for
example, of the fact that he has hostile feelings toward a particular
minority group, or if he is aware of such feelings but is afraid of the
consequences of revealing them, then the self-rating procedure is of
little value. Another difficulty arises from the fact that even if the
individual is capable of reporting his beliefs or feelings objectively,
his conception of what constitutes a moderate or an extreme position
Illay be quite different from that of others making comparable self-
ratings.
Despite the hazards involved, self-ratings have proved useful in
the measurement of social attitudes. For beliefs and feelings that the
individual can be expected to be aware of a?d willing to report-for
example, attitudes toward television programs-self-ratings are a useful
source of information. With respect to certain attributes of attitudes-=-
for example, intensity, importance, etc.-self-ratings have so far proved
to be the o~ly satisfactory source of information. Attention by the
investigator to such matters as clearly specifying the dimension to be
rated and defining the frame of reference or standards against which
the ratings are to be made may serve to reduce /
the possibilities of
distortion in self-ratings and increase their usefulness.
CONSTRUCTION AND USE OF RATING SCALES: SOME CAUTIONS
Since a large element of judgment enters into the use of rating

scales, there is considerable room for systematic errors to be introduced
by the personal bias of the rater or raters. One rather common syste-
matic error is the 'fhalo effect." If more than one characteristic of a
I
person is to be judged, raters frequently carryover a generalized impres-

sion of the person Ifrom one rating to the next, or they try to make
their ratings consistent.3 Thus, if a rater considers a person to be shy
and he believes shy, people to be poorly adjusted, he is likely to rate
I,
3 The tendency to make ratings consistent is not unlike what Newcomb (1931)
has described as a "logical error"; i.e., judges often give similar ratings on traits
that seem to them to be logically related.
the person poorly adjusted as well as shy. It is apparent that the halo
effect reduces the validity of the ratings of some traits and introduces
a spurious degree of positive correlation among the traits that are
rated.
Another frequent type of constant error is the "generosity error."
Here the tendency of the rater is to overestimate the desirable qualities
of subjects whom he likes. Still other frequent errors have been iden-
tified. Thus, raters tend to avoid making extreme judgments and to
assign individuals to the more moderate categories. Murray et at
(1938) have identified the "contrast error," in which there is a tendency
on the part of the rater to see others as opposite to himself in a trait.
They found, for example, that raters who were themselves very orderly
rated others as being relatively disorderly, whereas raters who were
themselves less orderly tended to see others as more orderly.
Obviously, one way of reducing constant errors such as those
described above is to train the raters carefully and, especially, to make
them aware of the possibility of such biases. Specific steps may be
taken to reduce the likelihood of specific types of error. For example,
the tendency to avoid using the extreme positions may be counteracted
by giving somewhat less-than-extreme labels to these positions. People
are more likely to check "I am well satisfied with my job" than "I am
completely satisfied with my job"; at the other extreme, they are more
likely to check "There are many things about my job that I do not
like" than "There is nothing about my job that I like." The "gen-
erosity error~' may be reduced by using relatively neutral descriptive
terms for the scale positions rathdr than evaluative ones; for exau;ple.
"does not readily accept new opinions or ways of doing things" rather
than "rigid." Halo effects may be reduced, or eliminated altogether, by
having the various ratings of a given person made independently-
either by different raters, or by the same rater at different times. with-
out awareness that he is rating the same person. Obviously, this latt~r
condition can be met only when the ratings are made on the basis of
recorded material, such as responses to interview questions, accounts
of behavior, etc.,.from which identifying information can be removed.
Systematic errors, of course, reduce the validity of ratings. There
1Jlay also be random errors that reduce their reliability. One frequent
RATING SCALES 353
source of unreliability among different raters is the fact that some
frame of reference is implicit in any rating; different raters may use
different frames of reference in describing individuals in terms of the
characteristic in question. For example, the rating of a person as "con-
servative" or "radical" takes its meaning from the rater's reference
groups-the group norms he has in mind as he makes his rating. Lack
of correspondence between ratings by different observers is frequently
due to the fact that they make ratings with different reference groups
in mind.
[Reliability can be increased not only by careful training of raters
but also by attention to the construction of the rating scal~ Clear
definitions of the characteristic being measured and of the various
positions on the scale, as well as clear specification of the reference
group, help to reduce unreliability. Whenever possible, the definitions
of the scale points should include concrete illustrations of question
responses, types of behavior, or communication content. Careful can
sideration should be given to distinguishing between adjacent positions
on the scale; for example, to the difference between "favorable" and
"very favorable." The example given on pages 348-349 illustrates this
procedure.
Un constructing a rating scale, one must decide how many scale
positions or categories are to be used, unless ohe is using a graphic
scale on which the rater is free to check any point on a continuous
line. TI1ere is no simple rule for determining the optimal number of
positions. A basic consideration, .of course, is the degree of differentia-
tion wanted in the measurement. But regardless of what is demanded
by the research problem, other factors must be taken into account:
, (1) the discriminative ability of the judges or raters, including the
extent to which tl;ey I
are trained and experienced; (2) the kind of
characteristics to b~ judged; e.g., whether "they are complex "inner"
attributes or more" manifest "outer" attributes; (3) the conditions
under which the ratings are to be made; e.g., whether they are based
on extensive data, (long periods of observation of the subject or a great
deal of communicationI,
content) or on limited data (brief observa-
tion or limited communication content). These factors interact' in
their effect on the; degree of fineness possible in the rating scale. If
relatively manifest behaviors are to be rated, then a more differentiated

scale may be used even with relatively unsophisticated judges, provided
they are trained appropriately. On the other hand, if complex attributes
are to be judged and experienced judges are not available, it may be
necessary to use a less differentiated scale in order to obtain reliable
ratings. In any case, all of these factors must be considered in deter-
mining the number of distinctions to be included in the rating scale.
Very often preliminary testing is necessary before a final decision can
be made.
Reliability of ratings is usually enhanced considerably by having
several raters working as a team-making independent judgments, com-
paring their ratings and discussing discrepancies, and making second
independent judgments that are then pooled or averaged to give a final
score (see Murray et al., 1938). Much research has demonstrated the
superiority of the average, or consensus, of the judgments of several
people over that of one individual (see Murphy, Murphy, and New-
comb, 1937). Poffenberger (1942) has written:
From the studies of judgment that are available, it would seem

that three independent estimates of the traits commonly judged
is the minimal requirement for satisfactory work. In many cases
where the variables affecting the judgments are numerous, the
number should be even larger.
Obviously, the pooling of independent ratings is more feasible

when the rating is being done by coders working from recorded
material or by observers of a behavioral situation; it is often not
feasible to have more than one rater (interviewer) present in an Inter-
viewing situation.
In many studies ratings are made in the process of analysis by
coders who have studied the recorded material. Often, however, rat-
ings are made on the spot by an interviewer, observer, etc., since there
may be overtones in the subject's manner or behavior that cannot be
recorded adequately and therefore cannot be taken into account in a
t rating made by, an analyst who has not had direct contact with the
subject. Little systematic evidence of the relative merits of these two
procedures is available. However, Maccoby and Maccoby (1954) sug-
RATING SCALES 355
gest that, at least in the case of interviewing, the advantage of the
interviewer in being able to observe additional cues may be more than
offset by two disadvantages: (1) interviewer ratings are especially
susceptible to halo effects, since the interviewer necessarily knows that
it is the same individual he is rating and since all the ratings of a given
individual must be made at the same time; (2) each interviewer tends
to develop his own frame of reference for his ratings, based on the
particular sample of respondents he has interviewed, whereas the
standards of coders working in the same office and subject to frequent
checking, if necessary, can be kept more consistent.
It would be rather fruitless to attempt to discuss in any detail
what constitutes a good rater, since there is very little definitive research
on this topic. However, several general conclusions seem warranted.
First of all, not only do individuals differ in their ability as raters, but
the same individual differs in his ability to rate different character-
istics. A good rater with respect to one characteristic may be poor with
respect to another. In other words, the ability to judge is not general;
it is a highly specialized function within the individual. Second-con-
trary to common expectation-acquaintance with the person to be
rated does not lead to increasing validity of ratings. Guilford (1954)
has pointed out that long acquaintance often /results in substantial
"generosity errors." Third, confidence in one's judgment is not neces-
sarily an indication of its validity. At least one investigation (Kelly and
Fiske, 1950) has reported an inverse relation between confidence and
the validity of predictions. However, Guilford (1954) cited an early
study by Cady, which indicated that judgments of which the rater is
confident are much more reliable than those of which he is less con-
, fident. Fourth, mo~t of the studies of validity of ratings seem to in-
dicate that people "rho are high on a characteristic generally considered
undesirable are poor judges of that characteristic both in others and
in themselves. I
The simplicity:of rating scales commends their use to many. Their
simplicity, howeve~, may be more apparent than real. Often so much
time is required t9, establish adequate reliability that the procedure
becomes uneconomical. Yet it is also true that, despite the difficulties
involved, a rating 'scale may be the only instrument available for a
given purpose, at least in the present stage of development of research

techniques. Furthermore, as we indicated earlier, rating-scale methods
have a wide range of application; they can be used in connection with
communication content, observed behavior, responses to question-
naires, or data collected by almost any other method.
Perhaps the most significant problem in the use of rating scales
has to do with their validity. Even a cursory examination of the
relevant literature reveals that, whereas evaluation of the reliability of
rating scales is common, measures of validity are rarely reported. The
reason, of course, is the dearth of available external criteria against
which ratings can be compared. As a matter of fact, ratings themselves
have often been used as criteria for checking the validity of other types
of measuring instrument, such as personality tests.
This does not mean, of course, that investigators have completely
ignored the question of the validity of their rating scales. In effect, they
have assumed that their scales were valid when the following conditions
obtained: (1) the attributes being measured were relatively "objec-
tive," so that their meaning would be uniformly understood by the
raters using the scales; (2) the ratings themselves were obtained under
optimal conditions, including carefully constructed scales, trained
judges, and specified common frames of reference. Under these condi-
tions, one may not go too far wrong' in assuming that if the obtained
ratings are reliable, they are probably also valid. However, if either one
of these conditions is not met, then the assumption of validity is
hazardous. The first condition has been overlooked by some investiga-
I
tors. There is little reason to assume that intuitive judgments about

complex attributes are inherently valid, even when the judgments are
made by highly trained and intelligent people.4 If the concept of what
is being measured is vague, as it is in some rating scales, it is unlikely
that the ratings will be clear in meaning. When the concept of what is
being measured is ambiguous, the ordering of individuals may actually
be quite arbitrary, and even distinctions of greater or less become
meaningless.
4 For example, Kelly and Fiske (1950) found that ratings by experienced
clinical psychologists based on unstructured interviews had little value in predicting
performance in situations that were not clearly specified.
QUESTIONNAIRES THAT FORM SCALES 357
Questionnaires that Form Scales 5

As we have pointed out, a large element of judgment enters when
a rater (whether it be the individual himself or someone else) places an
individual on a rating scale on the basis of his observed behavior, his
answers to open-ended questions, etc. In an effort to devise procedures
that would make it possible to place individuals on a scale with less
likelihood of error, carefully standardized questionnaires have been
constructed. In this approach, the individual does not directly describe
himself in terms of his position on the dimension in question. Rather,
he expresses his agreement or disagreement with a number of state-
ments relevant to it; on the basis of these responses, he is assigned a
score. In the process of standardizing the questionnaire, the investi-
gator has established a basis for interpreting scores as indicating posi-
tions on the dimension. Since this technique has been used most often
in the measurement of attitudes, our discussion will focus on attitude
scales. 5
Attitude scales differ in method of construction, method of re-
sponse, and basis for interpreting scores. Different types of attitude
scale will be discussed in this section.
The separate items or questions in an attitude scale are usuallY,not
of interest in themselves; the interest is, rather, in the total score or in
subscores that result for each individual from the combination of his
responses to various items. In effect, any set of items works as well as
any other set provided they give the same final scores on the particular
attitude being measured.
In selecting items for inclusion in a scale, two criteria are com-
monly used. First, the, items must elicit responses that are psycho-
logically related to the dttitude being measured. For example, in a scale
measuring anti-semitism, the following item has a manifest relation to
the attitude being measu{ed: "Anyone who employs many people
should be careful not to. hire a large percentage of Jews" (Adorno et aI.,
1950,). However, the relationship does not necessarily have to be so
j_
5 For more detailed discussions of attitude scaling, see Edwards (1957a),

Green (1954), and Peak (1953).
evident. In fact) there is a considerable advantage to using items that,

on the surface, have no bearing on the attitude being measured. This
may prevent the respondent from concealing or distorting his attitude.
Thus, in their study of anti-democratic ideology, Adorno et a1. used
many items that have no apparent relationship to this attitude-e.g.,
"When a person has a problem or worry, it is best for him not to think
about it, but to keep busy with more cheerful things." This item is one
of several that indicate an individual's readiness or lack of readiness to
adopt a psychologically insightful view of other people and of himself.
The theory is that people who are lacking in psychological insight and
understanding have a personality structure (e.g., greater repressed
hostility, weaker ego, etc.) that predisposes them to an anti-dem-
ocratic ideology.6
The second criterion requires that the scale differentiate among
people who are at different points along the dimension being measured.
To discriminate not merely between opposite extremes in attitude but
also among individuals who differ slightly, items that discriminate at
different points on the scale are usually included. Thus, a test of
opinions about child-rearing practices, aJong the 'dimension "permis-
siveness-strictness," would contain not only items representing a very
strict approach and others representing a very permissive approach, but
intermediate items representing moderate strictness, moderate permis-
siveness, etc. Some types of scale, however, provide for the identifica-
tion of moderate positions by permitting the expr~ssion of various
degrees of agreement or disagreemen~ with extreme items 'rather than
by the inclusion of intermediate items.
The way in which a scale discriminates among individuals depends
on the construction of the scale and the method of scoring. In some
I
6 Such indirect items cannot, of course, be used as measures of the attitude

being studied simply on the basis of theoretical assumptions about their relation
to the attitude. Before they are accepted as adequate measures, their relation to
the attitude must be demonstrated. (This statement is equally true for items that
seem to have a manifest relation to the attitude being studied.) Thus, in the study
of antidemocratic ideology, the hypothesis that items such as the -one quoted -
were related to anti-democratic ideology was tested-and borne out-by analysis
of the difference between responses to such items made by people known on other
grounds to have a democratic ideology and those made by people known to have
an antidemocratic ideology.
scales the items form a gradation of such a nature that the individual
agrees with only one or two, which correspond to his position on the
dimension being measured, and disagrees with statements on either
side of those he has selected. Such scales, in which a person's response
localizes his position, are sometimes called differential scales. In other
scales, the individual indicates his agreement or disagreement with
each item, and his total SCOre is computed by adding the subscores
assigned to his responses to all the separate items; such scales are
sometimes called summated scales. Still others are set up in such a way
that the items form a cumulative series; theoretically, an individual
whose attitude is at a certain point on the dimension being measured
will answer favorably all the items on one side of that point and answer
u~favorably all those on the other side. Each of these types of scale is
discussed in more detail in the following paragraphs.
DIFFERENTIAL SCALES
Differential scales for the measurement of attitudes are closely

associated with the name of L. L. Thurstone. The methods he devised
represent attempts to approximate interval scales. An interval scale,
it will be recalled from Chapter 5, is one on which the distances be-
tween points on the measuring instrument are known, and on which
equal numerical distances represent equal distances along the con-
tinuum being measured. Such a scale enables one to compare differ-
ences or changes in attitude, since the difference between a score of
3 and a score of 7 is equivalent to the difference between a score of 6
and a score of 10 and to the difference between any other two scores
that are four points ap4rt.
A differential scal,e consists of a number of items whose position
on the scale has beenjdetermined by some kind of ranking or rating
operation performed by judges. Various methods of securing judg-
ments of scale position: have been used: the method of paired compari-
sons (see Thurstone, 1927, 1928); the method of equal-appearing inter-
vals (see Thurstone, ~929, 1931, and Thurstone and Chave, 1929);
and the method of successive intervals (see Saffir, 1937). It is beyond
the scope of this volume to give the details of these procedures; we shall
only present in broad outline the method of equal-appearing intervals,

which is the most commonly used.
In selecting the items for the scale and assigning values to them,
the following procedure is used: (1) The investigator gathers several
hundred statements conceived to be related to the attitude being in-
vestigated. (2) A large number of judges-usually from 50 to 300-
working independently, classify these statements into eleven groups.
In the first pile the judge places the statements he considers most
favorable to the object; in the second, those he considers next most
favorable; and in the eleventh pile, the statements he considers most
unfavorable. The sixth, or "neutral," position is defined as the point at
which there is neither "favorableness" nor "unfavorableness."7
(3) The scale value of a statement is computed as the median position
(or pile) to which it is assigned by the group of judges. Statements
that have too broad a scatter are discarded as ambiguous or irrelevant.
(4) A final selection is made, taking items that are spread out evenly
along the scale from one extreme position to the other. It is often
possible to construct duplicate forms of the "'Scale from items not used
on the original form.
The resulting Thurstone-type scale is a series of statements,
usually about twenty; the position of each statement on a scale of
favorable-unfavorable attitude toward the object has been determined
by the judges' classification. The subjects, in filling out the question-
naire, fre asked either to check each statement with which they agree
or to~~lieck the two or three items that are closest to their,position.
The following illustration of items from a Thurstone-type scale
is taken from MacCrone:s study of attitudes toward natives in South
Africa (1937):
7 Throughout this section, for the sale of simplicity, the discussion is w9rded
in terms of scales measuring favorablenessunfavorableness toward some object.
A scale may, of course, be concerned with some other dimension; for example,
liberalismconservatism of social, political, or ~conomic views; permissiveness
strictness of views on child-rearing, etc. In developing Thurstone scales, ithe in
structions to the judges specify the dimension along which the items are. to be
placed. Thus, in developing a scale to measure liberalism-conservatism, the judges
would be instructed to place in the first pile the items they consider most liberal,
in the eleventh those they consider most conservative. The same principles and
procedures apply whether the dimension to be measured is favorableness-un-
favorableness or some other.
Scale Item
value no.
10.3 l. I consider that the native is only fit to do the
"dirty" work of the white community.
10.2 2. The idea of contact with the black or dark skin
of the native excites horror and disgust in me.
8.6 15. I do not think that the native can be relied
upon in a position of trust or of responsi-
bility.
8.4 17. To my mind the native is so childish and irre-
sponsible that he cannot be expected to know
what is in his best interest.
3.8 22. I consider that the white community in this
country owe a real debt of gratitude to the
missionaries for the way in which they have
tried to uplift the native.
3.1 3. It seems to me that the white man by placing
restrictions such as the "Colour Bar" upon
the native is really trying to exploit him
economically.
0.8 11. I would rather see the white people lose their
position in this country than keep it at the
expense of injustice to the native.
I
The scale values, of course, are not shown on the questionnaire,
and the items are usually arranged in random order rather than in order
of their scale value. The mean (or median)8 of the scale values of the
items the individual checks is interpreted as indicating his position
on a scale of favorable-unfavorable attitude toward the object.
Theoretically, if a Thurstone-type scale is completely reliable and
if the scale is measuring,! single attitude rather than a complex of
attitudes, an individual sh9uld check only items that are immediately
contiguous in scale value-~.g., items 15 and 17 above. If the responses
of an individual scatter witIely over noncontiguous items, his attitude
8 Thurstone, on the assumption that scales constructed by this method were
true interval scales (see Chapter 5), advocated the use of statistics appropriate
to interval scales-the mean and the standard deviation. Other investigators,
operating on the more cautiou~ assumption that the intervals are not truly equal,
have favored the use of the median as appropriate to ordinal scales. For a discus-
sion of whether the assumption that these are true interval scales is justified, sec
pages 363-365.
score is not likely to have the same meaning as a score with little scatter.
The scattered responses may indicate that the subject has no attitude
or that his attitude is not organized in the manner assumed by the
scale. There is no a priori reason to expect that all people have attitudes
toward the same things or that attitudinal dimensions are the same
for all.
The Thurstone method of equal-appearing intervals has been
widely used. Scales have been constructed to measure attitudes toward
(war, toward the church, toward capital punishment, toward the
Chinese, toward Negroes, toward whites, etc. In addition, an attempt
has been made by Remmers and his colleagues (1934) to develop
generalized Thurstone scales that might be used to measure attitudes
toward any group, social institution, etc. For example, the Kelley-
Remmers "Scale for Measuring Attitudes Toward Any Institution"
consists of forty-five statements, ranging from "The world could not
exist without this institution," through such items as "Encourages
moral improvement" and "Is too radical in its views and actions," to
"Is the most hateful of institutions." In applyi~g this generalized scale
to the measurement of attitudes toward a given iristitution (war, the
family"the church, advertising, or whatever), the subject is instructed
to check each of the statements with which he agrees in reference to
the given institution, or the statements may be reworded to include
mention of the specific institution being considered (e.g., "War is the
most hateful of institutions").
The Wright-Nelson study of editorial positions of newspapers
concerning Japan and China (see page~ 340-341) illustrates-an applica-
tion of Thurstone-type scaling to the analysis of available data. In this
study, the sorting of statements by the judges served simultaneously
to establish the position of each item on a scale of favorableness-
hostility and to determine the scores of the various newspapers. In
effect, each newspaper was treated as having "checked" all the state
ments selected from its editorials; its score was obtained by averaging
the score values of the specific statements.
Several objections have been raised against the Thurstone-type
scale. First, many have objected to the amount of work involved in con-
structing it. Undoubtedly, the procedure is cumbersome. However,
Edwards (1957) has expressed the opinion that~ in view of 'recent
developments in time-saving techniques, the amount of time and labor
involved in constructing a scale by the method of equal-appearing
intervals is not substantially different from that involved in construct-
ing a summated scale. In any case, it is doubtful that simple methods
for the rigorous construction of scales will ever be developed. The
precise measurement of attitudes is perhaps inevitably a complex affair.
A second criticism has been that, since an individual's score is the
mean or median of the scale values of the several items he checks,
essentially different attitudinal patterns may be expressed in the same
score. For example, on the scale of attitudes toward natives of South
Africa given earlier, an individual who checks the two moderately
"anti" items 15 and 17 receives a score of 8.5 (the median of their
scale values). Another individual, who checks items 1, 15, 17 and 22
(perhaps because 22 has a meaning for him which is different from
that which it had for the judges), also receives a score of 8.5 (the
median of the scale values of these items). The two individuals are
rated as having the same degree of prejudice, even though the latter
checked the most unfavorable item in the scale and the former did not.
Dudycha (1943), after six years' use of the Peterson test of attitude
toward war (a test constructed by the method of equal-appearing
intervals) with college students, reported that the average student,
instead of checking only two or three contiguous items, covered more
than a third of the scale; some students endorsed statements ranging
from those placed at the "strongly favorable:' end of the scale to shlte-
ments at the "strongly opposed" end. (One must, of course, comider
the possibility that such students had no clear attitude toward war and
that it was therefore inappropriate to try to measure their altitude by
any technique.) Dudycha questioned the meaning to be given to a
median derived from such a range of responses. However, the criti-
I
cism that identical scoJ,'es do not necessarily indicate identical patterns
of response is not unique I
to the Thurstone-type- scale; it applies at
least as strongly, as we shall see, to summated scales.9
A still more seriohs question has to do with the extent to which
the scale values assigned to the items are influenced by the attitudes
of the judges themselves. Do the attitudes and backgrounds of the
9 The fact that different patterns may lead to identical scores is not necessarily
as serious a limitation as it might seem. This point is discussed on pages 369-370.
judges affect the position of the various items on the scale? This
obviously is a matter that is open to experimental inquiry. A number
of early studies supported the view that the scale values assigned did
not depend on the attitude of the judges. Hinckley (1932) found a
correlation of .98 between the scale positions assigned to 114 items
measuring prejudice toward Negroes by a group of Southern white
students in the United States who were prejudiced against Negroes
and those assigned by a group of unprejudiced Northern students.
Similarly, MacCrone (1937), in the study of race attitudes in South
Africa referred to earlier, found that the scale positions assigned various
items by South Africans of European background and by educated
Bantus, natives of South Africa, were similar except for a few items.
Studies of the construction of scales measuring attitudes toward a
particular candidate for political office (Beyle, 1932), toward war
(Ferguson, 1935), toward "patriotism" (Pintner and Forlano, 1937),
and toward Jews (Eysenck and Crown, 1949) all found correlations of
.98 or higher between the scale positions assigned to the items by
groups of judges with opposed attitudes towar,d the object of the
scale.
More recent research, however, has sharply challenged the con-
clusions of these studies. Hovland and Sherif (1952), using the items
employed in the Hinckley study mentioned above, found marked
differences between the scale values assigned to items by anti-Negro
white judges on the one hand, and those assigned by pro-Negro white
judges and Negro judges, on the other. Items rated as "neutral" or
moderately favorable by Hinckley's subjects were likely fo be seen as
unfavorable by the pro-Negro white judges and the Negro judges. This
discrepancy between the earlier and the later findings can be accounted
for by the different procedures used. Hinckley followed a rule suggested
by Thurstone, that any judge who placed more than one fourth of the
statements in a single category should be eliminated as "careless."
Hovland and Sherif, however, found that judges with extreme attitudes
tended to place many statements in the same category; checks within
their procedure convinced these investigators that this was nbt a
matter of carelessness. Application of the rule followed by Hinckley
would have eliminated over three fourths of their Negro judges and
two thirds of their pro-Negro white judges; when they did eliminate
these judges, they found that the scale values assigned by the remain-
ing white judges were very close to those assigned by Hinckley's judges.
These findings strongly suggest that Hinckley's procedure had the
effect of ruling out judges with extreme attitudes.
A subsequent study by Kelley et al. (1955), using twenty of the
Hinckley items, found marked differences between the scale values
assigned to items by white and by Negro judges, with the statements
fairly evenly distributed from "favorable" to "unfavorable" by the white
judges, but bunched at the two ends of the continuum by the Negro
judges. lO Granneberg (1955), in constructing a scale of attitudes to-
ward religion, found not only that a religious group and a nonreligious
group differed significantly in the scale values they assigned to items,
but that judges of superior and of low intelligence differed, and that
there was an interaction between attitude and intelligence which
affected the scale position to which items were assigned.
Such findings, of course, cast serious doubt on the meaning of
the scale positions and the distances between them. It should be noted,
however, that even those studies that found marked differences be-
tween groups of judges in the absolute scale values they assigned to
items found high agreement in the rank order in which judges with
differing attitudes arranged the items along the favorable-unfavorable
continuum. Thus, although the assumption that Thurstone-type scales
are true interval scales seems dubious, it is still possible for them to
constitute reasonably satisfactory ordinal scales; that is, they provide
a basis for saying that one individual is more favorable or less favorable
than another. If in practice individuals agreed with only a few contig
uous items, so that a given score had a clear meaning, the Thurstone
methods would provide highly satisfactory ordinal scales. But, as noted
above, individuals m~y agree with items quite widely spaced on the
scale, and in such ca~es the median of the items checked may not
provide a meaningful: basis for ranking the individual in relation to
others.
10 These investigators found that other methods of constructing Thurstonetype
scales were less subject tlian the equal-interval technique to the effect of extreme
attitudes on the part of the judges. The method of successive intervals showed less
difference between whiteland Negro judges, and the method of paired compari~c:1>
eliminated the differences almost entirely.
SUMMATED SCALES
A summated scale, like a differential scale, consists of a series of

items to which the subject is asked to react. However, no attempt is
made to find items that will be distributed evenly over a scale of
favorableness-unfavorableness (or whatever dimension is to be meas-
ured). Rather, only items that seem to be either definitely favorable
or definitely unfavorable to the object are used, not neutral or "slightly"
favorable or unfavorable items. Rather than checking only those state-
ments with which he agrees, the respondent indicates his agreement or
disagreement with each item. Each response is given a numerical score
indicating its favorableness or unfavorableness; often, favorable re-
sponses are scored plus, unfavorable respo~ses, minus. The algebraic
summation of the scores of the individual's responses to all the separate
items gives his total score, which is interpreted as representing his posi-
tion on a scale of favorable-unfavorable attitude toward the object.
The rationale for using such total scores as a basis for placing individ-
uals on a scale seems to be as follows: The probability of agreeing with
anyone of a series of favorable items about an object, or of disagreeing
with any unfavorable item, varies directly with the degree of favorable-
ness of an individual's attitude. Thus, one could expect an individual
with a favorable attitude to respond favorably to many'items (that is,
to agree with many items favorable to the object and to disagree with
many unfavorable ones); an ambivalent individual to respond un-
favorably to some and favorably to others; an individual with an un-
favorable attitude to respond unfavorably to many items. ,
The type of summated scale most frequently used in the study of
social attitudes follows the pattern devised by Likert (1932) and' is
referred to as a Likert-type scale. In such a scale, the subjects are asked
to respond to each item in terms of several 'degrees of agreement 01
disagreement; for example, (1) strongly approve, (2) approve, (3)' un-
decided, (4) disapprove, (5) strongly disapprove. 11 Reproduced below
are several items, with directions, from a Likert-type scale, the so-taIled
"Internationalism Scale" used by Murphy and Likert (1938).
11 Although Likert used five categories of agreement-disagreement, some in-
vestigators have used a smaller and some a larger number of categories, Many
summated scales call simply for an expression of agreement or disagreement, with-
out indication of degree. '
Directions: The following list of sentences is in the form of
what should or should not be done. If you strongly approve of the
statement as it stands, underscore the words "strongly approve,"
and so on, with regard to the other attitudes (approve, unde-
cided, disapprove, strongly disapprove).
18. In the interest of permanent peace, we should be willing to
arbitrate absolutely all differences with other nations which
we cannot readily settle by diplomacy.
Strongly Strongly
approve Approve Undecided Disapprove disapprove
(5) (4) (3) (2) (1)
19. A person who loves his fellow men should refuse to engage
in any war, no matter how serious the consequences to his
country.
Strongly Strongly
(5) (4) (3) (2) (I)
22. We must strive for loyalty to our country before we can
afford to consider world brotherhood.
Strongly Strongly
(I) (2) (3) (4) (5)
The numbers under the scale positions do not appear on the
questionnaire given to the respondents. They are shown here to
indicate the scoring system.
The procedure for constructing a Likert-type scale is as follows:
(1) The investigator ;assembles a large number of items considered
relevant to the attitude being investigated and either clearly favorable
or clearly unfavorable: (2) These items are administered to a group
of subjects representa~ive of those with whom the questionnaire is to
be used. The subjects indicate their response to each item by checking
one of the categories of agreement-disagreement. (3) The responses
to the various items ar~ scored in such a way that a response indicative
of the most favorable attitude is given the highest score. It makes no
difference whether 5 is high and 1 is low or vice-versa. The important
thing is that the responses be scored consistently in terms of thct

attitudinal direction they indicate. Whether "approve" or" disapprove"
is the favorable response to an item depends, of course, upon the
content and wording of the item. (4) Each individual's total score is
computed by adding his item scores. (5) The responses are analyzed
to determine which of the items discriminate most clearly between
the high scorers and the low scorers on the total scale. For example,
the responses of those subjects whose total scores are in the upper
quarter and the responses of those in the lower quarter may be ana-
lyzed in order to determine for each item the extent to which the
responses of these criterion groups differ (see Chapter 5, pages 184-
185 for an illustration of this procedure). Items that do not showa sub-
stantial correlation with the total score, or that do not elicit different
responses from those who score high and those who score low on the
total test, are eliminated to ensure that the questionnaire is "internally
consistent"-that is, that every item is related to the same general at-
titude.
The Likert-type scale, like the Thurstone scale, has been used
widely in studies of morale, of attitudes toward Negroes, of attitudes
toward internationalism, etc. It has several advantages over the Thur-
stone scale. First, it permits the use of items that are not manifestly
related to the attitude being studied. In the Thurstone method, the
necessity of agreement among judges tends to limit items to content
that is obviously related to the attitude in question; in the Likert
method, any item that is found empirically to be consistent with the
total score can be included. Second, a Likert-type scale is generally con-
sidered simpler to construct. Third, it is likely to be more reliable than
a Thurstone scale of the same number of items. Within limits, the re-
liability of a scale increases as the number, of possible alternative re-
sponses is increased; the Likert-type scale item permits the expression of
several (usually five) degrees of agreement-disagreement, whereas the
Thurstone scale item allows a choice between two alternative re~ponses
only. Fourth, the range of responses permitted to an item given in a
Likert-type scale provides, in effect, more precise information about
the individual's opinion on the issue referred to by the given item,12
12 For a detailed comparison of the Thurstone and Likert methods, see
Edwards and Kenney (1946) and Edwards (1957a).
The Likert-type scale does not claim to be more than an ordinal
scale; that is, it makes possible the ranking of individuals in terms of
the favorableness of their attitude toward a given object, but it does
not p.rovide a basis for saying how much more favorable one is than
another, nor for measuring the amount of change after some experience.
From the point of view of the level of measurement we would like our
instruments to provide, this is, of course, a disadvantage. Whether it
constitutes a disadvantage of the Likert scale in comparison with the
Thurstone scale depends on one's judgment of whether Thurstone
scales really meet the criteria for interval scales.
Another disadvantage of the Likert-type scale is that often the
total sc~re of an individual has little clear meaning, since many pat-
terns of response to the various items may produce the same score. We
have already noted that Thurstone-type scales are also subject to this
criticism, but it applies even more strongly to the Likert scales since
they provide a greater number of response possibilities. Using the three
items in our illustration of the Likert-type scale, for example, an in-
dividual may obtain a total score of 6 by indicating: (a) disapproval of
18 and 19, approval of 22; (b) approval of 18, strong disapproval of
19, and strong approval of 22; (c) indecision on 18, disapproval of 19,
and strong approval of 22; (d) other combinations of responses. It
seems reasonable to suppose that, although the total scores are the
same, their meanings may be markedly different. Thus one may raise
a serious question whether the Likert-type scale actually conforms to
the requirements of an ordinal, much less an interval, scale. Despite
the lack of theoretical rationale for scalability, however, pragmatically
the scores on the Likert-type questionnaire often provide the basis for
a rough ordering o,f people on the characteristic being measured.
The fact that ~ifferent patterns of response may lead to identical
scores on either a: Thurstone or a Likert scale is not necessarily as
serious a drawback as it may at first appear. Some of the differences in re-
sponse patterns leading to a given score may be attributable to random
variations in respotise. Others may arise because specific items involve
not only the attitude being measured but also extraneous issues that
may affect the response. Thus some of the differences in response
patterns leading t.o the same score may be thought of as error from
the point of view lof the attitude being measured, rather than as true
differences in attitude that are being obscured by identical scores. The

fact that the scale contains a number of items means that these varia-
tions on individual items unrelated to the attitude being measured
may cancel each other out.
Moreover, different ways of getting to the same place may be
equivalent from the point of view of the measurement goal that is
being served. For example, if one weights addition and subtraction
equally in a concept of arithmetic ability, it makes sense to score two
individuals as equivalent in arithmetic ability, even though one is
relatively strong in addition and the other relatively strong in subtrac-
tion. Similarly, it may make sense to say that the net degree of ani-
mosity toward a given attitudinal object is the same in two individuals
even though the animosity expresses itself differently.
The problem is to determine when the fact that the same score
can be ar"rived at in different ways has consequences for the meaning-
fulness of the score, and when it does not. In part, this problem is one
of conceptual clarity; in part, it involves questions of fact. If the inves-
tigator is not clear about what he is trying to _measure, and why, this will
be only one of many problems with which he will be unable to cope.
But even if his concepts are clear, he will still want to know (although,
unfortunately, he may not be in a position to find out) the answers
to such questions as: Do the response patterns of individuals remain
i stable over time? If alternate forms of the test are available, do indi-
dividuals receive the same scores on different forms? Do different
individuals achieving the same score in different ways react in the same
way to particular stimuli, problems"incentives, etc.?
Ultimately what is involved is a question of the validity of the
scale. Questions of validity always involve questions of fact, which
cannot be settled by armchair argument. The problem of whether
different combinations of responses can meaningfully be assigned the
same score is one for empirical investigation.
CUMULATIVE SCALES
Cumulative s,cales, like differential and summated scales, are made

'up of a series of items with which the respondent indicates agreement
or disagreement. In a cumulative scale, the items are related to one
another in such a way that, ideally, an individual who replies favorably
to item 2 also replies favorably to item 1; one who replies favorably to
item 3 also replies favorably to items 1 and 2; etc. Thus, all individuals
who answer a given item favorably should have higher scores on the
total scale than the individuals who answer that item unfavorably. The
>ndividual's score is computed by counting the number of items he
answers favorably. This score places him on the scale of favorable-
unfavorable attitude provided by the relationship of the items to one
another.
Sometimes the items as they appear in the scale are arranged in
order of favorableness; sometimes they are randomly arranged. Ordi-
narily, no attempt is~made to determine whether the intervals between
items are equal; thus, in practice, cumulative scales are ordinal scales.
One of the earliest scales used in the measurement of attitudes,
the Bogardus social-distance scale (see Bogardus, 1925, 1928, 1933)
was intended to be of the cumulative type. The social-distance scale,
which has become a classic technique in the measuring of attitudes
toward ethnic groups, lists a number of relationships to which mem-
bers of the group might be admitted. The respondent is asked to in-
dicate, for specified nationality or racial groups, the relationships to
which he would be willing to admit members of each group. His atti-
tude is measured by the closeness of relationship/that he is willing to
accept. The Bogardus-type scale is illustrated below:
Directions: For each race or nationality listed below, circle
each of the classifications to which you would be willing to admit
the average member of that race or nationality (not the best
members you have known, nor the worst). Answer in terms of
your first feeling reactions.
I
I
To Toem- As
close 'romy ployment To citi- visitors Would
kinship tlub as To my in my zen ship only exclude
by pfrsonal street as occupa- in my tomy from my
marriage c;:hums neighbors tion country country country
English 1 2 3 4 5 6 7
Negro 1 2 3 4 5 6 7
French 1 I, 2 3 4 5 6 7
Chinese 1 2 3 4 5 6 7
Russian 1 2 3 4 5 6 7
etc.
The items used in the Bogardus scale (that is, the column headings
in the i11ustration above) were selected on logical grounds. It seems
reasonable to expect that an individual who circles 4 in relation to
Chinese, indicating that he would be wi11ing to accept them to em-
ployment in his occupation, would ordinarily also circle 5 and not
circle 6 or 7. (Here, as in other scales, the content of the item must be
taken into account in deciding whether a "Yes" response is to be scored
as favorable or unfavorable. Since 6 and 7 are essentially statements of
exclusion, absence of a circle constitutes the favorable response to
these two items. Thus, neither 6 nor 7 should be circled for a given
group if any of the other numerals is circled.) If the individual did not
circle 3 (willing to admit to my street as neighbors), one would expect,
on logical grounds, that he would also not circle 2 or 1.
On the whole, the assumption that these items form a cumulative
s{ale has been borne out. Nevertheless, in practice some reversals c;lo
occur. Some individuals, for example, who would object to living in a
buildingwitH ~uerto Ricans would not object to having Puerto Ricans
in an informal social club (see Deutsch and Collins, 1951). Although
individuals not infrequently show such reversals in replies on the social-
distance scale, it is relatively uncommon to find an entire group revers-
ing items. Thus, the social-distance scale has been used rather effect-
ively in comparing the attitudes of different groups of people toward
various nationalities. It may be noted that re.versals can almost always
be interpreted by postulating the intrusion of some factor other than
the individual's own attitude toward the group in quest~on-e.g., the
respondent's image of hqw other people would interpret his living in
a certain neighborhood, or his expectation concerning the impact on
real estate values of admitting minority group members to residence
on his street, etc.
With the appearance of the Thurstone and Likert scaling methods
in the late nineteen-twenties and early thirties, attention shifted away
from cumulative scales. However, the forties saw a revival of interest
and a rapid development of techniques for determining whether the
items of a scale do in fact have a cumulative relationship, regardless of
whether they appear cumulative in common-sense terms. This renewed
interest was linked to an emphasis on the development of unidimen-
sional scales-that is, scales consisting of items that do not raise issues,
or involve factors, extraneous to the characteristic being measured.
A number of investigators had pointed out that the Thurstone and
Likert scales, although ostensibly measuring "an attitude," contained
statements about various aspects of the object under consideration.
Thus, Carter (1945) pointed out that Form A of the Peterson scale of
attitude toward war (a Thurstone-type scale) had as its most favorable
statement, "War is glorious"; as its most unfavorable statement, "There
is no conceivable justification for war"; and as its mid-point, "I never
think about war and it doesn't interest me." He commented that it is
difficult to think of these statements as falling along a straight line. He
suggested that such statements as, "The benefits of war rarely pay for
its losses even for the victor" and "Defensive war is justified but other
wars are not," belong on two different scales, one having to do with the
economic results of war, the other with the ethics of war activity. It
was argued that combining items referring to different aspects of the
object made it impossible to specify exactly what the scale was measur-
ing, and also accounted for the scattering of responses, which made
it difficult to assign any clear meaning to the score based on the median
of the items checked. /
There have been several approaches to this problem. We shall
discuss here only the technique developed by Guttman, commonly
called scale analysis or the scalogram method. 13 One of the main pur-
poses of this technique is to ascertain whether the attitude or charac-
teristic being studied (technically termed the "universe of content"
or the "universe of attributes") actually involves only a single dimen-
sion. In the Guttma~ procedure, a "universe of content" is considered
to be unidimensional,only if it yields a perfect, or-nearly perfect, cumu-
lative scale-that is, if it is possible to arrange all the responses of any
number of responderj.ts into a pattern of the following sort:
13 For a more comprehensive discussion of the Guttman technique, see

Stouffer et a!. (1950). For critiques of it, and alternative approaches to the same
problem, see Festinger (1947) and Loevin,eer (1948).
Says "Yes" to Says "No" to

Score item item
3 2 1 3 2 1
3 x x x
2 x x x
1 x x x
0 x x x
The important thing about this pattern is that, if it holds, a given

score on a particular series of items always has the same meaning;
knowing an individual's score makes it possible to tell, without con-
sulting his questionnaire, exactly which items he endorsed. Consider,
for example, the following items, with which respondents are asked
either to agree or to disagree:
1. A young child is likely to face serious emotional problems if

his parents get divorced. _
2. Even if a husband or wife or both are unhappy in their mar-
riage, they should remain together as long as they have any
young children.
3 Divorce laws in this state should be changed to make it more
difficult to get a divorce.
If these items were found to form a perfect cumulative scale, we

would know, for example, that all individuals with a sCOre Qf 2 on the
scale believe that divorce of the parents presents serious emotional
problems for a young child and that a couple with young children
should remain together even if they are unhappy, but do not believe
that the divorce laws should be made ~ore stringent.
In practice, perfect cumulative, or unidimensional, scales are rarely
or never found in social research, but approximations to them can
often be developed. Scalogram analysis uses several criteria for decid-
ing whether or not a particular series of items may be usefully re-
garded as approximating a perfect unidimensional scale. The- most im-
por~ant of these is th~ reproducibility of the responses-the propqrtion
of responses of a large number of subjects which actually fall into the
pattern presented above.14 This pattern contains an the responses to
particular items that would be predicted from a knowledge of the indi-
vidual's total score on the series of items (his "scale type"). Thus the
proportion of actual responses which fall into the pattern provides a
measure of the extent to which particular responses are "reproducible"
from the total score. Guttman and his co-workers have set .90 as the
minimal reproducibility necessary for a series of items to be regarded
as approximating a perfect scale. Examples of such scales are presented
in Stouffer et al. (1950).
The Guttman technique is a method of determining whether a
set of items forms a unidimensional scale; as a number of writers have
pointed out, it offers little guidance for selecting items that are likely
to form such a scale. Edwards and Kilpatrick (1948) 15 have suggested a
method of selecting a set of statements likely to form a unidimensional
scale. Called the scale-discrimination technique, it combines aspects of
the Thurstone and Likert approaches to scale construction, in the
following steps: (1) A large assortment of items dealing with the issue
of study is collected. Items that are ambiguous, irrelevant, neutral, or
too extreme are eliminated by inspection. (2) As in the Thurstone
method of equal-appearing intervals, a large number of judges place
the remaining items in eleven piles, according to their Judgea favor-
ableness or unfavorableness toward the issue. The extent to which the
judges agree on the placement of each item is determined, and the
half of the items on which there is greatest variability or scatter of
judgments is eliminated. Each of the remaining items is assigned a
scale value corresponding to the median position in which it has been
placed by the judges. (3) These items are then transformed into a
Likert-type scale by providing for the expression of five or six degrees
of agreement-disagreement in response to each item. This scale is
administered to a largei group of subjects, and their responses are an-
alyzed to determine w~ich of the items discriminate most clearly be-
tween the high scorers 'and the low scorers on the total scale. The re-
sulting "discriminatory coefficients" of the various items are then
plotted against their scale values. From the total list of items, twice the
14 For a detailed discJssion of this and other methods of determining whether
a scale is unidimensional, see White and Saltz (1957).
15 This article appears also, with minor changes, in Edwards (1957).
number wanted in the final scale are selected. The items selected are
those which have the highest discriminatory coefficients in their scale
interval; for example, of all the items with scale values between 8.0 and
8.9, those with the highest discriminatory coefficients are selected. An
equal number of items is selected for each interval. (4) The items
in the resulting list are arranged in order of their scale value. The list
is then divided into two equated forms of the questionnaire by assigning
all the odd-numbered items to one form and all the even-numbered
items to the other.
The Guttman and related techniques represent major contribu-
tions to the methoddlogy of questionnaire construction and analysis.
However, two qualifications related to the use of unidimensional scales
should be kept in mind: (1) Such a scale may not be the most effec-
tive basis either for measuring attitudes toward complex objects or for
making predictions about behavior in relation to such objects; (2) a
given scale may be unidimensional for one group of individuals but not
for another.
Let us consider the first reservation. Suppose we have devised a
unidimensional scale to measure attitude toward the economic results
of war, another to measure attitude concerning the ethics of war ac-
tivity, still others to measure whatever other aspects of attitude toward
war can be identified and measured by unidimensional scales. No single
one of these scales may give an accurate reflection of an individual's
attitude toward the complex concept "war," or provide a basis for pre-
dicting how he would vote on the question of his country's participa-
tion in a specific war. This is, of course, the same qualification noted
in connection with the discussion of internal consistency in Chapter
5, pages 178-179; a complex measure may be needed as a basis' for
predicting complex behavior.
As for the second reservation, it is sometimes assumed that uni-
dimensionality is a property of a measuring instiument, rather than of
the patterning of an attitude among a given group of individuals. For
one group, a number of items may be arranged unidimensionally in a
given order; for another group, the same items may fall into a different
order; for still another group, they may not form a unidimensional pat-
terti at all. The way i~ which the experiences of different groups can
lead to different patternings of items is illustrated in a study by Harding
and Hogrefe (1952). These investigators interviewed three groups of
white department-store employees. The members of Group 1 worked
in departments in which there was at least one Negro in a job equal
in status to their own, or of higher status than their own; those in
Group II worked in departments in which all the Negroes were in jobs
of lower status than their own; those in Group III were in departments
where there were no Negroes. The interviews included six "social-dis-
tance" questions, having to do with: sitting next to Negroes in buses
or trains, sitting at the same table with a Negro in a lunchroom, taking
a job in which there were both Negroes and white people doing the
same kind of work as you, working under a Negro supervisor, living in
a building in which there were both white and Negro families, and
having a Negro for a personal friend. The investigators found that
these six questions formed satisfactory Guttman-type scales for each
of the three groups, but that the question about taking a job in which
there were both Negroes and white people doing the same kind of
work as the respondent fell in a different position for each of the three
groups. For Group I-the people who were actually in this situation-
this question tied with the one about buses and trains for the "most
acceptable" position. For Group II-those working in departments
with Negroes, but in positions of unequal status-sitting next to
Negroes in trains and buses was more acceptable than working with
them on an equal status. For those in all-white departments, both
sitting next to Negroes in buses and trains and sitting at the same
table with a Negro in a lunchroom were more acceptable than working
with them on an equal status.
In other, words, as Coombs (1948) has pointed out:
. . . in a highly organized social order with standardized
education, there will tend to be certain traits generated which
will be common to the population subjected to the same pattern
of forces. There is, however, at the same time, opposition, con-
tradiction, and interaction of these forces on organisms that are
not equally endowed in the first place-with the result that the
structuring of a psychological trait is less complete in some
individuals than in others . . . . A psychological trait, in other
words, mayor may not be a functional unity and it mayor may
not be general, i.e., common to a large number of individuals.
Some Modifications of Scaling Techniques

Not all attempts to quantify attitudes fit into the classification of
scales we have described. Especially within recent years, a number of
approaches have been developed that use some aspects of scaling pro-
cedure but cannot properly be described as differential, summated, or
cumulative scales. We shall discuss two such approaches: the Q-sort
and the semantic diHerential.
The operations involved in a Q-sort are similar to the first steps

in the construction of a Thurstone scale by the method of equal-ap-
pearing intervals. The subject is presented with a large number of state-
ments believed to be relevant to the topic under investigation, and is
asked to sort them into a specified number of piles-usually nine or
eleven-according to some criterion. In the Thurstone technique, it
will be remembered, this sorting process has the -purpose of assigning
scale values to statements; the criterion for sorting is not the sorter's
agreement or disagreement with a statement but his judgment of its
degree of favorableness or unfavorableness toward the object. In the
Q-sort, the purpose of sorting is to get a picture of the individual's own
view of, or attitude toward, the object being considered. The criterion
for placing statements in the various piles is the extent of his agreement
with them; the pattern into which he sorts the statements constitutes
the data for analysis of his position. A restriction not present in the
Thurstone sorting procedure is introduced into the Q-sort; this is speci-
fication of the number of cards to be placed in a given pile. In order to
simplify the statistical analysis, the numbers are usually specified in
such a way that the sort forms a roughly normal distribution.
Let us consider an example of the use of a Q-sort in the context
in which it has most frequently been employed-the study of person-
ality. More specifically, this was a study of changes in an individua1',s
image of himself, of his ideal person, and of "the ordinary person," and
16 For a detailed discussion of "Q methodology," see Stephenson (1953). For
shorter discussions, see Cron bach (1953) and Mowrer (1953) .
MODIFICATIONS OF SCALING TECHNIQUES 379
of changes in the relations among these images, as a result of psycho-
therapy (Rogers and Dymond, 1954). Each subject was given a set of
one hundred ~tatements of personal characteristics, each printed on a
separate card. The set consisted of such statements as: "I am a sub-
missive person . . . I am a hard worker . . . I really am disturbed
. . . I am afraid of a full-fledged disagreement with a person . . . I
am likable." The subject was asked first to "sort these cards to describe
yourself as you see yourself today, from those that are least like you
to those that are most like you." The distribution was to be in nine
pilee, with one card in each of the two extreme positions, four cards in
each of the next most extreme positions, and so on. Next the subject
was asked to "sort these cards to describe your ideal person-the person
you would most like within yourself to be." Finally, he was asked to
sort them to describe "the ordinary person." The three sortings
were carried out first before the person entered psychotherapy and
again after a period of treatment.
The appropriate statistical techniques to be used in the analysis
of data from Q-sorts are a matter of some controversy. Some authors
have held that factor analysis 17 is essential; others have disagreed. (For
a presentation and discussion of these points of view, consult the refer-
ences cited on page 378.) In the study we are considering, factor
analysis was not used. Rather, relations such as the following were in-
vestigated by means of correlations: the correspondence between self-
image and ideal image before and after therapy, with the finding that,
on the whole, an individual's ideal and his self-image were more alike
after t~erapy; differences in the extent of change in the self-image,
the ideal image:, and the image of the ordinary person, with the finding
that the self-image changed more during therapy than the other
images; etc.
In this study, the dat~ yielded by the Q-sort were _also summarized
into a single score, as in aisummated scale, to yield a score on "adjust-
ment." This latter procedure made possible (although this was not its
main purpose) estimatiorl of the reliability and validity of the instru-
ment used in this way. It was found that the "adjustment" scores of a
t,
17 Factor analysis is a method of finding the common element or elements
that underlie a set of measures. For a discussion of factor analysis, see Cattell
(1952) or Thomson (1946).
control group of individuals who did not receive therapy but who
were tested several times showed a s~lity coefficient of .86 computed
on the basis of two administrations of the instrument at least six
months apart. Comparison of the "adjustment" scores with therapists'
ratings of the success of therapy showed considerable, though by no
means complete, agreement.
Although Q-sorts have been used most often in studies such as
this, where the emphasis is on self-image and other person-images, the
method is applicable to the study of other attitudes. Subjects might be
presented with sets of statements about methods of child-rearing, or
about labor-management relations, or about Negroes, and asked to
sort them in terms of the extent of their agreement or disagreement
with each statement. The resulting data might be used, for example,
to compare a given individual's views about different ethnic groups,
or to compare the views of different individuals about a given group
m~~
THE SEMANTIC DIFFERENTIAL'
Osgood, Suci, and Tannenbaum (1957), who developed the "se-

mantic differential," describe it as a method for measuring the meaning
of an object to an individual. It may also be t1}ought of as an attitude
scale. The subject is asked to rate a given concept (e.g., "Negro,"
"Republican," "wife," "me as I would like to be," "me as I am," "Pi-
casso's Guernica") on a series of seven-point bipolar rating scales. Any
concept-whether it is a political issue, a person, an institution, a work
of art-can be rated. The seven-point scales include such bipolar scales
as the following: (A) fait-unfair, clean-dirty, good-bad, valuable-worth-
less; (B) large-small, strong-weak, heavy-li-ght; (C) active-passive, fast-
slow, hot-cold.
One may use the individual's responses to determine whether, for
him, two concepts are alike or different. For example, does his picture
of "me as I am" coincide with his picture of "me as I would like to be,"
or are the two quite different? In this use, one draws a "profile" of the
meaning of each concept to the individual simply by drawing lines be-
hyeen the points checked on each of the scales for a given concept. A
technique for measuring the extent of similarity between the two pro-
.files has been developed. Similarly, one may compare two individuals'
concepts of a given object by measuring the similarity of the profiles
provided by their checks on the various scales.
Used in this way, the semantic differential does not place individ-
uals on an underlying scale representing some dimension of attitude;
it simply provides a method of measuring the similarity or difference
between their concepts of a given object. However, the responses to
subgroups of the scales can be summated to yield scores that are inter-
preted as indicating the individual's position on three underlying di-
mensions of attitude toward the object being rated. These dimensions
have been identified by using factor-analytic procedures in examining
the responses of many individuals concerning many concepts or objects.
It has been found that, more or less consistently, the scales labeled (A)
on page 380 seem to group together; that is, an individual tends to
place a given object in a similar position on each of these sca1es. Simi-
larly, the scales labeled (B) seem to group together; so do the scales
labeled (C). The manifest content of each scale does not clearly indi-
cate the underlying attitudinal dimension to which it is relevant; what
does it mean, for e~ample, to rate Picasso's "Guernica" as hot or cold,
as clean or dirty? But from the consistent grouping of the scales, Osgood
and his colleagues have inferred that the three subgroups measure the
following three dimensions of attitude: (A) the individual's evalua-
tion of the object or concept being rated, corresponding to the favor-
able-unfavorable dimension of more traditional attitude scales; (B)
the individual's perception of the potency or power of the object or
concept; and (C) his perception of the activity of the object or concept.
The diagram on the following page is an adaptation of one given
by Osgood and his colleagues, illustrating the use of the semantic differ-
ential in plotting the meaning of several different concepts to a group
of Taft Republican~ in a: study that was made shortly before the 1952
presidential election. From the diagram one can see, for example, that
to these subjects theI concept of "Truman" is very different from the
concept of "Taft" bUt is very similar to the concept of "socialism." The
concept of "Stalin" 'is similar to the concept of "Taft" on the potency
and activity dimensions but very different on the evaluative dimension.
Osgood and h~s colleagues suggest that the semantic differential
makes possible the: measurement and comparison of various objects
Model of, the "Semantic Space" lor Talt Republicans18

Each pI'eight concepts is numbered according to the key
beI6.w. The point at which the line for a given concept starts
shows' the rating of that concept on the scales fair-unfair and
strong:weak. The length of the line shows the rating on scale
active-passive; a solid line indicates a rating toward the active end
of the scale, a broken line a rating toward the passive end. Thus,
for example, it can be seen that Stalin (4) was rated as very
unfair, very strong, and very active; Truman (5) as somewhat
unfair, somewhat weak, and somewhat active.
Key
1. Taft 5. Truman
2. Policy in China 6. Atombomb
3. Socialism 7. United Nations
,4. Stalin 8. Eisenhower
by diverse subjects; they imply that the'measuring instrument is not

grossly affected by the nature of the object being measured or by the
! type of person using the scale. If this is "realIy true, the semantic differ-
ential would be a solution to many of the problems of attitude meas-
18 Adapted from Osgood et al. (1957).
urement. Osgood, Suci, and Tannenbaum provide some evidence that
different types of subjects use the scales in similar ways. Thus, they
indicate that Taft Republicans, Eisenhower Republicans, and Steven-
son Democrats use the same underlying dimensions in their reactions;
so do normals and schizophrenics; so do Korean exchange students,
Japanese exchange students, and Americans. This is impressive evi-
dence that the scales are comparable across different types of subjects,
although it seems possible that if people with widely different educa-
tional backgrounds were compared, the same results might not obtain.
However, Osgood and his colleagues present considerable evidence
indicating that the measuring instmment is not completely compar-
able across concepts. The meanings of scales and their relation to the
other scales vary considerably with the concept being judged. What
is good, for example, depends heavily on the concept being judged;
"strong" may be good in judging athletes but not in judging women.
The implication of this is that it may be quite difficult to develop rating
scales that provide a consistent measurement of the underlying di-
mensions independently of the concepts being judged.
In concluding our discussion of attitude scales, it should be
pointed out that the field of scale construction and analysis is in a
period of rapid development and change. In addition to the work we
have discussed, major contributions have been l)1ade in recent years
by Coombs (1948, 1950, 1953), Lazarsfeld (1957), Loevinger (1947,
1948), Suppes and his co-workers at the Applied Mathematics and
Statistics Laboratory of Stanford University (Davidson, Suppes, and
Siegel, 1957), Torgerson (1958), and others.
A Concluding Note
1
In this chapter )Ve have discussed various methods of scaling, that
is, of distinguishing among objects or individuals in terms of the degree
to which they possJss a given characteristic. Here, as in connection
with other measurement techniques, we have raised questions about
reliability and validity. Although many users of scales have investigated
the reliability and ~alidity of their measures, it is probably still true,
as one writer (Ferguson, 1957) has remarked, that there has been
"more measurement than validation." As we have pointed out earlier,

there are understandable reasons for the lack of attention to validity-
notably the difficulty of determining what would be appropriate cri-
teria of validity for measures of complex attributes and of gathering
the necessary d'lta. However, a number of investigators have demon-
strated that, with ingenuity, one may develop useful criterion measures.
It is perhaps to be expected that in relatively new research fields, such
as the social.sciences, more attention should at first be paid to the
development of a variety of measuring techniques than to their valida-
tion. Nevertheless, it seems clear that research in the social sciences
would now be advanced "\:>y greater attention to the validity of the in-
struments used.
Scaling procedures may either be built directly into the data-
collection instruments or may be used in the course of analysis of data
collected in other ways. In the next chapter we shall consider the
problems and procedures of analyzing data collected by whatever
method in order to abstract from them the relevant information for
answering research questions.
11
ANALYSIS AND
INTERPRET ATION
The Influence of Anticipated Analysis and Interpretation

on Previous Research Steps
The Establishment of Categ9ries

Coding: The Categorization of Data
Tabulation
Statistical Analysis of Data
, Inferring Causal Relations
1
The Use of Nonquantified Data in Analysis and Interpr~t.a~~oI1

A l?TER THE DATA have been collected, the social scientist turns
ft his full attention to their analysis and interpretation, a process
consisting of a number of closely related operations.)t is the purpose
of analysis to summarize the completed observations in such a manner
that they yield answers to the research question~\ It is the purpose of
interpretation to search for the broader meanin~ of these answers by
linking}hem to other available knowledge. Both these purposes, of
course, govern the entire research process; all preceding steps have been
undert~ken in order to make th~iffulfillment possible~1
A '
The InBuen.ce oE Anticipated Analysis and Interpretation

on Previous Research Steps
Before we discuss the processes involved in analysis and interpre-

cation, let us consider a specific study from the point of view of the
ways in which concern for analysis and interpretation enters into the
earlier s~ages of an inquiry. We shall use as an example a nonexperi-
mental study designed to investigate the effects of association with
Arpericans on attitudes of foreign students toward Americans and
toward various aspects of life in the United States (Selltiz, Hopson, and
Cook, 1956). _
The study was planned against Ithe background of a number of
investigations which had .suggested that, under certain conditions~
when members of one racial or religious group within the United States
are brought into face-to-face contact with members of a different.group
toward which they have been prejudiced, their attitudes are likely to
become more favorable. The interest in this study was in findIng
whether association between individuals of different nationality or
cultural background would also lead to favorable attitude changes.
More specifically, the question was phrasoo: Do foreign students who
have more, and more intimate, interaction with Americans tend to
develop more favorable attitudes toward the United States than for-
386
INFLUENCE ON RESEARCH STEPS 387
eign students who have less, and less intimate, interaction with Amer-
icans?
The ideal study design would have called for selecting a random
sample from the total number of foreign students entering this country
in a given year and then randomly assigning these students to situations
in which they would have varying degrees and kinds of associations
with Americans. Obviously, this was not feasible. The compromise
solution that was arrived at was to select educational settings that
seemed likely to differ in the extent to which they would provide op-
portunities for interaction between foreign students and Americans.
Specifically, it was hypothesized that small colleges in small towns
would provide relatively many opportunities for frequent and intimate
contact, for the following reasons: Students at such colleges typically
live in dormitories or fraternity houses, where the daily opportunities
for interaction are great; the social life is likely to be concentrated on
the campus, again increasing opportunities for interaction between
American and foreign students; there are likely to be relatively few
other foreigners, either students or comm~nity members, so that there
is little possibility for a foreign student to carry on all his social life
within a foreign group. It was thought that large universities in metro-
politan centers would provide considerably less opportunity for close
association with Americans, and that large universitieS'in small towns
would fall between these two extremes.
Students attending these three types of educational institution
were to be compared, then, in the following terms: (1) the. extent to
which. the situations they were in encouraged interaction with Amer-
icans, (2) the extent to which they entered into interaction with
Americans and the nature of these interactions, and (3) their attitudes
I
toward Americans and the United States. However, the way in which
the students were select~d for the study-taking individuals who were
attending institutions of these three types-left open the possibility
that students might have chosen different types of educational insti-
tution because of the different possibilities they offered for association
with Americans. In orqer to check whether the three groups were in fact
comparable to begin with in their inclination to seek out association
with Americans, it was ,decided to ask the students their reasons for
coming to the United States, their reasons for attending the particular
388 ANALYSIS AND INTERPRETATION
college or university, whether they had chosen their living arrangements

or been assigned to them, whether they preferred being at a college
with relatively many or few other students from their home country,
and how many Americans they expected to get to know. In order to
make sure that students in the three types of institution did not differ
in their initial attitudes toward the United States, it was necessary
to secure measures of these attitudes as soon as possible after the stu-
dents' arrival in this country (ideally, before their arrival).
If comparisons on these measures showed that students in the
three types of institution did not differ in their initial attitudes nor in
, their inclination toward association with Americans, they could then
be compared in terms of the extent to which the situations they found
themselves in Ifavored interaction with Americans-that is, in terms
of the interaction-potential of the situations in which they were placed.
It was decided that two measures of interaction-potential would be
used: the nature of the student's living arrangements, and the fre-
quency with which he was in various situations in which there were
Americans whom he knew wen enough to speak to.l
If it turned out that students in different types of institution were
in situations with different interaction-potential, it would then be legit-
imate to consider whether students in situations with relatively high
interaction-potential do in fact interact more with Americans. -This
brought up the question of how interaction was to be measured: in
terms of the number of Americans the student knew? the variety of
situations in which he participated with them? the nature of the activ-
ities in which he engaged with them? his own judgmen-t as to whether
he had made any close friends? Earlier studies offered little guidance
as to what would constitute the most satisfactory measure of inter-
action. Since one of the goals of the study was to try to analyze what
it is about personal interaction thdt leads to attitude change (assuming
a relation between interaction and attitude were found), it was decided
that it would be desirable to have several different measures of interac
tion.
Next, if it was found that students in situations with greater inter-
action-potential ~lso scored higher on actual interaction-thus permit-
1 Note that these measmes constituted the "working definitions" of the con-
cept of interaction-potential.
ting the conclusion that differences in extent and nature of interaction
were not primarily a matter of personal choice-it would be appropriate
to consider whether students with different interaction scores differed
in their attitudes toward the United States. When it came to measur-
ing attitudes, a great deal of prior work was available as background.
This help"ed in two ways: by suggesting aspects of the United States
which are of special concern to foreign students; and by indicating that
provision should be made for measuring various aspects of attitude-
information, opinions, feelings, understanding.
The fact that the students came from a variety of countries, to-
gether with, the impression based on earlier exploratory studies that
students from different countries react differently to the United States,
led to planning in advance for another aspect of the analysis. It was
decided that it would be valuable not only to compare the total groups
of students in the three types of educational institution, but to con-
sider separately the responses of students from countries that were
judged to be relatively similar in culture to the United States (that is,
countries of Europe and English-speaking countries) and those of
students from countries that were judged to offer relatively sharp cul-
tural contrasts to the United States (countries of Latin America, the
Near and ,Middle East, Asia, and Africa) .
To a considerable extent, then, the analYJis of this study was
shaped before the data were collected; and the anticipation of the
future task of analysis and interpretation determined what went be-
fore. Although the extent of early concern with these two subsequent
steps will vary from study to study, it is safe to say that no scientific
inquiry should be planned without anticipating what will be done when
the data are in or without being concerned with the possible interpreta-
tion of findings. Otherwise the investigator is likely to discover when it
is too late that he cannot perform the analysis he wants be.cause relevant
data are missing.
The process a~d direction of interpretation at various steps of an
inquiry will undoubtedly bear the mark of the investigator's knowl-
edge, imagination, and wisdom. No one study can ever plan to follow
through on all interpretive ideas that emerge in the course of its con-
duct. But this sitJation is hardly improved by postponing concern
with interpretation to later stages, when there is little chance for any-
390 A~ALYSIS AND INTERPRETATION
thing but speculation in attempts to circumscribe the meaning of one's

results or to discover alternative explanations.
The relationship between analysis and interpretation, and the
particular form they take separately and jointly, vary from study to
study. Problems of analysis and interpretation differ, in different study
designs. As a rule, they present greater difficulties in exploratory than
in experimental investigations. However, since the basic principles in-
volved in analysis-even though not all the specific procedures-apply
to studies of every type, we shall not discuss different research designs
separately except in cases where a certain design presents special prob-
lems in connection with a given analytic operation.
Since interpretation is often inextricably interwoven with analysis,
so that it becomes a special aspect of analysis rather than a distinct
Qperation, it may be well to precede the discussion of analytic pro-
cedures with a clarification of the process of interpretation.
(! As noted earlier, interpretation is the search for the broader mean-
ing of the research findings. This search has two major aspects. First,
there is the effort to estavlish continuity.in social research through
linking the results of one study with those of another. In the study of
the attitudes of foreign students which we have been describing, the
attempt was made to establish continuity between findings about the
effects of personal association between members of diffefent racial or
religious groups within the United States and findings about the effects
of personal association between members of different national or cul-
tural groups when the members of one of the groups are in a foreign
country. In a somewhat different, sense, interpretation~ is involved in
the transition from exploratory to experimental work. The interpreta-
tion of the former often leads to hypotheses for the latter..
Secondly, interpretation leads to the establishment of explanatory
concepts. Had it turned out that forejgn students who engaged in
greater interaction with Americans were more favorable in their atti-
tudes or showed greater attitude change, the investigators hoped to be
able to deduce something about the process by which personal associa-
tion ~ffects attitudes. Had the attitude differences be~n gre~test be-
tween students who differed in the variety of their experiences with
Americans, the investigators might have tentatively drawn the con-
clusion that association changes attitudes by breaking down stereo-
types. Had the attitude differences been greatest between students
who differed in the intimacy of the activities engaged in with Amer-
icans, the inference might have been that association influences atti-
tudes by giving the individual an opportunity to observe qualities of
people that are not apparent in more superficial contacts. Had the
difference been greatest between students who said they had made a
close American friend and those who had not, the deduction might
have been that it is through increasing the warmth of personal feeling
, that association leads to attitude change. This function of interpreta-
tion is so closely linked to the function of theory for social research
that it will be dealt with in Chapter 14, which is concerned with the
relation between research and theory.
Our discussion of the procedures used in analysis and interpreta-
tion will cover the fonowing topics: the establishment of categories,
the application of ctLt~gories to the raw data through coding, the
tabulation at responses, statistical analysis ot the data, drawing infer-
ences about causal relations, and the use of nonquantified data.
The Establishment of Categories

CLASSIFICATORY PRINCIPLES AS THE BASIS FOR SETS OF CATEGORIES
/
In a neighborhood survey, a sample of the population was asked

the following question: "What kinds of people live in this neighbor-
hood?" The purpose of the question was not to obtain objective in-
formation about the neighborhood (which could have been obtained
more accurately and with less effort from census figures), but rather
to ascertain to what extent the people in the neighborhood thought in
terms of ethnic gro'up distinctions. Here are some of the answers:
I
Mostly poor people like myself.
Colored people ~nd
I
us.
Negroes and Italians and Jews and a lot of others.
There's plenty to say about the people around here. But I don't
want to get i~to trouble. You have to live where you can.
Colored people.
It's a tough neighborhood. All kinds of people live here.
There are some Irish, I know.

Many Jews.
A lot of niggers.
I don't really know, I'm new around here.
Clearly, if several hundred such responses are to be organized so
that they can be used in answering the research questions, they must be
grouped into a limited number of categories. In order to decide on the
relevant categories, some principle of classification must be selected.
The research question, or the hypotheses if any have been formulated,
provide the basis for selecting principles of classification. Suppose that
this study was concerned with awareness of the Negro group; the ap-
propriate principle of classification would be one based on explicit
reference to Negroes. This classificatory principle immediately suggests
two categories:
Mentioned Negroes
Did not mention Negroes
These two categories form a "category set." A category set musl
meet certain basic rules:
1. The set of categories should be derived from a single classifi-
catory principle.
2. The set of categories should be exhaustive; that is, it should
be possible to place every response in one of the categories of
the set.
3. The categories within the set should be mutually exclusive;
it should not be possible to place a given response in more
than one category within the set. .
A set may consist of more than two categories, provided, these
rules are not violated. For example, a :perusal of the answers listed
above soon reveals that "Did not mention Negroes" includes many
diverse responses. The person who did not want to talk about the
neighbors for fear of trouble, or the one who sail! he was new to the
neighborhood, is quite different from those who mentioned groups
other than Negroes. This suggests the need for subcategories to
distinguish among the different types of response that did not mention
Negroes. For example, it may seem psychologically meaningful to
THE ESTAB~ISHMENT OF CATEGORIES 393
distinguish people who did not mention Negroes but did mention
other groups from people who did not mention either Negroes or any
other groups. The set would then consist of three categories:
Mentioned Negroes (with or without mention of other groups)
Did not mention Negroes, but mentioned other groups
Did not mention either Negroes or other groups
A decision has to be made about how far the extension should
go. Reference to the actual answers suggests that this set of categories
still does not do justice to their content. The groups other than
Negroes that are mentioned vary in significant ways: some respondents
describe the groups in economic terms (poor people); others refer to
personal attributes (tough); still others mention religious or nationality
groups (Jews, Irish). Each of these attributes selected by respondents
to describe the groups of which they were aware presents a possible
classificatory principle that could lead to another set of categories,
much as we have demonstrated for Negroes. In principle it is, of course,
possible to use all these attributes for the establishment of category
sets. In practice this is often uneconomical and unrewarding, because
not all of these classificatory principles bear on the purposes of the
investigation.
If, as we have supposed, the study was concerned with awareness
/
of Negroes, it might seem that all of these other classificatory principles
are irrelevant. But such a view is based on too narrow a notion of the
task of analysis. That a certain percentage mention Negroes is not yet
a sufficient indication of awareness of Negroes. It may well be that any
other ethnic group in the neighborhood would have been mentioned to
the same extent. Whether or not there is a special awareness of
Negroes can be decided. only if the mention of other racial, national, or
religious groups has been coded too. However, classification of the
replies in terms of lmentioning economic status or personal attributes
of people in the neighborhood would probably not serve any function.
In order to provide for these additional classifications of refer-
ences to other ethnic groups, additional sets of categories must be set
up. Each of them; however, must obey the rules given on page 392.
The following list,' constituting the "code" for the question, might be
the final result:
Racial groups
1. Negroes: mentioned-_; not mentioned--
2. Other racial groups: mentioned--; not mentioned_ _
Nationality groups
3. Irish: mentioned--; not mentioned--
4. Italians: mentioned-_; not mentioned--
5. Other nationality groups: mentioned--; not men-
tioned--
Religious groups
6. Jews: mentioned--; not mentioned--
7. Catholics: mentioned-_; not mentioned--
S. Protestants: mentioned_ _; not mentioned-_
9. Other religious groups: mentioned--; not men-
tioned--
10. Mention of groups listed above
Mentioned one or more--; mentioned none, but men-
tioned other human grouping(s ) _ _; mentioned none,
stating "I don't know" _ _; did not answer question--
Notice that, although we are dealing with the answers to only one
question, our specific research interests have led us to provide ten
sets of categories in which to classify the respondents. Every respondent
can be placed in one of the two categories of each of the first nine sets
and in one of the four categories of the final set.
Since the failure to check a specific "mentioned" may be taken to
imply the corresponding "not mentioned," the appearance of the
code can be simplified by omitting all of the "not mentioned" cate-
gories; but each of the first nine set~ will stiU remain a two-category set,
one category .indicated by a check mark and the other by the absence
of a check mark. Such simplification is, howeve;, not always wise. The
failure to check a particular "mentioned" category may represent an
oversight in coding. Such oversights can 1:Je ,discouraged (and certflinly
made detectable) by requiring that the placement of a respondent in
a "not mentioned" category should call for as positive an act .as his
placement in a "mentioned" category. Also, to anticipate a bit, when
coding such data for machine tabulation, it is highly desirable that all
! categories should have an explicit identification. The machines in most
common use count the number of cases in each category separately.

The sum of the counts in all of the categories in a set should equal the
THE ESTABLISHMENT OF .CATEGORIES 395
total number of cases; if the "not mentioned" as well as the "men-
tioned" category is explicitly coded, we can check whether all cases have
been accounted for.
Several qualities of this code should be pointed out. First, four
major classificatory criteria have been used in developing the ten sets
of categories: racial groups mentioned versus racial groups not men-
tioned; nationality groups mentioned versus nationality groups not
mentioned; religious groups mentioned versus religious groups not
mentioned; any of these types of group mentioned versus none of these
types of group mentioned. It should be noted that each of these four
criteria of classification yields a number of sets of categories. Thus,
"nationality groups mentioned versus nationality groups not men-
tioned" includes the following sets of categories: Irish mentioned
versus Irish not mentioned; Italians mentioned versus Italians not men-
tioned; other nationality groups mentioned versus other nationality
groups not mentioned.
Second, the categories within each set are mutually exclusive, and
they also include an the possibilities relevant to that set of categories,
so that each response can be unambiguously assigned to one or the
other category; for example, a given respondent either mentioned
Negroes or he did not mention Negroes. (He may also have mentioned
other racial groups, and Irish and Jews as well. Each of these elements
of his response is taken account of separately in this code, but each
can be clearly placed within its relevant set of categories.)
Third, the total code is exhaustive in the sense that a place is
provided for every possible response (mainly because of the residual
"catch-all" category, number 10). But, fourth, it is not exhaustive in
terms of all possible :classificatory principles that could be applied to
the answers. In addit~on to those already ruled out as irrelevant to the
research problem, it does not, for example, provide for distinctions be-
I
tween the manner, iderogatory or otherwise, in which groups are
referred to. r
Finally, the inclusion in the list of the categories "Catholics" and
"Protestants" should, be mentioned. Subsequent tabulation showed
that 3 per cent of theirespondents had named Catholics and 0 per cent
Protestants. This was actually anticipated when the categories were
established. The idea of including these two groupings came in part

from knowledge of the actual composition of the neighborhood, in
part from an expectation and an interest in demonstrating that these
two groups, although they had many members in the neighborhood,
were not considered as separate or distinct to the same extent as other
groups.
In this case the "categorization of absent data" -i.e., the pro-
vision of a category for a response that is expected not to occur-may
have led to an obvious result, the demonstration that members of
dominant or socially accepted groups are often not perceived in terms
of their group membership. The principle involved, however, is far
from obvious and is all too often neglected. It is based on the notion
that ideas for categorization should always come from two sources: an
intimate acquaintance with the evidence in hand and general knowl-
edge and anticipatory analysis of the possible types of response, based
on theoretical, logical, or practical considerations. The application of
this rule for the establishment of categories leads to the possible
discovery of the significant absence of some response, which might
otherwise have gone unnoticed.
There are types of studies and types of data for which the estab-
lishment of categories is even simpler than in this illustration. Where
check-list questions have been used, or any form of rating scale,
whether applied in an interview or in an observational situation, the
analyst of the data need give virtually no attention to the problem of
establishing categories. His' data have been precategorized in previous
stages of the inquiry. This is also true in all studies in which the
observations are recorded in prescribed categories. Such precategoriza-
tion, when undertaken as part of the development of research instru-
ments, is, of course, very similar to the procedure employed in establish-
ing categories after the data have been collected.
DEFINITION OF COMPLEX CATEGORIES
In some situations the establishment of categories is considerably

more difficult and more time-consuming than in the neighborhood-
study illustration. What made the establishment of a set of categories
THE ESTABLISHMENT OF CATEGORIES 397
relatively easy in that case was that the answers were fairly simple and
clear-cut; the categories could easily be defined in a completely un-
ambiguous way. Although this is the way categories should always be
defined, the task is much more difficult with some types of content
than with others.
For example, Merton and his colleagues (forthcoming) asked
white housewives, "How would you say the colored residents feel about
living in the same community with whites?" and asked Negro house-
wives, "How would you say the white residents feel about living in the
same community with Negroes?" The answers ranged from imputations
of highly favorable attitudes to imputations of highly unfavorable
attitudes. Here are some examples:
They want to live here so they can say they are equal.
They seem very friendly to us.
They like the idea.
I don't think they mind; we get along.
Some of them like it, some don't.
They think it lowers them.
I don't come in contact with them so I wouldn't know.
They hate it.
According to the preceding discussion, it would/ not be difficult to
evolve a simple set of categories based on the classificatory principle
of favorable versus unfavorable attitude imputed to the other race. In
that case one category would be needed for favorable attitude, one for
unfavorable, one for neutral, plus a residual category. However, distinct
shades of meaning appear among both the imputed favorable attitudes
and the imputed unfavorable attitudes. A person who says, "They
want to live here so 'they can say they are equal," conveys something
I
different from the bne
I
who says, "They like the idea." Similarly,
one who says, "They think it lowers them," may mean something
different from one ~ho says, "They hate it." What is the distinctive
criterion? In both cases, it would appear, some of the respondents
attribute the alleged feeling of the other race to a value believed
to be held by that" group. Hence, the imputation of such values
was used as another classificatory principle. A set of categories was
developed in which: one category stood for attributing the favorable
attitude of the other race to a material, social, or spiritual value or

benefit they derived from the situation (or for attributing their unfavor-
able attitude to a corresponding loss Or disadvantage), and a~'lOther
category stood for an absence of statements about values. For the sake
of simplicity in the subsequent coding operation, the two sets of
categories were combined, thus:
Favorable attitude imputed to the other race, explained in
terms of values or benefits they derive from living in the same
project with respondent's race.
Favorable attitude imputed to the other race, without ex-
plicit explanation in value terms.
Neutral or accommodative attitude imputed to the other
race.
Unfavorable attitude imputed to the other race, explained
in terms of values or benefits they derive from segregation.
Unfavorable attitude imputed to the other race, without
explicit explanation in value terms.
Other answers, no answer, don't knQw.
Before a list of such categories is put into use, it is necessary to
specify the content of each category as accurately as _possible. This is
done by adding an explanatory sentence or'two to each category and
\llustrating the meaning by examples.
In the case of the first category in the list above, for example, the
explanatory sentence read: "These answers imply that the other race
gets something out of living with the respondent's race. E.g.: 'They
want to live here so they can say they are equal.' " .
Working with such complex categories requires considerable care
and effort in classification. Even when the categories have been worked
out carefully, their use will present more problems than the use of
categories narrowly and exactly defined. If one respondent answe,rs, for
.' example, "They like it all right here, they know why," it is a' moot
question whether or not this statement Implies a benefit. Additional
rules have to be established to deal with such answers. In this case,
I an appropriate ruling for the first category might read as follows: "This
category applies only to those statements which specify the nature of
the value or benefit."
THE ESTABLISHMENT OF CATEGORIES 399
SELECTING CLASSIFICATORY PRINCIPLES FOR CATEGORIZING
UNSTRUCTURED MATERIAL
Special problems arise in the categorization of unstructured

material, such as observational protocols, case histories, speeches of
agitators, unstructured interviews, etc. We have alluded to these dif-
ficulties in earlier chapters when discussing participant observation and
the use of personal documents and communication content in research.
In a study using structured instruments for gathering data
relevant to dearly formulated research questions or hypotheses, the
appropriate principles for classification of responses are fairly clearly
prescribed by the nature of the stimulus or of the questions and by
the responses. In working with unstructured evidence, however, the
first problem is to arrive at decisions about which aspects of the
material are to be categorized-that is, what classificatory principles
are to be used in establishing sets of categories.
The establishment of principles of classification is especially dif-
ficult in exploratory studies, since such studies, by definition, do not
start with explicit hypotheses. At the time of data collection, the in-
vestigator does not know which aspects may turn out to be most im-
portant. Therefore he must usually collect a large amount of data;
thus, in the analysis, he has the problem of dealing not only with
unstructured material but also with a vast quantity of it, much of
which may prove irrelevant to his purpose.
The first step in analyzing the data of an exploratory study is to
develop working hypotheses that will yield classificatory principles.
The investigator usually proceeds by reading carefully through all his
material, keeping alert for clues in the data. There are several proce-
dures that may help him in his task. One is to study, if it is available,
material on a group that contrasts with the one he is investigating, in
order to get ideas ab,out the important differences between the two.
In a study of delinquncy, for example, it is appropriate to read social
agency case records nbt only of delinquents but also of other juveniles
under the care of an agency. That is, one contrasts cases that differ
noticeably in the cha~,acteristic being investigated in order to see what
other concomitant differences can be perceived in the contrasting cases.
Or the investigator may sort his cases into groups that seem to
400 '. ANALYSIS AND INTERPRETATION
belong together, and then ask himself what led him to feel that those
he has placed in a single group are alike. Thus, for example, Chein
et aI. (1952), in a study of views of prominent Jewish educators and
group workers about a number of issues in Jewish education, found it
appropriate first to sort their respondents into groups that could be
characterized in terms of their total outlook on the meaning of being
Jewish. The investigator may find that he has grouped his cases on the
basis of common characteristics; he may then examine them to see
whether those who have similar characteristics have undergone similar
experiences. Or he may find that his grouping is on the basis of similar
experiences; he may then re-read the cases to see if these similar ex-
periences seem to have led to similar consequences.
,Another approach that may stimulate the formulation of working
hypotheses is to: note matters that seem surprising in view of eHher
common-sense or theoretical expectations, and then to search for #
possible explanations of the surprising phenomenon. For example,

Lambert and Bressler (1957), in a study_ of Indian students in the
United States, noticed that these students often seemed more disturbed
by questions about Indian problems asked by relatively well-informed
Americans than they were by the stereotypes or the ignorance of those
who had no information about India; even when the questions were
asked with no unfriendly intent, the students seemed to interpret them
as hostile criticism. This observation led to the hypothesis that, for
memb'ers of colonial or formerly colonial nations, certain :'sensitive
areas" develop. These, "sensitive areas" are aspects of the nation's cul-
ture that have historically been the object of hostile .criticism by the
dominant country or countries and that have been used as 'an excuse
for keeping the nation in a subordinate status. Any reference to these
I
areas, no matter what the intention of the individual who makes it,
carries with it the historical connotations of hostility. But the outsider
(in this case, the American) must have some knowledge of the stu-
dent's home country in order to be aware of these problem areas; thus
the questions of relatively well-informed people are more, likely to
touch on one of the sensitive areas. Questions of totally l1ninformed
people, which usual1y do not refer to these sensitive areas, are set down
to ignorance; que~tions referring to sensitive areas, which can be
THE ESTABLISH:MENT OF CATEGORIES 401
asked only by persons with some information about the country, are
interpreted as evidences of hostility.
Even with clear hypotheses, however, the analysis of unstructured
material presents special problems. Since, by definition, the material
is not of the kind in which the same question has been answered by all
respondents or the same observations have been made on all subjects,
there is always the possibility that information on a given point may be
missing from some of the documents. On the other hand, there is
likely to be a great deal of material that is not directly relevant to the
hypotheses. Moreover, there is a problem of deciding on the size of
the units of material to which the categories are to be applied. When
data have been gathered through a structured instrument such as a
questionnaire, ordinarily each question provides a natural unit for
categorization, although some sets of categories may apply to larger
units, such as the questionnaire as a whole. But when one is dealing
with unstructured material, there are no such convenient "natural"
units. For example, if the investigator is using case records kept by
social service agencies, he may categorize every act or statement made
by the client, or every session of the client with a case worker, or the
entire case record. He must decide which of these units is most appro-
priate for providing answers to his specific research questions.
/
Coding: The Categorization of Data

Coding is the technical procedure by which data are categorized.
Through coding, the raw data are transformed into symbols-usually
numerals-that may be tabulated and counted. The transformation is
not, however, aut~matic; it involves judgment on the part of a coder.
The jUdgmen't that assigns a response to a category is often made
by someone other than the person who goes by the official title of
I
"coder." Frequently, it is the respondent himself who assigns his
response to a cate'gory. This is true for many poll-type and multiple-
choice questions, for example, when the individual is limited to a
response of "Yes/' "No," "Don't know," or to "Agree," "Disagree,"
"Uncertain," or to indicating by a check mark his position on a rating
scale. Or the person who collects the data may categorize as he collects
them. This is, of course, what is being done when an interviewer or an

observer employs a rating scale to describe a person's behavior.
There are a number of advantages to having the interviewer or the
observer code the data. For one thing, he is in a position to notice the
situation as well as the individual's behavior. Thus, he has more in-
formation upon which to base a judgment than the coder working
from the written record. Another advantage is that categorization by
the data collector saves both time and labor.
Notwithstanding these advantages, categorization of complex data
is usually done by coders after the data have been collected. This pro-
cedure allows time for reflection; on-the-spot judgments of an inter-
viewer or observer may not be as discerning as judgments made with
more time for deliberation. The judgments of data collectors may be
colored by irrelevancies such as the appearance and mannerisms of
the respondent, his accent, responses to previous questions, etc. More-
over, if, each interviewer or ob~erver categorizes only the data he col-
lects, ,J.mreiiability is likely to 'be increased. There is a tendency to
develop_!! .frame of reference with respect ttl the material that one is
coding. E~en if the data collectors were all perfectly consistent with
one another in their use of categories initially-an un~ikely assumption
-they would tend to develop varying frames of reference appropriate
to their lir}iited materials, which would make their categorizations
unreliable after a time. A common frame of reference is easier to obtain
and check in an office coding operation than in the field.
PROBLEMS OF RELIABILITY IN CODING2
There are many things that may operate to make the judgments of
coders 3 unreliable. These factors lpay arise from the data to be cate-
gorized, from the nature of the categories that are to be applied, .from
the coders themselves, etc. Let us briefly consider some of these factors
and possible safeguards against them.
DIFFICULTIES ARISING FROM THE DATA. Many of the difficulties that
occur in coding result from the inadequacies of the data. Frequently,
2For a general discussion of problems of reliability, see Chapter ?
3Weare speaking here of persons who code the data after they have been
collected, not of respondents or data collectors.
CODING: THE CATEGORIZATION" OF DATA 403
the data do not supply e.nough relevant information for reliable coding.
This may be the consequence of inadequate data-collection procedures
-poorly worded questions, untrained observers, etc. Perhaps more
often, however, the difficulties are of a sort that can easily be corrected
by carefu1 editing of the data.
When the interviewer or the observer hands in his material, the
possibility of eliminating many potential coding difficulties still exists.
A careful examination of the data as soon as they are collected and, if
necessary, a systematic questioning of the interviewer or observer will
avert many coding problems. The process of scrutinizing the data to
improve their quality for coding is commonly caned editing. 4
Not only does editing help to avoid later coding problems; it may
also markedly improve the quality of data collection by calling atten-
tion to points at which the interviewers or observers have misunder-
stood instructions, are not recording data in sufficient detail, etc. To
serve this function, editing shou1d be done in the course of pretesting
the interview or observation schedule and of training the interviewers
or observers, as well as throughout the period of data collection. In
any case, if editing is to remove coding problems, it must be done
while the interviewers or observers are still available for questioning.
Each interview or observation schedule should be checked for:
1. Completeness. An items shou1d be fined. in. A b1ank next to a
question in an interview schedule may mean "don't know," "refused
to answer," or that the question was not applicable, that the question
was omitted by mistake, etc. For many purposes, it is important to be
able to distinguish among these potential meanings.
2. Legibility. If the coder cannot decipher the handwriting of
the interviewer or observer, or the abbreviations and symbo1s he
employs, then coding is impossible. It is a simple matter to check for
legibility when th~ material is handed in and to have it rewritten if
necessary, but it iSioften extreme1y time-consuming to have the coder
attempt to deciph~r the handwriting or to track down the interviewer
once coding has begun.
3. Comprehensibility. Frequently a recorded behavior or response
seems perfectly cQ~prehensible to the interviewer or observer but is
4 For a detailed discussion of the process of editing in largescale surveys,
consult Parten (1950; Chapter 13) ..
not comprehensible to another. The context in which the behavior or

response occurs is known to the interviewer but not to the coder, who
thus cannot visualize exactly what the subject did or understand what
his answer meant. Systematic questioning of the interviewer or observer
to dispel confusions and ambiguities will considerably improve the
/ quality of the coding.
4. Consistency. Marked inconsistencies within a given interview
or observation schedule not only make for problems in coding; they
may indicate errors in collecting or recording the data. For instance, if
in an interview on Negro-white relationships the reply to one question
is that the respondent does not know any Negro families but a later
comment reports that the respondent visits back and forth with a
Negro family, there is an obvious need to inquire immediately into
this inconsistency. Often the interviewer, on questioning, clears up the
inconsistency. If he is not able to do so, it may be desirable to get in
touch with the' respondent again if the point is important to the
analysis ,of the ciata. '
5. Uniformity. By and large, adequate instructions to the inter-
viewers or observers will result in uniform procedures for collecting
and recording the data; however, it is necessary to check the uniformity
with which these instructions have been followed. For example, if
income is recorded in units different from those specified in the instruc-
tions-say in terms of monthly rather than weekly income-and the
units of recording are not clearly indicated, coding may be disrupted.
6. Inappropriate responses. Occasionally a response simply is not
germane to the purposes of the investigation; this is particularly likely
to occur if a question is not clearly worded or not intelligently asked.
It is helpful to the coder to have such responses sorted ou't from. the
appropriate responses, since the categories developed to cO,de the
answers will probably not be applicabk. , '
DIFFICULTIES ARISING FROM THE CATEGORIES. The value of the
categorization of data depends entirely on the soundness of the cate-
gories employed. Categories must be well defined from a conceptual
point of view and must be relevant to the purposes of the research.
As pointed out on page 392, each set of categories must also meet
certain formal requisites: it must be based on a single ,classificatory
principle, the categories must be mutually exclusive, and they must be
exhaustive. Even if the categories satisfy all these conditions, however,
CODING: THE CATEGORIZATION OF DATA 405
coding will be unreliable if the categories are not clearly defined in
terms of indicators that are applicable to the immediate data. Usually
the categories are defined by means of examples from the data. The
examples should include not only responses that typify the category
but also, if possible, responses that help to distinguish the boundary
lines between similar categories.
TRAINING OF CODERS. It is obvious that the reliability of coding is
affected by the competence of the coders. The training of coders
usually proceeds by the following steps: (I) The various codes are
explained
, and illustrated with examples from the material to be
categorized. (2) The coders then all practice on a sample of the data.
Problems that arise are discussed by the coders as a group with the
supervisor in order to develop common procedures and definitions.
(3) Frequently, as a result of the practice coding, the categories are
revised to make them better applicable to the material and to put in
writing the procedures and definitions that have evolved during the
preliminary coding. (4) At the point in the practice period when
relatively few new problems arise, the coders work on an identical
portion of the data without consulting one another or the supervisor.
The consistency or reliability of the coding is then computed to deter-
mine whether it is feasible to begin coding in earnest.
In computing the consistency of coding, one may use the coding
of the supervisor as a criterion against which to test the various coders,
or use some measure that reflects the consistency of the group as a
whole, or compare each coder with every other coder. Depending on
the results of the reliability checks, one may decide to eliminate
categories that seem too unreliable to be of value, or to spend more
time in training, or to eliminate the coders who are most inconsistent,.
or to use special prpcedures with more difficult items (such as having
the more expert an,alysts work with the more difficult codes or having
two or more peop-Ie, in consultation, work with the more difficult
codes). (5) After \ the coding of material has begun, periodic con-
sistency checks arelnecessary to ensure that the coders do not become
careless as they become more experienced and that they do not develop
idiosyncratic methods
f.
of handling new problems in the material. The
supervisor must insist that all new problems be discussed with him.
To ensure uniformity, any decisions that are made after coding has
begun in earnest have to be instantly communicated to all coders.
COMPUTING THE RELIABILITY OF CODING
Obviously, the consistency and appropriateness with which a

given type of answer is assigned to a given category will have an
important bearing on the outcome of the analysis. Therefore it is
important to check the reliability of coding, and to increase the agree-
ment among coders as much as possible. It is difficult to set any given
level of reliability as a standard that should be achieved. Different
types of material present different degrees of difficulty in achieving
reliability; in general, the more highly structured the material to be
coded and the simpler the categories used, the higher the reliability
will be. Moreover, different research purposes may demand different
standards of agreement. In any case, it is desirable not only to check
the reliability of coding but to ma1<e this information available to the
reader when the study is reported. Increasingly, this is becoming
common practice. ..,. #
There are various methods of computing the reliability of coding.

Since the more carefully developed methods require a more detailed
statistical 'presentation than is appropriate here, we shall refer the
reader to representative articles. 5 It should be noted, however, that
perhaps the most frequent error in reliability checks is the computation
of reliability coefficients that are inappropriate to the use to which
the data are to be put. Thus, it is a coinmon but mistaken procedure
to report only the reliability for the gross categories and to omit
reliability checks on subcategories. For example, if in analysis one is
going to employ subcategories of "initiating behavior"-say, "initiating
behaviqr which is followed by other members of a group" and "initiat-
ing behavior which is not followed by other members of the group"-
it would be misleading to perform or to report a reliability check on
the o~er-all category of "initiating behavior" only.
Tabulation '
Tabulation ~s a part of the technical process in the statistical

,analysis of data. The essential operation in tabulation is counting to
5 See Guetzkow (1950); Lasswell, Leites and associates (1949); Robinson
(1957); Schutz (1952); Scott (1955).
TABULATION 407
determine the number of cases that fall into the various categories.
The term marginals is commonly used to refer to simple counts of the
frequencies with which the various categories in each set occur in the
data; for example, the number of people who have not gone beyond
grammar school, the number who have attended high school but not
graduated, etc. The temlS cross-tabulation or breakdown are often
employed to refer to the tabulation of the number of cases that occur
jointly in two or more categories-for example, tabulation of the
number of cases that are both high in education and low in income.
Crosstabulation is an essential step in the discovery or testing of
relationships among the variables in one's data. Later in this chapter
we shall discuss this use of cross-tabulation.
Tabulation may be done entirely by hand, or it may be done by
machine. Each method has advantages and disadvantages. Briefly,
manual tabulation is generally less expensive and less time-consuming
when there is only a small or moderate number of cases, when the
numb,er of category sets to be counted is small, and when not many
cross-tabulations are to be done. As the number of cases or the number
of cross-tabulations increases, the use of machine tabulators becomes
progressively more economical. The number of cross-tabulations is,
perhaps, the most important factor in determining the relative effi-
ciency of one rather than the other procedure In tabulation.
Generally, the efficiency of hand tabulation tends to be under-
estimated. With proper techniques, hand tabulation can be quite
rapid and accurate. One of the most efficient techniques uses, for each
case, a small (3 x 5) code card that can be easily sorted and counted.
Scores from as many category sets can be placed on a card this size as
on the SO-column punch card that is commonly used in machine
l
tabulation. By use of such devices as colors, heavy lines, etc., the codes
can easily be dis~inguished and the cards efficiently sorted. Parten
(1950) 6 reports trat a relatively unskilled clerical worker can sort
1,000 of these cards into six stacks in less than five minutes. Counting
I
is also quite rapid if small cards are used; 1,000 cards can be counted in
less than five minutes, if good technique is employed. At such a rate
of speed, sorting !and counting by hand is likely to rival, for short
6 See her Chap~er 15 for a detailed discussion of tabulation procedures.,
4G5 ANALYSIS AND INTERPRETATION
periods of time, the efficiency of machine tabulation for even several

thousand cases.
Both machine and hand tabulation presuppose that the data have
been coded and that the coding has been checked. Scores for each
individual are usually transcribed onto a card for hand tabulation or
onto a sheet from which a card puncher punches them onto cards fof
machine tabulation. Machine tabulation actually involves more cleri-
cal and other specialized operations than hand tabulation does. Card
punching, the chetking of machine tabulations, the transposition ot
the results from machine tabulation forms to tables, are all steps that
are not required in hand tabulation. With a large number of cases or
with many cross-tabulations, however, the speed of the machines more
than compensates for the time involved in these operations.
There are a number of types of machine available; developments
in this field have been extremely rapid within recent years. Some
machines simply sort and count cards; others sort, count, and print the
results; still others are capable of performing the most complicated
statistical operations. These latter machines_are extremely complex;
they must be "programmed~'jfoJ a given operation by a specialist in
this work. If a program-iS";n9t already available, programming may be
quite expensive. "
The cost of using a machine is very high if one considers hourI,
'-'rates. However, if there is a large amount of data requiring complicated
statistical treatment, the speed of the machine may more than com-
pensate for the expense; in some cases it makes possible ~nalyses that
would not otherwise be feasible. For example, with a few hundred
subjects, the computation of all the intercorrelations among a hundred
variables 7 is a job that might take a statistical clerk, using an ordinary
calculating machine, many months;, an electronic computer, once it
has been properly set up for the operation, can complete the job in an
hour or SO.8
The rapidity of even the simpler machines presents a temptation
For one hundred variables, there are 4,950 possible intercorrelations!

7
, For a more detai)ed discussion of the various machines, see Baehne (1935),
8
Eckert (1940), Leahy (1931), Parten (1950), Paton (1935), Pease (1949), and
Wrigley (1957); or consult representatives of the commercial agencies that rent
the machines.
TABULATION 409
to "run wild" and simply cross-tabulate every variable against every
other in the hope of finding some relationships rather than to plan the
analysis by considering the probable value of each operation. This
procedure is most undesirable. To say nothing of the expense, it has a
number of fallacies which we shall discuss later in this chapter (page
421).
Statistical Analysis of Data
Let us suppose that we have asked a thousand people, who have

been selected as a sample of the adult population of England, a series
of questions to gather information about their movie-going habits.
Assume that we have developed a number of sets of categories and have
coded the responses of each individual. The coding may be considered
: method of summarizing the responses of each individual; if we were
concerned only about each individual, this might be as far as we would
go in the analysis. However, the purposes of our research are broader
than this. We wish to know more than that a given government clerk
in London goes to the movies four times a month and that an innkeeper
in Yarmouth goes only once a month. Our inquiry is directed toward
providing infor,m,at,io!l about the adult population of England. 9 ,
9 Even if we were interested p;imarily in individual cases, we would 'still 'find a
characterization of the total population useful for study of the individual. A fact
about an individual frequently takes on its significance in relation to facts about
the population of which the individual is a part. Thus, if one knows that it is
customary behavior for members of a group to go to the movies twice a month,
we have a frame of reference against which to interpret the individual who secs
five movies a month. Our picture of this same individual might be different if the
average member of the group went to the movies ten times a month. Such
normative interpretatio~-the evaluation of the individual's behavior in the light
of the group standard--;is the most common type of interpretation. This is not to
deny the value of ipsative interpretation, in which the frame of reference is the
individual himself; Thus, the fact that an individual reports going to the movies
five times a month may take on varying significance depending upon whether, in
the past, it was customary for him to go ten times or not to go at all. Additional
meaning may be given by other intra-individual comparisons-for example, by
comparing the amount of time the individual spends at the movies with the
amount of time he spends reading books. For a statistical approach to the
ipsative study of the individual, see Baldwin (1950) and Cattell and Luborsky
(1950).
As a necessary step in characterizing this population, we must

describe or summarize the data we have obtained on the sample we
have studied. Tabulatio,n is a part of this step. In addition, we must
estimate the reliability of generalizations from the obtained data to the
total population. Statistical methods are used to fulfill both these func-
tions. The term descriptive statistics is often applied to characterize
the methods employed in summarizing the obtained data, and sam-
pling statistics to characterize the methods utilized in making and eval-
uating generalizations from the data. Since the procedures are described
fully in most statistics textbooks, we shall not discuss them in detail
here. However, let us briefly indicate what is involved.
DESCRIPTION OF THE DATA
In giving an adequate description of a mass of data, we usually

wish to do one or another, or several, of the following things:
v l. To characterize what is "typical" in the group. We wish to
know, for example, how many movies, on th~ average, the people in our
sample attend, or, perhaps, we want to know what kind of movies most
of them prefer. In the terminology of statistics, we wish to get some
indication of the central tendency. There are various measures of
central tendency, each of which makes assumptions about the nature of
the data. If those assumptions are not met, then the measure of
central tendency may be misleading. Thus, an arithmetic mean or
average (the sum of individual scores divided by the number of in-
dividuals) implies equality of intervals, or an interval scale.lO It is
appropriate to such data as the number of times people go to ,the
movies each month, since here the scale is that of number itself, with
equal intervals between every two whole numbers.n On the other
hand, suppose that we had asked the respondents to rate each of
10 See Chapter 5 for a discussion of various types of scale.
11 Number of times a person goes to the movies is, of course, on a ratio scale,
and a fortiori on an interval scale. But, strictly speaking, even this scale violates
one of the assumptions involved in the use of the arithmetic average. This'scale is
discrete; i.e., only integers are possible. What, for instance,.J3oes it mean to say
that a person has gone to the movies 1.763 times? 'fife arithmetic average
presuppuses a continuous scale; Le., one in which all fractional values are possible.
Sometimts, however, the attribution of continuity to a discrete scale is a con
venient fictiun.
STATISTICAL ANALYSIS OF DATA 411
several types of movie (comedies, documentaries, etc.) on a scale that
attempted to gauge preference in terms of frequency of seeing such
films; for example, "I go to see almost all the films of this type that are
shown, . . . I occasionally go to see films of this type, . . . I seldom
go to see films of this type, . . . I avoid films of this type." If we
were to assign the numbers 1 to 4 to these four scale positions and
average the rating for each type of film, the average would have no
clear meaning, since there is no reason to believe that the scale positions
are equal}y distant from one another. In such a case, it would be
preferable to employ the median (the point on the scale above which-
and correspondingly, below which-50 per cent of the cases lie). Or
suppose that we had simply asked each respondent to check which of
various types of film he preferred. Here the data would be in the form
of a nominal scale, since various types of film have no relation of order
to one another. In this case, the only appropriate method for measuring
the central tendency would be the mode (the score that occurs with the
greatest frequency-in this case, the type of film mentioned by the
greatest number of respondents) .
The various measures of central tendency not only make different
assumptions about the nature of the data but provide somewhat dif-
ferent kinds of information. The arithmetic mean may be thought of
as the point on the scale around which the case,_s balance: one case,at
one extreme of the distribution, for example, may be counterbalanced
by one or more cases at the other extreme. If the cases are not sym-
metrically distributed around this point, the arithmetic mean may be
very misleading in certain respects. Thus, it takes a very large number
of cases of low income to counterbalance one case of a person whose
annual income comes to over a million dollars; as a consequence, the
arithmetic mean OF income is far above the income of the vast majority
of the population. It does not follow that knowledge of the mean
income is point1es~ information; whether it has point depends on what
you want to do with it. The median also implies a concept of balance,
but, since it takes account only of the ordinal positions of scores rather
than of their absolute values, one millionaire is counterbalanced by one
pauper. I
The use.of th~ mode dispenses with the idea of a point of balance.
In fact, it is possible to have several modes in one distribution; for
example, frequency of movie-going may. be bimodal, with a larg~

number of people going once a week, another large number going lesz
than once a month. As a rule, multimodal distributions such as this
come about as a result of the intermingling of distinct populations. In
our illustration, for example, going to the movies once a week may be
the mode for people under 25; going less than once a month, the mode
for an older group. However, it is not always easy to identify the basis
of distinction.
It is also possible for a distribution of scores to have no mode at
all, every score occurring about as frequently as every other. In this
case, as in the case of bimodal and multimodal distributions, it is
perhaps misleading to speak of central tendency, a term suggesting that
the cases tend to cluster around SOme point more or less in the middle
of the distribution. But even in these instances, the arithmetic mean
and the median may be useful in bringing out aspects of the distribu-
tions which they describe.
2. To indicate how widely individuals in the group vary. We
might wish to know, for example, whether the people in our sample
are similar in their film preferences, so that most people prefer films of
a given type, or whether there is great diversity. Or we might wish to
know whether there is much variation in the frequency of movie-
going among the sample being studied. There are many measures of
interindividual variation. The purpose of each of them is to indicate
how similar or how different the individuals in the group are with
respect to a given characteristic. Some of the common measures are the
range, the average deviation, the standard deviation, and the quartile
deviation. '
As with measures of central tendency, each of the measures of
variability makes assumptions about the nature of the measurements
and gives somewhat different kinds of information. The range shows
the extremes of variation in the group; it might show, for example,
that at least one person never goes to the movies, while at least one
other goes every day. Obviously, the range is affected by extreme cases,
and therefore may be misleading as a picture of the group as a whole.
!To avoid this difficulty, other measures of variatio~cus on the limits
within which half of the group fall, or on the average distance of
individuals from the group mean. The quartile deviation shows the
points within which the central half of the cases faIl; like the median,
it assumes that the data correspond to an ordinal scale. The average
deviation and the more frequently used standard deviation are measures
of the average distance of individuals from the group mean; like the
mean, they assume that the data correspond to an interval scale.
_./ 3. To show other aspects ot how the individuals are distributed
with respect to the variable being measured. For example, is the number
of people who do not go to the movies at all about the same as the
number who go three times a month? Or do relatively many people go
three times a month, while relatively few do not go at all and relatively
few go six or more times a month? If you plotted the figures on a graph,
using the frequency of movie-going on the horizontal axis and the,
number of people reporting each frequency on the vertical axis, wha~
would be the shape of the resulting graph? Is it rectangular (that is,
are there equal numbers of people at each point on the scale, resulting
in a graph that takes the form of a straight line across the page); or
is it a bell-shaped or "normal" distribution; or is it an asymmetrical
curve, with a piling up of cases at one side or the other; or does it have
more than one mode (that is, is there a piling up of cases at two or
more points along the scale, with relatively few cases in between) ?12
Knowing the shape of the distribution curve is fundamental to
the use of efficient statistical methods, since the more
/
efficient methods
make specific assumptions about the nature of the distribution curve.
It is common to assume that the distribution curve for any variable is
normal, but this may not be so. If it is not, one may see whether other
known distribution curves fit the data or whether one can transform
the raw data by mathematical manipulation to a known distribution.
4. To show the relation at the different variables in the data to
one another. We; wish to know, for exampl~, whether, within our
sample, the frequency of movie-going or the preference for different
I
types of movies seems related to income, to sex, to age, etc. That is, we
wish to know whether a variation in one characteristic is associated
with or parallele4 by variations in another characteristic. There are
several method~ of determining the relationship between variables.
12 For a fuller c!jscussion of these aspects of distribution, look up skewness and
kurtosis in almost aJ;lY statistics textbook. An unusually thorough and not overly
technical discussion 'of various types" of distribution may be found in Smith and
D....l~n (1945). '
414 ANAL YSIS AND INTER.PR.ETATION
None of these methods in itself, however, permits the conclusion that

an association or correlation between variables in one's data is indicative
of a causal relationship. The imputation of causality requires an ap-
proximation to the logical model discussed in Chapter 4, pages 80-94.
Later in this chapter we shall discuss the use of analytic procedures in
drawing inferences about causal relationships in nonexperimental
studies.
5. To describe the differences between two or more groups of
individuals. For example, we may wish to compare the movie-going
habits of those members of our sample who live in communities of
less than 10,000 (whom we shall, for convenience, call rural residents)
and of those members who live in communities of 10,000 or more
(whom we shall call urban residents). This is, of course, a special case
of showing the relationship between two variables. We distinguish it
solely because in much social research interest is focused on She
comparison of groups. Although such comparisons most commonly
involve measures of central tendency, they need not be limited to this;
they may include comparison of measur~s of variation within the
groups or of the relations among variables in the two groups. It might
be, for example, that the average frequency of movie-going is similar
in the two groups but that there is greater variation among urban than
among rural residents, or vice versa.
GENERALIZATION TO THE POPULATIONS FROM

WHICH THE SAMPLES WERE DRAWN
Suppose that we have studied samples of rural and urban English-

men and that our results show differences between the two samples.
One may ask whether the differences that have been obtained reflect
true differences between rural and urban Englishmen, or whether the
( two samples might have differed to this extent by chance even though
the total rural and urban populations are alike in their movie-going
habits. Through statistical procedures, one is able to answer such a
question in terms of a statement of probability.
When we are contrasting samples or study~the differences be-
tween experimental and control groups, we usually wish to test some
hypothesis about the nature of the true difference between the larger
populations represented by the samples. Most commonly, in the social
sciences, we are still concerned with relatively crude hypotheses (for
example, that urban residents go to the movies more often than rural
residents); we are usual1y not in a position to consider more specific
hypotheses (for example, that they go to the movies twice as often).
Suppose our data show that our sample of urban Englishmen attend
an average of three motion pictures a month and that our sample of
rural Englishmen attend an average of only two motion pictures.
Clearly, the findings within our samples are in line with the hypothesis:
urban residents attend the movies more often than rural residents. But
we know that the findings based on our samples are not very likely to be
exactly the same as the findings we would obtain if we had interviewed
all the adults in England. (See Appendix B for a discussion of this
point and of ways of estimating how much the findings for each of the
samples are likely to differ from the true state of affairs in the popula-
tion it represents.) Now we want to estimate whether, if we had
interviewed the total population, we would still have found more
frequen't movie attendance on the part of urban residents. This we do,
ordinarily, by testing the null hypothesis-in this case, the hypothesis
that in the English population as a whole, rural and urban residents do
not differ in frequency of movie-going. Various statistical techniques
( called statistical tests at significance) have been _gevised, which tell us
the likelihood that our two samples might have differed as much as
they do by chance even if there were no difference between urban and
rural Englishmen as a whole. 13
It may seem odd that, when interested in one hypothesis (that
there is a difference between the two populations represented by our
samples), we should test its opposite (that there is no difference). But
the reason is not too difficult to follow. Since we do not know the true
state of affairs in th~ population, all we can do is make inferences about
it on the basis of our sample findings. If we are comparing two groups,
13 Which method, of testing significance is appropriate depends on the nature
of the measurements used and the distribution of the characteristic. Most of the
tests commonly described in statistics textbooks assume that the measurements
are in the form of at least interval scales and that the distribution of the char-
acteristic is normal. These conditions, however, are seldom met in social research.
In recent years, a num"ber of statistical tests have been developed that do not rest
on these assumptions;. they are called nonparametric or distribution-free statistics.
For a presentation of tests of this type, see Siegel (1956) .
there are obviously two possibilities: either the two populations are
alike, or they are different. Suppose that our samples from the two
populations are different on a particular measure or attribute. Clearly,
this would be likely to happen if the two populations from which the
samples are drawn do in fact differ on that attribute. However, it does
not in itself constitute evidence that they do differ, since there is always
the possibility that the samples do not correspond exactly to the
populations they are intended to represent. We must consider the pos-
sibility that the element of chance which is involved in the selection
of a sample may have given us samples which differ from each other
even if the two populations do not differ. Thus the crucial question
is: Is it likely that we would have come up with samples that differ to
this extent if the two populations were actually alike? This is the
question the test of the nulI hypothesis answers; it tens us what the
chances are that two samples differing to this extent would have been
drawn from two populations that are in fact aIike.14 Only if the
statistical test indicates that it is improbable that two samples differ-
ing to this extent could have been drawn froIl} similar populations can
we conclude that the two populations probably differ from each other.
Suppose, however, that our findings show no difference between
the two samples; Jet us say that in our samples, both rural and urban
Englishmen attend the movies, on an average, two and one-half times
a month. Can we then concluder that the total populations of rural
and urban Englishmen are alike in frequency of movie-going? Not
with any certainty. Just as there. is the possibility that samples may
diller when the populations are alike, so there is the possibility that
samples may be alike when in fact the populations differ. Thus, if our
two samples are alike, all we can conclude is that we have no evidence
that the populations differ-in other words, that the idea that the
two populations are alike is tenabie. I
But to go 'back to the case where the two samples differ. We .tan
affirm that the two populatio!1s they represent probably differ if we
14 It should be kept in mind that all statistical tests of significance, and thus
all gener;>lizations from samples to populations, rest on the assumption that the
I>Jimples are not biased.!.._that is, that the cases to be included in the samples have
been selected by some procedure that gives every case in the population an equal,
or at least a specifiable, chance of being included in the sample. If this assumption
i.~ not justified, si~nificance tests become meaningless.
-
can reject the null hypothesis-that is, if we can show that the obtained
difference between the two samples would be unlikely to appear if the
two populations were in fact the same. It is, however, in the nature of
probability that even highly improbable events can sometimes happen.
Thus, we can never be absolutely certain of our generalizations to the
total population. Whenever we reject the null hypothesis, there is some
chance that we are wrong in doing so.
However, since we are always dealing with inferences and prob-
abilities, there is always also some chance that if we accept the null
hypothesis, we are wrong in doing so. That is, even if our statistical test
indicates that the sample differences might easily have arisen by chance
even if the two populations are alike, it may nevertheless be true that
the populations differ.
In other words, we are always confronted with the risk of making
one of two types of error. We can reject the null hypothesis when, in
fact, it is true; that is, we may conclude that there is a difference be-
tween the two populations when, in fact, they are alike. This is com-
monly referred to as the Type I error. Or, on the other hand, we can
accept the null hypothesis as tenable when, in fact, it is false; that is,
we may conclude that the two populations are alike when, in fact,
they are different. This is referred to as the Type Irerror.
The risk of making the Type I error is determin,ed by the level of
significance we accept in our statistical testing. Thus, if we decide that
we will conclude that the populations truly differ whenever a test of
significance shows that the obtained difference between two samples
would be expected to occur by chance not more than 5 times in 100
if the two populations were in fact alike, we are accepting 5 chances
in 100 that we will be wrong in rejecting the null hypothesis. \Ve can
reduce the risk of a Type I error by making our criterion for rejecting
the null hypothesis m9re extreme; for example, by rejecting the null
hypothesis only if the: statistical test indicates that the sample dif-
ference might have ap~eared by chance only once in a hundred times,
or once in a thousand times, or once in ten thousand times. Un-
fortunately, however, the chances of making Type I and Type II errors
are inversely related. The more we protect ourselves against the risk
of making a Type I error (that is, the less likely we are to conclude
that two populations differ when in fact they do not), the more likely
we are to make a Type n error (that is, to fail to recognize population

differences which actually exist) . Once we have determined the degree
of Type I risk we are willing to run, the only way of reducing the pos-
sibility of Type II error is to take larger samples and/or to use statistical
tests that make the maximum use of available relevant informationY'
The inverse relationship of the risks of the two types of error makes
it necessary to strike a reasonable balance. In the social sciences, it is
more or less conventional to reject the null hypothesis when the
statistical analysis indicates that the observed difference would not
occur more than 5 t.imes out of 100 by chance alone. If the statistical
analysis indicates that the difference between the two samples might
have appeared by chance more than 5 times out of 100, the null hypoth-
esis is not rejected. But these conventions are useful only when there
is no other reasonable guide. The decision as to just how the balance
between the two kinds of error should be struck must be made by the
investigator. In some instances, it is obviously more important to be
sure of rejecting a hypothesis when it is false than to fail to accept it
when it is true. In other cases, the reverse may. be true. This may be
seen clearly in an example from everyday life, outside the sphere of
statistical analysis. In many countries it is considered more important
to reject a hypothesis of guilt when it is false than to fail to accept this
hypothesis when it is true; a person is considered not guilty so long
as there is reasonable doubt as to his guilt. In other countries, the ac-
ceptance of a false hypothesis of guilt is deemed less costly than the
rejection of this hypothesis if it w~re true; a person charged with a
crime is considered guilty until he has Idemonstrated his lack of guilt.
In much research, of course, there is no clear basis for deciding
whether a Type I or a Type II error would be more costly, and so the
investigator makes use of the conventional level for determining statisti
cal significance. However, there are some studies in which one type of
error would clearly be more costly than the other. Suppose that in a
certain school it has been suggested that a new method of teaching
arithmetic would be more effective; suppose also that this method
would require expensive equipment for each arithmetic class. An
. 15 For a discussion of the extent to which different types of statistical test offer
protection against Type II errors, consult a statistics textbook that discusses this
matter; for example, Dixon and Massey (1957, Chapter 14); or Walker and Lev
(1953, pages 60-67).
experiment is set up to test whether the new method would in fact
lead to better learning of arithmetic. Two groups of children are
randomly selected; they may also be matched in arithmetic ability and
achievement and in other relevant respects. One is taught by the new
method, one by the old. Since the new method requires expensive
equipment, the school system would not wish to adopt it unless there
were considerable assurance of its superiority; in other words, it would
be costly to make a Type I error and conclude that the new method
is better when in fact it is not. On the other hand, if there were no
difference in the expense of the two methods, a Type I error would
not be especially costly, whereas a Type II error might lead to failure
to adopt the new method when in fact it is superior.
Let us go back to the fact that any generalization from samples
to populations is simply a statement of statistical probability, of the
chances that a given difference between samples reflects a true differ-
ence between the populations. Let us say that we have decided to work
with a 5 per cent level of confidence. This means that we will reject the
hypothesis of no difference between the populations only if a sample
difference as large as the one we have found can be expected to occur
by chance 5 times or less in 100; if such a difference can be expected
more than 5 times in 100, we will accept the null hypothesis. Is there
any way of estimating whether or not our finding represents one of the
5 times that such a difference might have appeared by chance? On the
basis of an isolated finding, there is not; but we can draw further infer-
ences by examining patterns within our findings.
Suppose that we are interested in testing the effects of a series of
lectures about the United Nations on attitudes toward that body. We
have set up a careful study design, with randomly selected experimental
and control groups, perqaps also matched in terms of initial attitudes,
with precautions to assure the frankness of responses, etc. Now suppose
that we use as our measure , of attitudes toward the United Nations
only one item-say, attitudes toward the establishment of a U.N. police
force. We find that those who have attended the lectures are more
favorable on this question than those who have not, and a statistical
test indicates that the cl~fference would not have appeared by chance
due to random sampling fluctuations more than 5 times in 100. How-
ever, this has the coroliary that it might have appeared by chance 5
times in 100, and we have no way of knowing whether this is one of

those 5 times. Let us say, however, that we have asked 20 different
questions that are reasonable indicators of attitude toward the United
Nations. If we are using a confidence level such that we accept as
significant a difference that might have occurred by chance 5 times in
100, then, if we had 100 questions, we might expect to find, by chance,
statistically significant differences on 5 of them; out of 20 questions,
we might expect to find, by chance, a difference on 1 of them. But
suppose we find that on 12 of our 20 questions those who have
attended the lectures are more favorable than those who have not.
We may feel much safer in concluding that there is a true difference in
attitudes, even though on each question the statistical test indicates
that the difference might have arisen by chance 5 times in 100.
What if, out of the 20 questions, only the one about a U.N. police
force shows a statistically significant difference between the l;wo
groups? This difference might well have occurred by chance; on the
other hand, it may be that the lectures actually did influence opinions
onqhis point though on no other. Unless our hypotheses specifically
predicted that the lectures would be more likely to affect beliefs about
an international police force than any of the other 19 items, we are
not justified in making this latter interpretation, no matter how con-
vincingly we may argue (after we have the finding) that the content of
the lectures was such that they were especially likely to affect beliefs
about an international police force. For if it 'bad happened that the
one item that showed a change dealt with the veto_ power in the
Security Council, we might (in the absence of specific predictions
about which items would be most affected) discover in the lectures
material especially likely to change beliefs about the veto power. In
other words, given any two variabl~s that show a statistically significant
relationship~ an investigator usually finds, it possible to propo,se an
explanation for the relationship. However, the critical test of an
obtained relation is not the ex post facto rationales and explanations
for it, but rather the ability to predict it or to predict other relationships
on the basis of it. 16 Thus, our unpredicted finding of a difference in
16 As pointed out in Chapter 5, "prediction" in this context is not limited in
meaning to the forecasting of future occurrences; it includes also the. announcing
of past or current events prior to knowledge of their occurrence.
attitudes toward a U.N. police force, even though "statistically signi-
ficant," cannot be considered as established by the study we have
carried out. It may, however, provide a fruitful hypothesis for a future
study.
This point is clear enough as an abstract principle, but its implica-
tions are frequently ignored in analyzing and reporting data. Sometimes
an investigator "runs wild" with an IBM machine and relates every
variable to every other variable; it is then to be expected that a specified
proportion of. the relationships will appear statistically significant. Or
he may first inspect the data for relationships and analyze only those
that appear to be statistically significant. In either case, if he reports
only those relationships that turned out to be statistically significant,
the report may be entirely misleading, since the reader has no basis for
judging whether these relationships form a consistent pattern within
the findings or whether they are such a small proportion of the total
number of relationships that it is more reasonable to assume their
chance occurrence.
Since statistical statements are always statements of probability,
we can never rely on statistical evidence alone for a judgment of
whether or not we will accept a hypothesis as true. Confidence in the
interpretation of a research result requires not only statistical confid-
ence in the reliability of the finding (i.e., that th~ differences are not
likely to have occurred by chance) but, in addition, evidence about
the validity of the presuppositions of the researchP This evidence is
necessarily indirect. It comes from the congruence of the given research
findings with other knowledge about which there is considerable as-
surance. Even in the most rigorously controlled investigation, the
establishment of confidence in the interpretations of one's results or in
the imputation of causal relationships requires repetition of research
and the relating of th~ findings to those of other studies.
I
It is important to recognize that even when statistical tests and
the findings of a number of studies suggest that there is indeed a con-
sistent difference between two groups, or a consistent relationship be-
tween two variables', this still does not constitute evidence of the
17 Some of the c0r.mon presuppositions are that the measures are relevant to
the variables included iI,1 the hypothesis and that the effect of extraneous variables
has been controlled or taken into account. These points were discussed in greater
detail in Chapter 4.
reason for the relationship. If we want to make causal inferences-

that is, to say that one variable or event has led to another-we must
meet assumptions over and above those required for establishing the
existence of a relationship. This point is discussed in detail in Chapter
4; analytic procedures that provide a basis for judging whether these
.\lssumptions have been met in nOn experimental studies are considered
in the following section of this chapter.
One further point should be mentioned. The fact that a result is
statistically significant does not necessarily mean that it is socially or
psychologically significant. Many statistically significant differences
are trivial. For example, given enough cases,18 an average difference
in intelligence between men and women of less than one IQ point may
be statistically significant, but it is difficult to see any real import in the
finding. On the other hand, there are cases where a small but reliable
difference has great practical importance. For example, in a large-scale
survey designed to give information about a population, a difference of
one half of 1 per cent may represent hundreds of thousands of
people, and knowledge of the difference may be important for policy
decisions. One must constantly be concerned with the social and
psychological meaning of one's findings as wen as their statistical
significance.
Interring Causal Relations

As pointed out in Chapter 4, if one wishes to draw the inference
that one variable (X) is the "ca~se" of another (Y);three types bf
evidence are necessary: (1) that X and Y vary together in the way
predicted by the specific hypothesis; (2) that Y did not precede X in
time; and (3) that other factors did not determine Y. It was pointed
out that carefully controlled experiments provide evidence nilevqnt to
all these points. They do so by setting up randomly selected groups,
one of which is then exposed to X while the other is not; the influence
of other factors is taken into account more or less extensively, depend-
18 The larger a sample, the less likely are findings based on it to deviatdrom the
, true state of affairs in the ~opulation. Tests of significance therefore' take into
account the number of cases m the sample. The larger the samples, the greater the
likelihood that a given difference between them is statistically significant. For a
more detailed discussion of this point, see Appendix B.
INFERRING CAUSAL RELATIONS 423
ing on the particular study design. Experimental studies present no
special problems of analysis, since the precautions against invalid
inference are built into the study design.
In nonexperimental studies designed to test causal hypotheses,
however, the investigator must find substitutes for the safeguards that
are built into experimental studies.1 9 Many of these safeguards enter
at the time of planning data collection, often in the form of providing
for the gathering of information about a number of variables that
might be alternative determinants of Y. By introducing these additional
variables into the analysis, the investigator approximates some of the
controls that are inherent in experiments. But the drawing of infer-
ences of causality always remains somewhat hazardous in nonexperi-
mental studies. In view of its many hazards, causal attribution in the
analysis of studies not following an experimental design will be ex-
amined in some detail in the following pages. 20
If a relationship or association between two variables has been
established in a study that did not follow a rigorous experimental de-
sign, and if the research interest is in causal relationships rather than
in the simple fact of association between the variables, the analysis has
taken only its first step. The investigator must consider whether Y
(or the variation in Y) might have occurred before X, in which case
it cannot be an "effect" of X. Assumptions about the time relations
between X and Y rest either on logical considerations, or on attempts
to establsh (either by measurement before exposure to X or on an
ex post facto basis) that the groups being compared did not differ in
ter~s of Y before exposure to X, or on both. Several of the studies
discussed in Chapter 4, pages 127-142, make use of assumptions and
evidence of this sort.
In addition-and,; sometimes, more important-the investigator
must ccnsider whether Ifactors other than X may be the determinants
of Y. In general, this i~ done by introducing additional variables into
the analysis and examining how the relationship between X and Y is
I
affected by these further variables. If the relationship between X and
19 See Chapter 4 for :discussion of nonexperimental studies designed to test
causal hypotheses, and problems of drawing inferences from them.
20 Much of the follo\\,ing discussion is indebted to Hyman (1955, Part III),
and to Kendall and La~rsfeld (1950). For a fuller discussion, consult these
sources.
Y persists even when other presumably relevant variables are intro-

duced, the hypothesis that X is a cause of Y remains tenable. The
procedure followed in the Deutsch-Collins study of the effect of
occupancy pattern in interracial hOllsing projects on attitudes toward
Negroes, described in Chapter 4, pages 138-139, provides an example.
Here it was found that introduction into the analysis of the variables
of political orientation, education, and religion did not affect the
relation between X and Y; whatever their religion, education, or politi-
cal views, white housewives living in integrated projects had mOre
friendly feelings toward Negroes in the project than did housewives
in segregated projects.
In other cases, however, introduction of additional variables may
change the relation between X and Y: it may reduce or eliminate it;
or it may enhance the relationship within one subgroup, reduce it
within another. If the relationship between X and Y is enhanced. in a
subgroup characterized by Z and reduced in a subgroup not character-
ized by Z, we may conclude that Z is a contingent condition for the
relationship between X and Y, or, in other words, that we have specified
a condition under which the relationship holds. If Z reduces or elimi-
nates the relationship between X and Y, we may conclude either that
X is not a determining condition of Y (that is, that the relationship
i~ spurious), or that we have traced the process by which X leads to
Y. Let us consider first the circumstances under which we would con-
-x
clude that an apparent relationship between and Y has been shown
to be spurious, next those under which we would conclude that the
process of the relationship has been traced, and finally those under
which we would conclude that a contingent condition has 'been
specified.
SPURIOUS RELATIONSHIPS
An apparent relationship between two variables, X and Y, is said

to be spurious if their concomitant variation stems, not from a con-
js
nection between them, but from the fact that each of them related
to some third variable or combination of variables that does not serve
as a link in the process by which X leads to Y. The study of attitudes
of foreign students described at the beginning of this chapter provides
an example of a bit of analysis concerned with testing whether an
apparent causal relationship might be spurious. One of the hypotheses
of .the study was that students who associated more with Americans
would be more aware of differences among individuals and subgroups
in the United States-that is, less likely to think of Americans in terms
of stereotypes. Since the phrasing of the interview questions suggested
that generalizations could be made (for example, "Which of the char-
acteristics on this list would you say was most typical of Americans?"),
the score on this variable was b~sed on the number of times the student
introduced into his replies the idea that a generalized answer must be
qualified in terms of individual or subgroup differences.
Let us suppose that, when students were classified as low, medium,
or high on an index of interaction with Americans, the findings were
as shown in Table IA; that is, that students who scored higher in
interaction with Americans were more likely to qualify generalized
statements about them.
TABLE lA
(Hypothetical data)
No. % making
qualifications
Low interaction 107 /
64
Medium interaction 101 72
High interaction 140 77
Total 348' 72
Such a finding would have been in line with the hypothesis.

Nevertheless, since the study was not an experimental one, in which
students were randoiply assigned to the different levels of interaction,
it would be necessary to consider whether other factors might have
accounted for this observed relationship between interaction with
I
Americans and reluctance to generalize about them. One such factor
might be the studeQts' nationality. Suppose that it had already been
found that stude~ts from Europe were more likely to score high in
interaction with Americans
I,
than students from other parts of the
world, and that European and non-European students differed in
many of their views flbout the.United States. A next step in the analysis
would have been to "control for" nationality-that is, to examine the

relationship between interaction with Americans and qualificatiqn of
generalized statements about them separately within the European
and the non-European group.
Suppose that the results were as shown in Table lB. From this
TABLE IB
(Hypothetical data)
Students from Other
Parts of the
Students from Europe World
% making % making
No. qualifications No. qualifications
Low interaction 17 84 90 63
Medium interaction 45 78 56 68
High interaction 90 82 50 68
Total 152 79 196 65
table, one would conclude that association with Americans does not
lead to qualification of generalized statements about them, and that
the observed relation between these two variables stemmed from the
fact that European students were both more likely to associate with
Americans and more likely to qualify their statements about them.
The "No." columns show more than half the European students
(90 out of 152) as scoring high on interaction with Americans,
and almost half the non-European students (90 out of 196) as scoring
low. The percentages in the "Total" row show European students as
more likely to qualify their statements; according to these hypothetical
figures, 79 per cent of them did so, compared with 65 per cent of the
non-Europeans. The columns headed "Percentage making qualifica-
tions" indicate that neither among the European nor among the non-
European students did the extent of interaction with Americans make
any marked difference in the frequency with which students qualified
their generalized statements. In other words, on the basis of ~uch a
pattern of findings, one would conclude that the relationship between
amount of interaction with Americans and the likelihood of qualifying
generalizations about Americans was spurious.
INFERRING CAUSAL RELATIONS 42'
TRACING THE PROCESS INVOLVED IN A RELATIONSHIP
The same study of foreign students provides an example of a

piece of analysis from which it was concluded that part of the process
by which X led to Y had been traced. The investigators had found, as
they had predicted, that foreign students in three types of educational
institution differed in their scores on a measure of "intimacy of associa-
tion" with Americans. Mean scores of students in the three types of
institution are shown in Table 2A.
TABLE 2A
No. Mean "intimacy" score
Students in small colleges 77 2.43
Students in non-metropolitan
universi ties 139 2.17
Students in metropolitan
universities 132 1.69
Such a finding suggests a question as to what about the three types

of institution makes for greatest intimacy of association between for-
eign students and Americans in small colleges, least in metropolitan
universities. This question had been.considered in planning the study.
It had been predicted that the three types of institution would differ
in the opportunities they offered for association between foreign stu-
deqts and Americans, and especially in the extent to which the living
arrangements provided for students would encourage such association.
Therefore each student was given a score on the "interaction-potential"
of his living arrangements-that is, the extent to which the circum-
stances under which he. lived were likely to provide opportunities for
association with Americans. 21 In order to check the hypothesis that
it was largely through 'such differences in living arrangements that
the three types of instit~tion produced differences in interaction with
Americans, type of institution and "interaction-potential" of living
arrangements were considered together in relation to mean scores for
21 For example, living in a fraternity house or having an American roommate
were scored as situations lof "high interaction-potential"; living, without an
American roommate, in an apartment or a rooming house or other building not
primarily occupied by Amrican students was scored as a situation of "low in-
teraction -potential." :
"intimacy of association with Americans." The results are shown m

Table2B.
TABLE 2B
Students in "Low Students in "Me- Students in "High
Interaction- dium Interaction- Interaction-
potential" Living potential" Living potential" Living
Arrangements Arrangements Arrangemen ts
. Mean Mean Me:m
"intimacy" "intimacy" "intimacy"
No. score No. score No. score
Students in
small colleges 7 1.43 20 2.45 50 2.56
Students in non-
metropolitan
universities 38 1.46 40 1.93 61 2.77
Students in metro-
politan uni-
versities 64 1.16 34 1.79 34 2.59
All students 109 1.28 94 1.99 145 2.65
Reading across the rows, we see that the "interaction-potential"

j .of.lh:ing irrangen;ents is strongly related to the student's score on
~intimacy of association wltp Americans. 'Whether we consider the
row for small colleges, for non-metropolitan universities, for metropoli-
tan univer~ities, or for "all students," in every case there is an increase
in the mean intimacy score wi~h an increase in interaction-potential
qf~;1i.vi.ii~ arrangements. Moreover? after reading the -!'No." entries
.a_9!:oss the rows, it becomes clear that type of institution and interaction-
potentia~of living arrangements are highly related; approximately two
.thi~ds the students in small. colleges (50 out of 77) are in living
-6/
arr~ng~ments with high interaction-po~en~ial, compared with less
,tha~;half of those in non-metropolitan universites, and only about a
quarter of those in metropolitan universities. Now, when we, read
the intimacy scores down the columns, we see that the relationship be-
tween type of institution and intimacy of association with Americans
! has been considerably reduced; in fact, for students in living arrange-
ments with high interaction-potential, there is no relationship be-

tween type of institution and intimacy score. From this set of relation-
ships, the investigators concluded that their prediction had been correct
-that is, that one of the major ways in which a given type of educa-
tional institution encourages association between foreign students and
Americans is by providing living arrangements that offer considerable
opportunity for such association. Rather than concluding that the
relationship between type of educational institution attended and
intimacy of association with Americans was spurious, they felt that
they had traced the process by which the former variable influenced the
latter.
CRITERIA FOR DECIDING WHETHER A RELATIONSHIP Is SPURIOUS
What are the differences between these examples that lead us to

conclude, in the first, that the original relationship was spurious, and,
in the second, that we have traced the process through which the
relationship came about? In both cases, introduction of a third variable,
which was related to each of the variables in the initial relationship, re-
duced or eliminated the relation between them. One difference is ap-
parent, however. In the former example, the third variable (national-
ity) was clearly prior in time to the other two (interaction with Ameri-
cans, and qualification of generalized statements about them). In the
latter example, this was not the case. The third variable ("interaction-
potential" of living arrangements) did not occur before the assumed
causal variable (attending a given type of educational institution); it
was concurrent with it, and might be thought of as starting after the
student had begun attending the institution.
The time sequence (known or assumed) of the variables, then, is
an important consideration in deciding whether an apparent causal
relationship is spurious.' If the third variable, Z, which removes the
relationship between tHe originally related variables X and Y, is as-
sumed to have occurre4 before both X and Y, we usually conclude
that the apparent caus~l relationship between X and Y ~s spurious.
If, on the other han~, i is known or assumed to have occ~rred at the
same time as X, or afteli X, we may conclude that what we have done
is trace the process by ~hich X leads to Y, rather than s owing the
..
apparent causal relationship between X and Y to be spuri us .
However, the decision whether we have demonstrated a relation_

ship to be spurious or have traced the process by which it comes about
need not depend only on the time relationships among the variables;
it may involve also a judgment about logical or psychologicaJ relation-
ships. Take the relationship between type of institution attended and
intimacy of association with Americans. Suppose the third variable, Z,
which reduced the relationship between type of institution and inti-
macy of association was not "interaction-potential" of living arrange-
ments but the way the student spent his Christmas vacation. Let us
say that many students who were attending small colleges, and few of
those attending metropolitan universities, spent their Christmas vaca-
tions with American families, and that it was this experience which led
to the difference in intimacy scores. Whether we would consider that
this bit of analysis showed the apparent causal relation between type
of institution and intimaty ,of association to be spurious, or whether
we would consider that It showed the process by which type of institu-
tion influence~d intimacy of association,. would depend on our judg-
ment oftne-;'nature dHhe relationship between type of institution and
mann~r 'Of spending Christmas vacation. If we had reason to believe
tharthose studehts who spent their Christmas vacation with American
fami,lies were visiting with classmates, we would probably infer that
attending a smaller college encouraged relationships that led to these
invitations, which in turn led to greater intimacy of association with
Americans. In this case, we would be inclined to feel that we had
traced something of the process by which attending a certain type of
institution influences intimacy of association with Americans. On the
other hand, we might have reason to believe that invitations. to spend
Christmas vacation with American families came through some central
organization and were distrib~ted without attention to the type of in-
stitution; that it was a matter of chance that more students from small
colleges spent their vacations with American families. In this case, we
would probably conclude that the apparent causal relation between
type of institution and intimacy of association was spurious. In other
words, in in~tances where the third variable, Z, occurs simultaneously
with or after the assumed causal variable, X, the inference that the
apparent causal relationship between X and Y is or is not to be con-
sidered spurious depends on a judgment as to whether Z is or is no1
a consequence of X. 22
In the an.alysis of nonexperimental studies, there is always a pos-
sibility of interpreting spurious relationships as causal. Thus, to have
any confidence in causal relationships inferred from such studies, it is
necessary to subject them to the critical test of eliminating other pos-
sibly relevant variables. For this reason it is important to collect, in the
course of the study, data on possibly influential variables other than
those with which the hypotheses of the study are centrally concerned.
SPECIFICATION OF A RELATIONSHIP
The inference that an assumed relationship is spurious or that

one has traced the process through which it occurs is made only if the
introduction of a third variable or combination of variables leads to
the reduction or elimination of the initially found relationship. How-
ever, the introduction of a third variable may have still another effect;
it may lead to an intensification of the relationship within one sub-
group and its reduction within another. In this case, we say that we
have specified a condition under which the relationship holds. 23
Let us illustrate the process of specification. Suppose that, in a
hypothetical study, we have found a relationship between income and
educational level, such as that shown in Table 3A.
TABLE 3A
(Hypothetical data)
Percentage Earning Percentage Earning
Number Less than $3000 $3000 or More
Graduated from high school. 500 32 68
Did not graduate from I
high school 500 82 18
' I
Total 1,000 57 43
The relationship ;is a fairly marked one. However, we may decide
that it requires furthef specification; we may wish to know more about
22 A number of procedures that may be built into the plan of data collection in
order to provide a basis for judgments about the time sequence of variables and
their logical connections are discussed in Chapter 4, pages 130-136.
23 "Specifying a condition under which a relationship holds" is essentially th~
same process as demonstrating "interaction" between variables by means of all
analysis of variance (see Chapter 4, pages 122-125).
the conditions under which it occurs. Suppose the thought strikes us

that racial discrimination might severely curtail the advantages of
education in the competition for remunerative employment. If so, we
would "break down" Table 3A by race. Let us say that we now obtain
Table 3B. The hypothetical data in Table 3B show a very different rela-
TABLE 3B
(Hypothetical data)
Number Perct:ntage Percentage
Earning Less Earning $3000
than $3000 or More
NEGRO WORKERS
Graduated from high
school 100 80 20
Did not graduate from
. high school 100 90 10
Total 200 85 15
WHITE WORKERS
Graduated from high
' school 400 20 80
~ Did not graduate from
school 400 - 80 ~ 20
Total 800 50 50
tionship between education and income for Negroes than for whites.
F9r whi~es, it ISO somewhat higher than in the original relationship; for
Negroes It is considerably lower. Thus, the breakdown of the original
~relationship by race has helped to specify some of the conditions
,tinder whibh it is more pronounced and some of the conditions under
'which it is)~ss pronounced. . --
- i_ I , 1 t
The Use of N onquantified Data24

in Analysis and lnterpretation
Every reader of social research publications is familiar with the
fact that raw data,25 in the form in which they were collected, are often
24 It is customary to refer to data that are presented essentially in the form in
which they were collected as "qualitative" data. The term is not too ,happily
chosen, because it implies a contrast to quantitative data. Both qualities and
quantities can, of course, be at least counted in social research. But not everything
that can be quantified in principle is actually quantified in a study. In other
words, non quantified material is not necessarily nonquantifiable material.
25 The term raw data is used synonymously with nonquantiJied data.
USE OF NONQUANTIFIED DATA 433
used in conjunction with data that have undergone analysis amI are on
a higher level of abstraction. For example, Jahoda and Cook (1952),
in considering conditions that enhance or inhibit the effects of the
United States loyalty and security procedures on freedom of thought
among government employees, state: " ... the work-relationship
factor which seems, from our exploratory interviews, to influence most
strongly the impact of the security measures is the relation of employees
to their supervisors." This statement !s almost immediately followed
by a verbatim quotation from an interview, a response to a question
about a hypothetical situation in which a loyal government employee
believes that he may be suspected of disloyalty because a friend has
been accused of being a Communist:
In our department I'd advise him to stay and fight, and I'd
be confident of a good chance of winning. I trust the head of
the department; he would support me. But that's the only thing
on which I base my advice. In every other case, no matter who
the suspect is, however honest, outstanding and competent, I'd
advise him to run as quickly as possible. In our place there is no
need to change. But I don't believe in being a martyr. Every-
where else I might act differently.
There can be little doubt that the insertion of such material in-
creases the readability of social science documents. Laud~ble-a;t~at
is, increased readability in itself presents no compelling scientific
. ~!
reason
for the inclusion of such anecdotes and verbatim quotations.~Our pur-
pose here is to discuss the scientific functions of using 's;U~h material
in the course of analysis. f(
Raw data can be used in analysis and interpretatio~ ,whether or
not they have been fJuantified in all aspects. Even if virtually 'every
aspect of a statementI has been converted into categories, recourse to
the original makes it gossible to see the whole context. More frequently,
however, not every aspect of the material has been categorized. In
these cases, recourse to the raw material has the additional advantage
of bringing neglected~aspects to the fore.
The use of raw aata in the course of analysis fulfills two distinct
functions: to illustr~te the range of meaning attached to anyone
category, and to stimulate new insights. We shall deal with these two
.tunctions separately.
ILLUSTRATING THE MEANING OF CATEGORIES
As we pointed out earlier in this chapter, the sharpness with

which categories can be defined varies according to the nature of the
raw data, the type of problem under investigation, and the situations
to which the data refer. Suppose that one wishes to study by observa-
tional methods the nature of the interaction between labor and man-
agement in a joint production committee, such as those established in
British industry. Depending on the anticipated level of analysis, sev-
eral types of categories for summarizing an observational session can
be set up. One might count the frequency of spontaneous remarks
exchanged between members of one group and members of the other,
or one might observe the seating arrangements; categories of this kind
are so easily given unambiguous definition that they need no illustra-
tion. But more complex categories can be established for the same
purpose, such as interactions between the two groups that indicate
feelings of equal status, subordination, or superior status. A worker
opposing the point of view of management might say, "I beg to dis-
agree"; a managerial representative might say to a worker, "You are
wrong." To classify these two answers in different categories of status
consciousness involves a considerable amount of judgment about
both the implications of the statements and the exact meaning of the
categories. In dealing with such data, illustrations are necessary for
three purposes: to help the coder understand the meaning of the cate-
gories, to help keep the actual nature of the material assigned to each
category clear in the investigator's mind as he carries out the analysis
and interprets his findings, and to help the reader understand the
nature of the variables described and the investigator's reasoning.
The use of illustrations in defining categories for coders has. al-
ready been discussed; their use In aiding the reader's understanding is
obvious. Let us consider their importance as guides for the analyst in
keeping aware of the real nature of the material included in various
categories.
In an analysis of reasons that prompted people to move 'into a
housing project, Merton and his colleagues (forthcoming) intro-
duced a distinction between those who had to leave their old residence
and those who wanted to leave it. In these broad terms the two types of
motivation appear to be clear-cut alternatives. However, illustration of
both categories indicated that the dividing line, although sharp in
some cases, was vague in others. One tenant explained, "The house
was sold from under us." Clearly, he had to move. Another one com-
mented, "I didn't like the people in the neighborhood." Obviously, he
wanted to move. But one tenant said: "We lived with my in-laws.
Sometimes I thought I'd go crazy with all the bickering. I had to get
out sooner or later. I heard about this place here and I liked what I
heard; so we moved right out." Did he have to move, or did he want to
move? Whether the economic compulsion under which the tenant
whose "house was sold from under us" acted is to be classified in the
same category as the psychological compulsion under which the tenan~
who could not get along with his relatives acted depends largely on
the purposes for which these categories are established. Whatever the
decision, it is necessary to keep in mind the wide range of motivation
covered by each category if the further use of the distinction between
those who had to. move and those who wanted to move is to be psycho-
logically meaningful. This is facilitated by reference to examples from
the raw data whenever the categories are used in a new context.
STIMULATING NEW INSIGHTS /
Whatever the design of a study, and no matter how refined its

statistical procedures, inspection of the raw data may lead to important
insights. These may bear on the course of the analysis, or they may help
to Illake clear the nature of the relationships between variables that
have been demonstrated statistically, or they may lead to hypotheses
for further research.
The inspection of ',nonquantified data may be particularly helpful
if it is done periodically throughout a study rather than postponed to
the end of the statistical analysis. Frequently, a single incident noted
by a perceptive observer contains the clue to an understanding of a
phenomenon. If the so~ial scientist becomes aware of this implication
at a moment when he can still add to his material or exploit further
the data he has already oollected, he may considerably enrich the quality
of his conclusions.
During the period, of data collection in a study of the social and
ANALYSIS AND INTERPRETATION
psychological effects of long-term unemployment in an Austrian vil-

lage (J ahoda-Lazarsfeld and Zeisl, 1932), one of the field workers had
a casual conversation with a small boy. This boy was at the age when
the notion of becoming chieftain of an Indian" tribe appeared a suit-
able vocational goal; "But," he added, "I'm afraid it will be hard to get
the job." This incident suggested that the bitter reality of unemploy-
ment had affected even the realm of fantasy. The investigators were
immediately confronted with the question whether this was a unique
case or whether the relationship was a general one. It was decided to
).-ansform the hypothesis-that parental unemployment restricts the
fantasies of children-into a situational context of wider applicability
than the wish to become an Indian chieftain, and to test it statistically.
The opportunity to do so arose just before Christmas. At that time a
sample of children of unemployed parents and a sample of children of
employed parents were asked for their Christmas wishes.
, The wishes of .
both groups transcended the possibilities of fulfillment, considering
the budgetary conditions even of employed parents; bicycles, cameras,
and other expensive articles were frequently mentioned. Yet when
these unrealistic wishes were translated into current prices, it was
found that those of children whose parents were unemployed would
cost significantly less than those of children whose parents were em-
ployed. This statistical finding was regarded as bearing affirmatively on
the hypothesis suggested by the qualitative account of the boy's voca-
tional dream.26
(Much of the analytical effort of social scientist~ is devoted to
establishing relationships between objective characteristics of a group
of people and their subjective reactions:) However, the demonstration
that a relationship exists does not in itself provide an understanding
of the way in which the factors are related. The scrutiny of the raw data
may be rewarding in the search for such understanding. Thus Merton
et a1. (1946), in a study of radio listeners' responses to a war bond
drive, based their quantitative analysis on the hypothesis that people
26 Since there was no corresponding measure from the time before the parents
I were unemployed, the hypothesis cannot be considered as 90nclusively demon
. strated. It is possible-though unlikely-that even before the period of unem
ployment, children of the subsequently unemployed parents mig?t have ex
pressed more modest wishes than the other group. This is an example of the
limitations of nonexperimental studies.
who had relatives in the armed forces would react differently to the
themes of the drive than those without relatives in the armed forces.
By and large, persons with relatives in the armed forces were, indeed,
more impressed by and willing to respond to an appeal to make sacri-
fices. Yet the relationship was not as strong as the investigators had
anticipated. Many persons with relatives in the armed forces remained
unresponsive. In a search for explanatory variables, Merton turned
back to the interviews. He reports:
An analysis of the interviews provides the clue. The unre-
sponsive persons had little basis for acute tears and anxiety con-
cerning the safety of sons or brothers in the service [italics added]
... their kin were stationed in this country. . . . Nor could a
chief gunner's mate, the husband of another informant, be a
source of anxiety, since at the time he was home on leave and
listening with his wife. . . . Thus, the emotional context for
selective listening was significantly different for the two categories
of informants although both had close relatives in the Army or
Navy.
Here, Merton discovered fear and anxiety-that is, the emotional
context-as a possible explanatory variable for the different responses,
through scrutiny of the actual interviews.
From what has been said so far, it should be,c1ear that this use of
nonquantified data as a source ofclues for the explanation of relation-
ships does not constitute a tracing of the process involved, in the sense
discussed in pages 427-429. Unless the suggested explanation is one
for which a measure can be found within the study, thus making pos-
sible the statistical testing of the suggested process, the insight pro-
vided by the raw data remains simply a promising suggestion as far as
that particular studylis concerned. But if research is regarded as a never-
ending search for in;terrelationships, in which one study provides the
basis on which another can build, the suggestion of plausible medi-
ating variables becofues as important a step as the statistical demon-
stration of relationships.
The search for explanatory concepts through reference to the raw
data often becomes theI,
only possible way of dealing with cases that do
not show the same "trends or patterns of relationship as most of the
cases in a study. Such cases, by the very fact that they are exceptions,
occur in such small numbers as to preclude a refined statistical analysis.

Yet study of the exceptions may make it possible to refine hypotheses
so that the frequency of exceptions can be reduced. Thus, Bettelheim
and Janowitz (1950), in their study of American war veterans, investi-
gated the relationship of frustration to anti-minority feelings. Among
other questions they asked the veterans, "Do you feel that you got a
bad break in your army career?" Of those who were tolerant of minor-
ities, 62 per cent21 said that they had had a "good break," while of
those who had outspoken and intense anti-minority feelings, only 43
per cent gave this reply. These figures establish a correlation between
feelings of being fairly treated and lack of prejudice, but with a rela-
tively high frequency of deviation. After turning back to their raw
data in an attempt to explain the exceptions, the authors report:
An examination of the interview records suggests interest-
ing hypotheses which may partially explain why some of the men
deviated from prevalent patterns. All three tolerant men who
claimed that they had gotten a bad break in the army gave the
same reason for their "bad break." In ? general way they all
Lr~sented the treatment they received from their officers. For
example, a thirty-three year old air corps sergeant said: "I wasn't
treated right by those damn officers. There was no respect due to
an enlisted man from an officer." Thus, their feeling of having
had a "bad break" was not due to an over-all attitude toward the
army but rather to a specific resentment of officers.
On the other hand, it is also striking that contrary grounds
were offered by the intolerant men for their thinking they had
gotten a good break in the army. . . . For example, one intol-
erant man stated that what he liked most was: "The discipline
and the strong order. . . . If I were single I'd make a career out
of it. I liked the physical culture, fitness. . . . "
Clearly these . . . intolerant men felt that they had' had a
good break for reasons markedly different from the matter-of-fact
attitude of having had a "good break" in the army which char-
acterized tolerant veterans.
As the authors develop their argument, it becomes clear tl:tat the
apparent deviation from the pattern established in the figures quoted
27 Of 61 tolerant veterans, 38 (62 per cent) said they had had a good break,
12 (21 per cent) that they had had neither a good nor a bad break, 3 (5 per
cent) that they had had a bad break, and 7 (12 per cent) made other remarks.
USE OF NONQUANTIFlED DATA 439
above is less of a deviation than originally suggested. One is led to
conclude that the statement "I had a good break" or "I had a bad
break" is only a crude indication of the basic relationship. The study
of the concrete answers of those who deviate from this crude relation
reveals that what is considered as a good or a bad break is more decisive
than the phrase. And this concept of what constitutes a "good break'~
is, in turn, a function of characteristics of the person.
Thus, the concern with deviant cases through a study of their
concrete responses may lead to a refinement of hypotheses so that they
result in fewer exceptions than did the original hypothesis. As indi-
cated before, this process of refinement is unending. And the refine-
ment itself needs to be regarded with caution until it can be checked
against new data. "Explaining away" the exceptions suffers from the
weaknesses of all ad hoc reasoning (see pages 419-421).
The examination of nonquantified material often leads to ideas.
for further research by revealing aspects of the phenomenon that have
not been sufficiently studied in the particular investigation. Recogni-
tion of those aspects needing further study may also help to point up'
limits to the generalizability of findings from the given study. An ex-
ample will illustrate the latter point. One of the research purposes in
Patterns of Social Lite (Merton et at, 'forthcoming) was to discover
whether living in a bi-racial housing project significantly influenced
prejudiced attitudes. Several statistical indices suggested that for a'
considerable part of the tenant population, there had been a definite
improvement in relations with the other race. For example, informal'
contacts, such as are implied in having friends and acquaintances.
among members of the other race, were more frequent among people
who had lIved in the project for a longer period. In the search for
illustrations, the analyst used an excerpt from an interview with 3i
white woman. This ~oman explained that originally she had been,
frightened by the ide\} of living in the same project with Negroes, that
she had avoided them when possible, and that she had generally re-
jected the idea of informal contact with them. She went on to say that
gradually she had ,realized how unfounded her fears had been, she
had made friends wl~h some Negro women, and now she regarded
herself as a champion of good race relations in the project. By way of
climaxing her account, she added: "1 am friendly with the men too.
When I walk thraugh the praject, many a Negra man shauts fram
across the street, 'Hello., Helen.' "
So. far, the excerpt has mainly an illustrative function. Fallawing
a table shawing the frequency af such changes in the caurse of time,
this vivid accaunt would probably lead all readers to assume that race
relations were considerably impraved by living in the praject, withaut
any awareness af qualificatians other than thase indicated in the table.
Hawever, the excerpt cantinues: " ... 'Hello., Helen.' [a pause]. Of
caurse, I'd .faint if they did this to. me in the main street in front af
everybody."
This afterthaught intraduces an impartant qualificatian af the
results. It suggests that the relatians af the white tenants to. the Negro
tenants are only in part a cansequence of their own beliefs and feelings
abaut Negraes; in part they are also. a cansequence af the white tenants'
perceptian af sacial approval or disappraval af interracial assaciation.
Within the hausing project, the white waman in the illustration ap-
parently perceived appraval af such assaciatian. Outside the praject,
"an the main street," she apparently perceived disappraval. Evidently
her behavior taward Negroes varied accardingly. The illustratian sug-
gested nat anly a qualificatian af the findings but a lead for further
research.
In the example given, the implication is so. dramatically obvious
that no. analyst who. examined the raw data wauld be likely to. averloak
it. In ather cases, suggestians far further research are nat so. abviaus;
they must be deliberately sought in the raw data. Although the con-
firmatian or refutatian af hypatheses requires that a study be set up
with these hypatheses in mind, the purpose af discavering promising
leads for investigatiQ~ is often served best by the painstahng inspec-
tian af nanquantified data.
12
TJ-tE .RESEARCH REPORT
What the Report Should Contain
Modi6.cations tor Shorter Reports

I
The Style of the Report
,\ thing is not necessarily true because uttered badly, nor false
because spoken magnificently. AUGUSTINE
T HE RESEARCH TASK is not completed until the report has been

written. The most brilliant hypothesis, the most carefully de-
signed and conducted study, the most striking findings, are of little
import unless they are communicqted to others. Many social scientists
seem to regard the writing of a report as an unpleasant chore tacked on
to the end of the research process but not re?lly an inherent part of it.
To be sure, this stage requires a set of skills somewhat different from
those cailed for by earlier stages of research; and much of the excite-
ment of discovery may have worn off by the time the investigator shifts
the focus of hi~ attention from analysis of his data to preparation of the
report. Nevertheless, communication of the results so that they enter
the general store of knowledge is an essential part of the investigator's
responsibilities, which should receive the same careful attention that
earlier stages do. Moreover, if sufficient time has been allowed 'for prep-
aration of the report (which is seldom true), the investigator may even
find himself enjoying the opportunity it presents to fit pieces' together
into a larger whole, to consider th,e implications of his findings, to mull
over the gaps or new questions raised and to think about what kinds
of future research might provide answers.
Perhaps the most important point to be kept in mind when
writing a report is its function: The purpose of a report is not com-
munication witb oneself but communication with the audiepce. This
statement may appear obvious. However, a perusal of social research
documents will show that all too many bear the stamp qf a struggle
for clarification of the author's own thoughts; they are not designed
.....
.,..
~
THE RESEARCH REPORT 443
to communicate with an audience about problems that it would find
of interest.
Emphasis on communication with an audience demands that one
be clear about the type of reader for whom a given report is intended.
A report directed to fellow social scientists will be different in many
ways from one addressed to administrators who may take action on the
basis of the findings; both will differ in some respects from a report
whose purpose is to inform the general public. Whatever the audience,
two broad questions should be considered in planning the report: (1)
What does this audience want or need to know about the study? (2)
How can this information best be presented?
In this chapter, we shall discuss the writing of a report for a scien-
tific audience. 1 The suggestions are directed primarily toward the prep-
aration of a detailed report, such as a thesis or monograph. 2 Shorter
reports-articles intended for publication in a social science journal,
for example-involve somewhat different emphases; some of these
differences will be pointed out.
What the Report Should Contain

The social scientist who reads a research report needs to be told
enough about the study so that he can place it in its general scientific
context, judge the adequacy of its methods and thus form an opinion
of how seriously the findings are to be taken, and-if he wishes-repeat
the study with other subjects. In order to give him the necessary in-
formation, the report must cover the following points:
1. Statement of t~e problem with which the study is concerned.
2. The research procedures: the study design, the method of
manipulating the independent variable if the study took the
form of an experiment, the nature of the sample, the data-
collection techniques, the method of statistical analysis.
I
1 The presentation of'research results to audiences other than social scientists

is discussed in the following chapter.
2 For other discussions of the preparation of reports, see Good (1941), Ogburn
(1947), or Parten (1950j. Chapter 17 of the latter reference gives not only
many practical suggestions but an extensive bibliography on the preparation of
reports.
444 THE RESEARCH REPORT
3. T}!~ t~~\.,lt.~.
4. The irnpliC'ations drawn from the results.
This list provides the major headings for an outline of the research
report. Each point is discussed in more detail below.
STATEMENT OF THE PROBLEM
It was pointed out in Chapter 2 that the first step in the research
process is a precise formulation of the question to be investigated.
Ordinarily, the research report also starts with this statement of the
issue on which the study was focused. \Enough background should be
given to make clear to the reader why the problem was considered
worth investigating. Since a social science audience is likely to be more
interested in contributions to general knowledge of human behavior
than in the solution of a specific practical problem, the report to such
an audience usually stresses the relevance of the investigation to some
aspect of psychological or sociological theory.3 For example, a study
undertaken at the request of an institution to ascertain reactions to a
proposed change in personnel policy may be planned and carried out
in such a way that it provides evidence on the manner in which an
individual's role within the organization influences his perception of
the new policy. The report to a social science audience 'would quite
properly stress this latter aspect rather than the concrete issue with
which the specific institutiqn is concerned.
However"it should be recognized that not all studies have a direct
bearing on theoretical issues, and th~t the relevance of a study to s~me
theoretical point may be_come apparent only as one seeks to 'under-
stand the findings. At the present time, in the social sciences, many
studies are of necessity carried out without the guidance of a systematic
theory. When this is the state of affairs, there is no reason to disguise
it; attempts to invent theoretical relevance usually strike the reader
as pretentious.
In addition to indicating the practical or theoretical importance of
the question inves~igated, the statement of the problem should include
3 For a discussion of the relation between research and theory, see <;::hapter 2,
pages 44-47, and Chapter 14.
WHAT THE REPORT SHOULD CONTAIN 445
a brief summary of other relevant research, so that the study may be
seen in context; the hypotheses of the study, if any were fonnulated;
and definitions of the major concepts employed (see Chapter 2). The
connections among these elements should be made clear; that is, the
logical sequence of ideas leading from the existing theory and relevant
research findings to the hypotheses and concepts of the study should
be explicitly indicated.
THE RESEARCH PROCEDURES
The scientific reader needs to know in considerable detail how the

study was carried out. What was its basic design? If the study was an
experimental one, just what were the experimental manipulations?
(For example, was "threat" established by telling the subjects that
they were about to take a very difficult test; which would determine
their grades in a course, or by shouting, "Fire!"?) At what point or
points were the measurements taken?
If the data were collected by means of questionnaires or inter-
views, exactly what questions were asked? (The questionnaire or inter-
view schedule is usually given in an appendix.) How much and what
kind of experience had the interviewers had, and how were they trained
for this part'icular study? If the measurements were based on observa-
tion, what instructions were given to the observers?
The reader also needs to know how the observations or replies
to qu~stions were translated into measures of the variables with which
the study was concerned. (For example, which questions were taken
into account in estimating "morale"; or what kinds of leader behavior
were classified as "democratic"?)
Regarding the sakple used in the study, he should be told: Who
were the subjects? How many were there? How were they selected?
These questions are crucial for estimating the probable limits of gen-
eralizability of the finaings. Are elaborate conclusions being drawn on
the basis of responses of ten college sophomores, selected because they
happened to be friends of the investigator? Were only housewives
interviewed? If so, isI, there any basis for extending the findings to
people in general? Intensive study of a small number of cases that do
not constitute a random sample of any specifiable population may be
extremely valuable. Nevertheless, the number and characteristics of

the subjects on which the findings are based should be clearly stated,
so that the reader can draw his own conclusions about the applicability
of the findings to other groups.
The scientific reader is also concerned with the statistical analysis
of the data. What techniques were used to determine whether
groups differing in certain background characteristics. or subjected
to specified treatments, showed differences on a given measure that
were significantly greater than might have been expected by chance?
Ordinarily it is sufficient simply to name the technique used, and to
indicate the level of confidence one has accepted in deciding whether
differences are to be considered significant (see Chapter 11, pages
414-421). Only if the statistical technique is a new one need formulas
be given. If one has used a technique that is, strictly speaking, not ap-
propriate to the data, this fact should be noted, and explanations made.
(For a brief indication of the kinds of data with which specific statisti-
cal tests may appropriately be used, see Chapter 5, pages 189-197; for
more detailed discussion, see any statistics te~tbook.)
THE RESULTS
The basic rule in presenting findings is to give all the evidence

relevant to the research question asked, whether or not the results are
in accord with the investigator's views. This is the cardinal rule of
scientific reporting. Unlike other writers, the scientifi~ author is not
free to choose what he will include'and what he will leave out in terms
of the effects he wishes to create.
Nevertheless, he must still make some selection in terms of what
is relevant; not every table worked out in the course of the analysis
can or should find a place in the report. How does one decide what is
relevant? Guidance comes primarily from the research problem and
from the hypotheses, if any, with which the study was concerned. Sup-
pose one wishes to evaluate the effects of two different metl)ods of
selecting discussion leaders. The setting chosen might be a class of a
'hundred students; divided into discussion groups of five students each,
with one of the students in each group serving as its leader. In half of
the groups the leader is elected by the group from among its members;
in the other half, he is appointed by the instructor on the basis of
scores on an initial questionnaire. Let us say that the investigation was
interested in the effects of the different methods of selecting leaders
on the following aspects: (1) the satisfaction of members with the
group's functioning, (2) the attendance of group members, (3) the
extent of participation in group discussion, (4) the concern of mem-
bers with having the group accept their views or with working out a
position acceptable to the entire group, (5) grades in the course.
Whether or not explicit hypotheses were formulated, the investi-
gator is obligated to report on each of these points-even if all he can
say about a given one is that he was unable to collect the data needed
to answer the question. He may h:lVe found, for example, that the
ratings by which he had intended to"get data relevant to point (4) were
made so unreliably by different observers that they were not usable.
This fact should be reported. If his data show differences between the
two types of group on points (1), (2), and (3), but not on (5), this
too should be reported.
But, as has been pointed out throughout this book, a study is not
completely "jelled" at the time the research problem is formulated.
During the course of the inves~igation, a more ad~quate statement of
the problem itself may be developed, new hypotheses may emerge,
unforeseen relationships may appear. Therefore, while the original
formulation provides the basic point of reference for the report, there
must also be room to include subsequent developments. For example,
Selltiz, Hopson, and Cook (1956), in a study described in Chapter 11,
pages 386-391, started out with hypotheses about the effects of
association with Ame~icans on the attitudes of foreign students toward
the United States. Information about certain background character-
istics of the students, including their nationality, was gathered primarily
in order to check on 'conditions that might influence the relation be-
tween personal association and attitudes. As it turned out, the hypoth..
eses about the effects of personal association were not supported, but it
became clear that th~!e were marked differences in attitude between
students from different parts of the worM. Therefore, although the
initial plan of the stu9Y was not directly co.(1cernci with :aatior~al back-
ground as a determinant of attitudes, it became appropriate to devote

a section of the report to discussion of that relationship.
For every finding that involves acomparison between groups or a
relationship between variables, the level of statistical significance
should be reported. Suppose, for example, that in our group leadership
study, attendance in the groups with elected leaders averaged 80 per
cent, in those with appointed leaders, 75 per cent. Unless the reader is
told the results of the test of significance, he is in no position to judge
whether this finding indicates a slight superiority of elected leadership
or whether the difference can reasonably be attributed simply to chance
variations. Ordinarily, in a detailed report for a scientific audience,
every finding that is considered sufficiently important'to be stressed is
accompanied also, either in the text or in an appendix, by a table or
graph or chart showing the relevant data.
The guides suggested so far originate mainly in the initial formu1 9-
tion of the research problem and the hypotheses and in the statistical
results of the analysis. However, not all studies begin with a research
problem stated in sufficient detail to provide a basis for outlining the
general points to be covered in the report, and' not all studies make use
of quantitative data. In an exploratory study, for example, the appro-
priate content and organization of the report are much les~ ,clearly
suggested by the study design and analysis of the data; the ipVeStigator
must rely more on his own~ judgment"in deCiding the outline of his
report. Neverthless, it is still necessary that he state clearly tne' prob-
lem with which he was concerned, the procedure by which he worked
on the problem, the conclusions at"which he arrived, and the bases for
his conclUSIons. 1
DISCUSSION OF IMPLICATIONS
A bare statement of the findings is usually not enough to convey

their meaning; usually the reader is interested in their implications for
the general understanding of human behavior. Discussion of these
implications is sometimes combined with the presentation of the data;
sometimes it is placed in a separate section. Wherever the discussion
appears, it usually includes three major aspects:
1. A statement of the'inferences drawn from the findings in this
particular situation which may be expected to apply in similar circum-
stances. The inferences may be at a level quite close to the data, or may
involve considerable abstraction. For example, in our group leadership
example, if the investigator has found more satisfaction, better attend-
ance, more participation, and higher grades in groups that elected their
own leaders, he may simply conclude that in similar situations, elec-
tion of the discussion leader will have similar effects. However, he may
wish to carry his inference to some higher level of abstraction, espe-
cially if there is some partially developed theory to which he may link
his findings, or if there have been other studies in which the specific
phenomena are different but can be understood in terms of the same
abstract principle. Thus, for example, he may treat election of the
group leader as an example of the more abstract concept autonomy.
2. As a qualification of these inferences, the investigator should
note conditions of his study that limit the extent of legitimate generaliz-
ation. He should, for example, remind the reader of the characteristics
of his sample and the possibility that it differs from a larger population
to which one might want to generalize; of specific characteristics of
his method th'at might have influenced the outcome; of any other
factors he is aware of that might have operated to produce atypical
/
results.
3. Finally, the discuSsion of implications of the findings will usually
include relevant questions that are still unanswered or new questions
raised by the study, perhaps with suggestions for the kinds of research
that would help to answer them.
I
THE SUMMARY
It is customary to conclude with a very brief summary, restating

in barest outline the problem, the procedures, the major findings, and
the major conclusions drawn from them.
I
,Modifictations for Shorter Reports

Frequently, a detailed report such as that discussed above does
not seem appropriate. The study may not seem important enough to
warrant a great investment of time and money in writing and producing

the report, or of the audience's time in reading it. In fact, the more
usual report is one prepared for a psychological or sociological journal,
which must ordinarily be kept to a few printed pages. Such reports
must still give the reader the basic facts he needs to know, but in ab-
breviated form.
In an artjcle written for publication in a social science journal, the
discussion of previous research may be omitted ~xcept for reference
to summaries that have appeared elsewhere. Or, if the significance of
the study being reported stems in large part from its relation to certain
other studies, these are mentioned and the crucial points on which
they focus are briefly summarized, but the reader is referred to reports
of those studies for additional details. Similarly, the relation of the
study to theoretical issues is stated with the minimum of detail needed
t? make clear the general bearing of the study on social science th~ory.
Space is saved also by omitting the data-collection instruments,
unless these are absolutely essential for understanding of the study.
Ordinarily, in a journal article, the reader _is told simply that the data
were collected by a participant observer, or by an observer watching
from behind a one-way screen, or by means of .interviews or anonymous
questionnaires. A few questions, observational categories, etc., may be
included. If the method of data collection has been described in print
elsewhere, reference is of course made to this description. If no pub-
lished description is available, and the methodological innovations
are of general import, the aut40r may publish separately a description
of his new techniques. He should, send his methodological material .to
the American Documentation Institute of the Library of Congress,
where it will be available to other investigators who wish to examine
it. 4
Limitations of space in a journal article require that one be much
more selective about what aspects of the findings are to be reported. To
be sure, the requirement that both positive and negative evid~nce be
reported still holds. However, negative findings may qe stated very
briefly, and without supporting tables or graphs. In presenting the
4 For a discussion of these and other procedures appropriate to the publication
of articles in scientific journals, see the Publication Manual of the American
Psychological Association, 1957 Revision.
MODIFICATIONS FOR SHORTER REPORTS 451
findings, if the study is a fairly elaborate one, the investigator may
have to decide which are most important, and present only those.
Points that seem interesting and might warrant discussion in a longer
report, but on which the data are only suggestive, must usually be
omitted from a journal article. Tables, charts, and graphs will be used
much more sparingly than in a longer report.
If the study has two or more aspects that can be discussed inde-
pendently of each other, it may be advisable to write separate articles
rather than to try to crowd too many different findings into a single
article.
The Style at the Report

The basic qualities of good scientific writing are accuracy and
clarity. A style that is pleasing from a literary or esthetic point of view
(I
is an "extra." The investigator reporting his research has no obligation
to be colorful or elegant in his writing, or to hold the reader's interest
by a lively style-though there is no reason why he should repel the
reader by asking him to wade through long, involved, unclear, or pom-
pous sentences.
The first step-and with it the battle is half won-is t9 decide just
what information you want to convey and how the various points are
related to one another. Writing a detailed outline is an invaluable aid
J.t this stage. Some experienced writers can sit down with pencil or
typewriter, with only a general notion of the final shape of the report,
and produce a clear and well-organized account of their research. For
most people, however-and especially for beginners-this is a very in-
efficient way to proceed. W~iting "from scratch" relies on ideas some-
how organizing themselves a,nd appearing in the proper order. If they
do not, the resulting draft jof the report will be a garbled account
requiring complete rewriting# it is to make sense.
Preparing a detailed ou~line allows one to concentrate exclusively,
at first, on what is to be said, without worrying about how to say it.
One constructs the skeletoq of the report; looking at the bare bones,
one can more easily see whether any important points have been left
out. Moreover, the outline, form, with its clear indication of major
topics, subtopics, and still further subdivisions, almost forces attention

to logical relationships within the material. If One is writing without
an outline, he may write two or three paragraphs describing his sample,
then go on to an account of his data-collection procedure, then add
an afterthought about the sample, without noticing what has happened
to the line of thought. The outline form clearly marks the shift from
one topic to another and thus facilitates the grouping together of all
the points about a given topic.
Once the outline has been written, it is a good idea to go baokover
it, to see whether anything important has been omitted and whether
ideas that have been grouped together do logically belong together.
It may be helpful to have someone else read and comment on it. Then,
if changes in the basic structure of the report seem to be needed, they
can be made before the actual writing starts.
Whether one worries about writing style in the course of pro-
ducing the first draft is to some extent a matter of individual disposi-
tion. Sonie experienced writers spend a long time over each sentence,
'carefullY'chQosing words that will best ,convey their meaning. How-
ever, when the purpose is to convey information rather than to achieve
a literary production, it seems likely that in the long run.time is saved
by writing the first draft as quickly as possible. Once it is on paper, one
can go back and rewrite sentences and :paragr!1phs, fortified by the
knowledge that he has produced aHeast a fir$t draft of the ~eport.
f[ At some point, however, attention must be given to style. There is
little that can be said about t~is, except to stress the value of simplicity
and correct grammatical structure. A common fault is pretentiousf1ess.
It is, of course, entirely appropriate to use terms that have a technical
meaning in social science. But the bulk of a social science report does
not consist of technical language, and there is no good reason for con-
sistently using four-syllable words instead of one-syllable wor9s with
essentially the same meaning. Similarly, two or three relatively simple
sentences may convey an idea more dearly than one complicated
sentence with a number of entangled clauses.5
Many students emerge from their language training with the view
5 For examples of unnecessarily complex writing and ways of simplifying, see the
Publication Manual of the American Psychological Association, 1957 Revision. For
a more detailed disC'tt<;sion of "readability," see Flesch (1949, 1954), and Flesch
and Lass (1955).
THE STYLE OF THE REPORT 453
that the rules of grammar are an arbitrary cOflstruction of pedants, with
which none but teachers of English need concern themselves. This is
a serious misconception. One of .the major functions of grammatical
rules is to help us use language in such a way that it will convey our
meaning with a minimum of ambiguity. The placing of commas
around certain kinds of clauses and not around others, for example, is
not an arbitrary matter; it is a way of specifying the relation of a clause
to the word it modifies.
A good dictionary is an essential part of a writer's equipment. A
good handbook of composition may also be helpful; for example,
Woolley and Scott (1944) discuss words, sentence structure, punctua-
tion, paragraphs, and organization. For specific problems in preparing
a manuscript for publication, one may need to consult more specialized
manuals; for example, the University of Chicago Press Manual at Style
or the Publication Manual of the American Psychological Association.
The preparation of tables, graphs, etc., also requires care. This is
not the place to discuss the preparation of such material in detail; how-
ever, a few general suggestions may be helpful. All tables and graphs
should be clearly labeled. The title should state briefly the subject
matter of the table; any necessary qualifications or explanations should
be given in a footnote to the table. Row and col~ headings should
be as short as is consistent with identifying the<fata being presented.
Both axes of a graph should be clearly labeled to show both the
variables being represented and the units of measurement. The number
of cases on which the findings are based should be indicated in the
table or graph as well as in the text; this is especially important when
figures are given in percentages. Although the tabular or graphic
presentation should be clear enough so that a reader can- identify the
major points it is in~ended to convey, it is customary to state the majer
findings in the text ~s well. In this way, the reader who does not want
to take time to studX tables can get all the necessary information simply
by reading the text.: (More detailed suggestiolils for the preparation of
tables and graphs are given in the Publication Manual of the American
Psychological Association and in a number of sources cited there.)
Once the repof~ has been written and the tables and graphs pre-
pared, the author should read it over, asking himself such questions as:
Is this sentence clear? Is it grammatieally correct? Does it say what I
mean? Could the point be expressed more simply? Does the material
given in the tables justify the conclusions I have drawn? Do the various
points fit together logically?
Having at least one colleague read the report just before the final
revision is extremely helpful. Sentences that seem crystal-clear to the
writer may prove quite confusing to other people; a connection that had
seemed self-evident may strike others as a non sequitur. A friendly
critic, by pointing out passages that seem unclear or illogical, and per-
haps suggesting ways of remedying the difficulties, can be an _invaluable
aid in achieving the goal of adequate communica'fiOn. -
13
THE APPLICATION OF
SOCIAL RESEARCH
Concern with A'pplication During the Research
Presentation ot Action-Oriented Research

/
Extending the Area ot Application

Research and Social Policy
Unapplied knowledge is knowledge shorn of its meaning.
A. N. WHITEHEAD
U P TO THIS POINT, we have discussed problems and procedures

common to all social science research, whether it be directed
toward increasing general knowledge about human behavior or toward
the solution of some practical problem. In this chapter, we shall discuss
a problem specific to the latter type of research-that of incre~sing
the likelihood that the findings will be used.
The immediaty applicability of research findings to current social
issues is largely a function of earlier decisions in the research process.
To be sure, inquiries that have not been geared to immediate applica-
tion from the outset may yield results that have practical significance.
In the long run, they may have even greater social usefulness than a
study undt:rtaken with the goal of immedi~te application in mind. It
is possible, for example, that modern theories of learning developed
in the course of laboratory experimentation may one day revolutionize
education, but, if this is to happen, someone must un~ertake the task
of transforming theoretical concepts into realistic and relevant terms
that will mike possible the testing of their validity for concrete educa-
tional situations.
It is largely a question of preference whether a social scientist :works
with the goal of immediate or potential future application in mind, or
whether he is completely indifferent to this issue. But th<:; social
scientist who is concerned with the immediate applicability of his work
must keep this objective in mind at every stage of his study. The selec-
tion of the research problem, the choice.of setting, the extent to which
action personnel are involved, the manner in which the findings are
presented-all may affect the likelihood of application.
456
CONCERN DURING THE RESEARCH 457
Concern with Application During the Research

When an investigator is concerned with application of his work,
planning for it must begin at the outset. One of the first things he must
decide is the extent to which the people who may be expected to use
the results should participate in planning and carrying out the research.
Usually some minimum of cooperation is essential. Research intended
to have immediate usefulness is frequently carried out at the request
of an organization that hopes to solve some problem. Even if the
investigator himself initiates the research, he is likely to want to carry
it out in the setting of a functioning organization. Although a labora-
tory study, or a survey of individuals unconnected with any given
organization, may sometimes be appropriate for research intended to
have action implications, more often such work is carried out within
a factory, an office, a housing project, a school, a community center.
In order to conduct research in such a setting, the investigator must
enlist the cooperation of the people in authority at least to the extent
of securing their permission to go ahead with his plans.
Collaboration with an action agency may sometimes be necessary,
or desirable, even if the investigator is not concerned with the applica-
tion of his findings. Any study design that invplves a functioning
organization demands at least a minimum of collaboration on the
part of the action personneJ.1 Much of the following discussion of
sources of difficulty in such collaboration and of methods of reducing
friction is relevant to any study conducted within an organizational
setting, whether or not there is concern with application of the findings.
How much the action personnel should participate in the research,
beyond the necessarY, minimum, is a matter for decision. Collaboration
with lay people is tiine-consuming and does not always proceed with-
out friction. Nevertheless, from the standpoint of application of re.
search findings, the advantages of collaboration far outweigh its dis~:
advantages. I ~
f
l_By action personnel we mean the members and staff ot the organization
within which the study ltakes place, whether this be a factory, a school, a religious
organization, a labor union, a P.T.A., a boys' club. Similarly, the terms "action
organization and action agency, or simply organization and agency, are intended
to refer to any such group in which research is being conducted.
458 THE APPLICATION OF SOCIAL RESEARCH
There are two major ways in which collaboration with an action

agency increases the likelihood that the research findings will be used.
First, participation usually increases the agency's interest in the re-
search and in its possible usefulness. It is a well-established psychologi_
cal principle that the greater the degree of active participation in an
endeavor, the greater the degree of involvement with its outcome. An
agency that has been in close contact with a research project is more
likely to act on recommendations emerging from the investigation
than one that has had nothing to do with the development of the
inquiry. Second, and equally important, collaboration with an action
agency helps to ensure that the research is really relevant to problems
as they appear in the daily operations of the organization and of others
like it.
CONTRIBUTIONS OF THE ACTION PERSONNEL
Perhaps the most crucial step in determining whether the findings

of a study will have immediate social usefulness is the selection of the
research problem. Obviously, research concerned with the reasons
housewives prefer 'one brand of scouring powder to another, or with
the effects of different methods of supervision on production in a
factory, or with the conditions under which race riots occur, is more
likely to find immediate application than research on the mating be-
havior of chimpanzees. But the selection of a general area which is
recognized as involving a practical problem or a matter for decision
regarding social policy, no matter how urgent, is not in itself a guarantee
of immediate applicability. The study of prejudice, for example, can
lead to change in social practices, but it need not do so. Whet}:J.er or
not it actually does depends largely on the manner in which the research
problem is conceptualized. If the problem is formulated so that the
hypotheses bear on a possible or actual course of action that persons
concerned with the situation envisage or perform, it may well lead to
immediate application. This was demonstrated in the Deutsch-Collins
I housing study ~escribed in Chapter 1. Public housing officials had
indicated that research results concerning the effect of occupancy
pattern on race relations might be taken into account in making deci
sions about future policy; accordingly, the hypotheses formulated were
directed at this question. A study focused on the relation between
personality type and the nature of response to close personal contact
with members of another ethnic group would be less likely to result in
practical implementation. The formulation of research for immediate
use requires a knowledge of the nature of decisions that are likely to
be made by persons concerned in a practical way with the problem
under investigation. This knowledge is best acquired in close collabora-
tion with such persons and agencies in the early stages of an inquiry.
The final statement of the problem in terms that make it a work-
able starting point for research is, of course, up to the social scientist.
He is the expert in the research process, and it is he who must take
responsibility for the adequacy of the research procedure. The practi-
tioner has a different kind of knowledge to contribute. Nothing but
confusion is to be gained by mixing the functions of the two. Each is
an expert, with a specialist's knowledge of the research matter, but
one is concerned with advancing knowledge of a given phenomenon
and the other with improving existing services. Sometimes friction
occurs because of these disparate goals and standards. The practitioner
may well become impatient with the experimenter's controls, and may
wish to put new insights into effect immediately, w~ereas the experi.
menter may become exasperated with the practitioner's desire to enact
changes when he has no solid evidence of their value.
Each partner makes his best contribution when it is highly
specific to the function for which he has been trained. This division
of labor is appropriate throughout the research, no matter how close
the collaboration. The social scientist is the expert on research design,
sampling, construction i;>f measuring instruments. But the practitioner
knows the specific situation. He can be of inestimable help in pointing
out possibilities for "natural" experiments, such as comparing sub-
groups within the agency in which different practices are followed, or
possibilities for setting up experimental manipulations in such a way
that they will appear realistic to the subjects. Suppose, for example,
that the research problem !,
has to do with the effects of grouping high
school students into classes that are relatively homogeneous or hetero-
geneous with respect to scholastic ability. If decisions on this matter
have been left up to the principals of the various schools in the system,
the superintendent of schools may be able to point out to the investi-
gator schools which are similar in such matters as general socio-
economic level of the students, range of scholastic ability, and caliber
of teachers, but which differ in the independent variable with which
the research is concerned-the basis on which students are grouped.
Or, if the existing situation does not contain arrangements that so
neatly fit the research Iequirements, the superintendent may arrange
to have homogeneous classes set up in some schools and heterogeneous
ones in others, or to have some homogeneous and some heterogeneous
classes within the same school.
The agency personnel can inform the investigator whether data
needed for the study are available in records compiled in the course of
the agency's regular operations, or they may be able to suggest ways
in which the collection of data can be worked into on-going pr0cedures.
They can advise the investigator whether his proposed design or
methods of data collection are likely to be feasible. For example, an
investigator planning a study of street-comer, gangs discussed his re-
search plan with a group worker who had had experience with such
gangs. The plan called for administering Rorschach tests to the gang
members. The group worker predicted that it would be impossible to
persuade the boys to respond to the Rorschach, and suggested that
the investigator try to devise ways of getting the information he needed
by means of participant observation. At times, a~tion personnel, view-
ing the situation from their particular vantage point, may overestimate
or underestimate difficulties; therefore the investigator should try to
become sufficiently familiar with the situation to make his own.estimate
of the feasibility of his proposed procedures. Nevertheless, the practi-
tioner may be an invaluable sourc;e of advice concerning resources and
possible obs,tacles.
The action personnel may participate in carrying out the experi-
mental manipulations or in gathering the data for the study. However,
their participation in analysis of the data is likely to be limited, except
in the case of studies set up as self-surveys and requiring only the
simplest kind of analysis. 2
2 For a discussion of self-surveys, see Wormser and Selltiz (1951a and b).
SOME PRACTICAL PROBLEMS
Collaboration does not always run smoothly, nor are the recipients
of research findings always eager to accept and act on them-even
though they may have requested the investigation in the first place.
Even if the organization has requested the study, this request may not
represent a consensus on the part of all the relevant people or interest
groups within the organization. One individual, or a few, in a central
positipn, may have made the decision. There may be others who are
not in favor of the research, and these may include people whose help
is needed in carrying out the study or who will have some responsibility
for applying the findings. Some of them may believe that no problem
exists; others may recognize a problem but think that research cannot
contribute to its solution; others may feel that the nature of the people
or groups sponsoring the research is such that the study will inevitably
lead to conclusions detrimental to their interests; still others may be
sure that they know the solution and so consider research unnecessary.
For example, the principal and board of trustees of a private school
became concerned about the school's apparent lack of effect on the
character of its students. They asked a research organization to under-
take an exploratory study to evaluate whether there was, in fact, a
discrepancy between the goals of the school and the putcomes in terms
of the students' character and, if so, to identify possible reasons and
suggest courses of action that might make achievement of the goals
more likely. The plan of the study called for interviews with faculty,
parents, and students. When the interviews with faculty members
were started, it rather quickly became apparent that the faculty were
antagonistic to the study. They saw the school's accomplishments as
more nearly in keepipg with its goals than did the principal and
trustees. Moreover, to; the extent that they recognized the results as
falling short of what )Vas desired, they held that the main source of
the difficulty was their own low pay and heavy work schedule. In their
opinion, the money spent for research might better have been used to
increase their salaries. Several meetings 'With the entire group and with
influential individualsi~ere necessary to overcome their opposition.
In other situations, the relevant people may recognize a problem
and believe that rese~rch can contribute to its solution, but some of
them may feel that the specific experimental manipulations or other

research procedures are in conflict with the goals of the organization
or with its best functioning. Such a situation has been described and
analyzed in considerable detail by Schwartz (1948). Two agencies
cooperated in this study. One was engaged in the practice of group
work on an interracial basis in a neighborhood child-recreation center;
the other was established for the purpose of research in the area of
interracial relations. On the basis of an "experience survey" (see
Chapter 3) of workers in such neighborhood centers, the research
agency formulated the following problem: What type of leadership
behavior in an interracial group produces more favorable changes in
the attitudes of children toward those of another skin color? Specifi-
cally, two practices were to De compared: (1) pointing out to the chil-
dren at every opportunity that they were actually enjoying an experi-
ence in an interracial group, and (2) refraining consistently from
mentioning the group composition. Another variable to be tested was
whether emphasis on group achievement or on individual achievement
was more conducive to good intergroup attitudes.
The research group, having formulated this problem, asked the
director of the neighborhood center whether he and his staff would be
interested in collaborating in such a study. The staff wanted to produce
more favorable intergroup attitudes and believed that research could
help them do a better job in this respect. They appeared to accept the
formulation of the specific problem and the research design, and
agreed to help carry it out.
The relationship between the two agencies in carrying out the
study was very close. The maintenance of the experimental design-
i.e., the following of s_trictly defined alternative practices in the des-
ignated children's groups-was completely dependent on the group
workers themselves. However, it came to light several months after
the initiation of the experiment that the group workers who had the
assignment of emphasizing individual achievement were not doing
so, because they regarded this as contradictory to their own professional
standards.
This parti,:ular collaborative situation illustrated still another
source of resistance. Quite apart from their feeling that one of the
experimental variables required behavior out of keeping with their
own standards, the group workers found other aspects of the research
procedure a burden and an interference with their major task. The
plans for data collection included use of a pictorial attitude test, the
arrangement of test situations, and the recording of observations. The
already heavy work load of the group workers was inpeased by demands
that they record what they observed during each session; the introduc-
tion of test situations at periodic intervals interfered with their own
spontaneity and that of their groups. Even though the research was
designed to make group work more effective in the long run, situations
arOse where it seemed necessary for the group worker to sacrifice the
research for the sake of carrying out his function in the group. It is
also possible, of course, that the natural difficulty of deviating from
habitual practices, rather than the requirements of the situation, inter-
:ered with the effort to introduce and study new methods.
Lack of understanding of the relevance of the research to solution
of the problem at hand is another source of reluctance to cooperate in
a study or to apply its findings. A large government agency once en-
listed the help of a research team to conduct a content analysis of the
output of its writers, in order to provide a basis for evaluating whether
the material really conveyed the messages intended. In planning and
carrying out the study, the social scientists worked only with the
administrators of the program; the writers thems~lves were brought
into the picture only when the report of the study had been completed.
Many of the writers took exception to the study, especially to the
method of analysis that had been used. Prolonged discussion and
negotiations were required before they were able to see the value of
the research.
This example suggests still another source of resistance to research:
apprehension about the use to which findings may be put. Individuals
!
may see the research las involving an evaluation of how well they are
performing their functions; they may fear that possible negative evalua-
tion will be used agaipst them. In a work situation, for example, they
may worry that evid~nce collected in the course of the research will
be used as a basis for determining promotions or dismissals. Even
where no such obj~ctive consequences are likely, people may be un-
t-
comfortable at feeling themselves "under the microscope." The chair-
man of a P.T.A. in ~hich attendance at meetings has been declining
may ask a social scientist for help in finding out the reasons for the
decline and possible ways of reviving interest; but even though she has
asked for the help research can give, she may be apprehensive that
the findings will point to inadequacies in her performance as chair-
man.
One final source of resistance is especially likely to come into
play at the stage of applying the findings: the reluctance to change
accustomed ways of doing thinGs. Application of research findings
often implies doing something differently, in a way that will pre
sumably be more effective. But ordinarily it is easier to keep on doing
things as we have been doing them; inertia may lead to rejection of
the research findings or of their implications for action. A dramatic
example from medical research illustrates both this source of resistance
and the one discussed immediately above-resentment of criticism.
Beveridge (1950) recal1s the fate of Semmelweis after he had dis-
covered the origin or puerperal rever:
. . . he instituted a strict routine of washing the hands . . . be-
fore the examination of the patients. As a result of this procedure,
the mortality from puerperal fever in the first obstetric clinic
of the General Hospital of Vienna fell immediately from twelve
per cent to three per cent, and later almost to one per cent. His
doctrine was well received in some quarters and taken up in some
hospitals, but such revolutionary ideas, incriminating the obste-
tricians as the carriers of death, roused opposition from en-
trenched authority and th<:; renewal of his position as assistant was
refused. -
METHODS OF REDUCING RESISTANCE TO RESEARca3
The reader has probably noticed that in several of the examples

above, opposition arose from persons who had not been consulted in
the course of planning the research. This fact points to a.basic principle
in achieving successful collaboration with an action agency. Before a
study gets under way, it should be discussed fully with all those who
will be affected by the research, either through helping to carry it out
3 For a more detailed discussion. see Likert and Lippitt (1953).
or through applying the findings. 4 This process is usually started with
top administrators, whose support is ordinarily essential to carrying
out research within an organization. However, cooperation on the part
of other personnel can seldom be assured simply by orders from above,
especially if the study makes heavy demands on them or extends over
a long period of time. Therefore, the discussions should be extended
to include all those who will be asked to help in the study or in apply-
ing its findings.
The more active the cooperation needed, the more essential it
is that everyone concerned understand the nature and purpose of the
research and be reasonably sympathetic toward it. It may be possible
to eliminate reservations and objections through open discussion before
the study is undertaken. If not, and if the objections corne from a
considerable number of people or from individuals who are in key
positions to help or hinder the study, it may be better to abandon the
plan rather than run the risk of finding the study blocked after it gets
under way. In discussing the reservations and objections, the investi-
gator must make it clear that he is willing to consider them seriously;
he should not attempt simply to "talk down" the opposition. Fre-
quently, arguments in favor of the research can be made more effec-
tively by other members of the action agency than by the investigator
himself. In any case, he must beware of promising, or of letting others
promise, more than the study can realistically be expected to perform.
Excessive promises are likely to boomerang.
Discussions with action personnel should not only give a clear
picture of the nature of the research problem and the general proce-
dures to be used; they should also state precisely what will be required
in the way of collaboration. This statement should include not only a
description of the Kinds of things that may have to be done and how
much time these tasks are likely to take, but an account of how the
4 Sometimes, of course, the nature of the research problem is such that knowl-
edge by the subjects ofl the specific details, or even of the fact that they are being
studied, may affect the results. If the invesigator believes that this is the case
in his study, he may make a conscious decision not to acquaint them with the
research plan. However, it is often possible to discuss the research with the relevant
personnel and still keep the actual subjects unaware of it. For example, in a study
of leadership practices in a recreation center, the group workers can be given a
full description of the study even though the children in their groups do not
know that they are being observed.
research procedures may interfere with normal operations, to the

extent that this can be foreseen. It may be possible to work out ways
of making the research operation less of a burden on the action
personnel; but research can seldom be completely unobtrusive, and
this fact should be made clear. If the agency personnel are to be asked
to undertake time-consuming duties in the conduct of the research,
arrangements should be made with top 1dministrators to release them
from some of their other tasks. If the research design calls for some of
the action personnel to behave in ways that are unusual or uncongenial,
it may be preferable to hire persons who can be especially trained for
the study. For example, in a study of the relative effectiveness of praise
and criticism on scholastic achievement, many teachers who were
sympathetic to the study might nevertheless be unwilling or unable
to carry out the "criticizing" role. An experienced teacher with training
in research, hired especially for the purpose of carrying out this role,
might have less conflict about it because of her greater identification
with the study .
. ~T'he" threatening aspect of research can often be reduced by
stressing the aim of finding better ways of accomplishing the organiza-
tion'~. g6als-that is, by emphasizing the potential findings as a source
of help rather than criticism. It should be made clear that the purpose
of the study is not the evaluation of individuals, and that the findings
will have no bearing on anyone's prospects for continued employment,
promotion, etc. In this connection, it is appropriate to stress that what
any given individual says will be kept in strict confidence by ~he research
staff, and that no information will be reported in such a way that
individuals can be identified. Such assurances should, of course, be
given only if they can be rigidly adhered to.
Presentation ot Action-Oriented Research
The way in which the research findings are presented, as well as

t~e extent to whicp the action organization has participated in the
research, helps determine whether the findings will be used. A special
problem is created by the fact that there is little opportunity for col-
la~oration during the stage of analysis. In the period between the end
PRESENTATION OF ACTION-ORIENTED RESEARCH 467
of data collection and the completion of the report, interest may
wane.
INTERIM REPORTS
When data collection has been completed, many activities in

which the sponsoring or collaborating agencies played a part come to
a standstill. After the exciting events that so frequently occur during
field work begins a period that, although it may be as eventful for the
social scientist as any of the preceding stages of research, has little
to offer his partner in the enterprise. This is inevitable. And it is also
inevitable that this period should appear long to those who do not
participate in the procedures of analysis and report-writing. What is
more, their impatience is on occasion objectively justified when a
practical solution of the problem under investigation cannot be post-
poned beyond a certain date. From this point of view, the selection
of such a problem for research is, of course, always a risk.
In 1948, for example, when Israel became an independent state,
an organization in the United States which was concerned with the
maintenance of civil rights for the Jewish population of this country
considered the possibility that this event might call for changes in its
program. The organization wished to base its decision on a scientific
/
inquiry into the effect of the foundation of the new state on the
attitudes of non-Jewish Americans toward American Jews. Accordingly,
a large-scale study was initiated. But the collection and analysis of data
took a long time. In the meantime, the world did not stand still.
Israelliuge1y disappeared from the headlines; public attention became
more and more absorbed in other matters. The agency, actively
engaged in its large-scale; activities and consulted daily about many
problems of Jewish life, developed
I
its strategy successfully without the
benefit of the research. The study made an important contribution to
research methodology ana produced some interesting results, which
were relevant to policy qecisions. Yet from the point of view of its
original purpose-to facilitate a decision on an acute problem of policy
-the inquiry failed. The results came too late.
Even when the natdre of a problem does not impose a time limit
on application, the delay between data collection and presentation of
findings may reduce the likelihood of application. This happens either

because new problems arising within the agency absorb the energy
and interest of the staff, or because the agency fails to understand
the reasons for the delay. What can be done during this period to
increase the probability that the findings will be applied?
One of the most effective ways to reduce, if not eliminate,
tensions arising during this waiting period is to present an interim
report. Such a report may contain either?the first results of the over-all
analysis or the final outcome of the analysis of some part of the
investigation. It should be remembered that an interim report, however
strongly its preliminary character is emphasized, is usually understood
as a definite commitment. In this sense, it involve certain risks. If it is
produced at too early a stage of analysis, the social scientist may, when
he writes a final report, find himself in the position of having to change
statements he has made-a situation that understandably reduc!!s the
collaborator's confidence. By and large, therefore, it is advisable to
present interim reports, not on the full scope of a study, but on one
completely analyzed section at a time. This possibility should be taken
into account in planning the analysis, and those sections analyzed first
that are most appropriate for separate presentation.
Whateve{ fqrm the interim report takes, it fulfills a number of
important fmlctlons, an of which help to increase the likelihood that
the findings of the study will actually be applied. If the agency has to
take action without waiting for the full report, the interim report may
contain some findings relevant to such action. Its delivery ,may help to
keep alive the agency's interest in the study and prevent misunder-
standings about delays.
In addition, the interim report helps the agency to maintain realis-
tic expectations with respect t,o social research. Frequently the level of
expectation of what social science can do is unrealistically high. A
collaborative research enterprise often arouses considerable enthusiasm
/ in those who have not previously had close contact with social science.
This enthusiasm can easily lead to the belief that now, at last, all '
questions and problems will be answered. A sample analysis presented
as an interi~ report reduces such expectations to a more!realistic leve1.
Furthermore, the interim report serves to spead: over a longer
period the time-consuming process of discussing research findings and
PRESENTATION OF ACTION-ORIENTED RESEARCH 469
their implications. The amount of time needed thoroughly to review
a large-scale report is sometimes so great as to be forbidding, and may
encourage a superficial examination which is detrimental to successful
application.
It is, of course, not only the volume of a report that prolongs
discussion and puts off application. Sometimes the results provoke
resistance, and much time is needed to adjust to the findings and
accept them. In extreme cases of unwillingness to accept results, the
interim report will be of little help; in less extreme cases, it will provide
time and occasion for a gradual readjustment to new ideas while other
parts of the analysis are still in progress.
Finally, the interim report serves as a test of whether the social
scientist has found the proper level for his communication; for instance,
whether there is enough illustrative material. Difficulties spotted at
this point may be remedied in the final report.
The preparation of an interim report and the ensuing conferences
are, of course, time-consuming for the research staff; often they have
not been anticipated in the budget. Yet the contribution they can
make toward increasing the chances of application would seem to
outweigh all other considerations.
/
THE FINAL REPORT TO THE COLLABORATING ORGANIZATION
In preparing both the interim reports and the final report to the
collaborating organization or to others who are expected to act On the
findings, the investigator must again remember that the main purpose
of the report is to communicate to the audience. And this audience
is quite different from the social scientists for whom technical reports
are written. As a rule, the collaborators or the other persons who will
apply the findings cOi_lstitute a lay audience as far as the theory and
methods of social science are concerned, but in the area of their
activities they are experts, and their concrete knowledge of the specific
situation being studie4 usually exceeds that of the social scientist In
reporting to them, thcrinvestigator must take into account the areas in
which they are experts and those in which they are not.
A report to this audience usually starts with a formulation of..-the
,/
470 THE APPLICATION OF SOCIAL l\ESEARCH
research problem in which the action aspects of the problem are

emphasized. If formulations of the issues have been jointly arrived at
in earlier collaborative efforts, these are presented in full. The formula-
tion of the problem is best followed by a summary of the highlights of
the findings; then by a detailed account of each of the findings. In
addition, a special section on the implication of the results for concrete
decisions is often included. Theoretical implications of the study, if
there are any, are presented briefly, in such a way as to give the lay
reader a sense of its general significance. The use of scientific terminol-
ogy in such reports is generally inadvisable. Where such terms have
to be introduced, they must be precisely and clearly defined. If statisti-
cal data are included, care should be taken to present them in such
a way that they can be understood by an audience with little experience
in dealing with such materiaJ.5 In the written report, the research de-
sign and methods are usually described in an appendix, and tables other
tha,n those essential to an understanding of the findings are usually
placed there also.
As far as the main body of a report to adion personnel is con-
cerned, however, the social scientist must remember that he is address-
ing experts. This audience should receive a comprehensive and detailed
presentation of findings, qualifications and limitations of results, and
possible alternative interpretations. In preparing and presenting a
report for such an audience, the social scientist must be fully aware
of the responsibility he carries in i~fluencing social practices through
his work. I
In addition to submitting a written report, the social scientist

often presents his findings orally to the collaborating agency. Many of
the principles an,d procedures suggested in connection with reducing
resistance during the course of the research ~PBly also to the presenta-
tion of results. Ordinarily, the findings are reported first to the top
officers of the organization, by the investigator himself. The findings
may be reported to the other persons concerned either in a large meet-
ing or in small groups, by the investigator himself or by members of
th~ collaborating organization. Presentation to small groups provides
5 For a detailed discussion of problems arising in the presentation of statisticai
data, see Zeisel (1957;'
PRESENTATION OF ACTION-ORIENTED ReSEARCH 471
greater opportunity for discussion and for consideration of action
implications for a specific unit or procedure. 6
Successful application is facilitated if the results are presented to
all those who will be expected to apply them, and perhaps also to those
who will be affected by the changes. They should have a chance to ask
questions, to make sure that they understand the findings. They
should be encouraged to discuss the findings, to suggest alternative
interpretations and additional possible applications. In fact, in pre-
liminary reports the investigator may present only the bare findings,
encouraging the action personnel to suggest interpretations and applica-
tions.
The threatening aspects of the findings may be reduced by
emphasizing first those things the study shows are going well. Suppose
that a survey among the students of a college shows that they are, on
the whole, satisfied with the quality of teaching, the content of courses,
and the library facilities, but that they have complaints about the living
arrangements, the lack of help in securing jobs after graduation, and the
scarcity of opportunities for meeting and talking with the faculty
outside of class. In a report to the faculty, presenting first those aspects
with which students are satisfied may reassure the faculty that the
research has not been focused only on shortcomings. The explicit recog-
nition that in many respects the college is doing a good job may make
them less sensitive to the criticism of other aspects and more willing
to consider whether the criticisms are justified and whether the un-
satisfactory situations can be changed. Sensitivity to implied criticism
may be reduced by focusing discussion on methods of remedying the
shortcomings revealed by the research rather than dwelling on the
evidence of their existence. However, it is sometimes necessary to stress
such evidence in order to create an awareness that there are problems
that need to be solved. .
OTHER AUDIENCES FOR ACTION-ORIENTED RESEARCH

I
In addition to those' who may make decisions and take action on
the basis of the findings, there are two other audiences for action-
L
. Likert and Lippitt (19:53) give a detailed account of a presentation following
II
thiS sequence.
oriented research: social scientists and the general public. Preparation

of reports for a scientific audience has been discussed in the preceding
chapter; the presentation of action-oriented research to this audience
docs not introduce any new considerations.
Presentation of social research findings to the general public in-
volves different problems altogether. Presentation to the lay public is
rarely undertaken by the social scientist himself. As a rule, the mass
media-press, radio, television, and motion pictures-employ specialists
in the art of popularization. These middlemen fulfill an important
function in the education of the public. Since they are usually respon-
sible to those in charge of the mass communication media rather than
to the social scientists who are their source of information, however,
the success or failure of these men is often judged in terms alien to the
concerns of science. The accuracy of their presentat~on is sometimes
considered of secondary importance. Omitting qualifications of results,
ovelsimplifi.cation, and, on the whole, gIeateI daims {or social science
than are justified by the research are frequent by-products of this situa-
tion.
This state of affairs is particularly dangerous in the presentation of
action-oriented research. The only way to deal with it satisfactorily lies
in the hands of the social scientist. If he gives his consent to a popular
presentation of his work, he should arrange to see the product before
it is released for popular consumption. His main task in checking the
presentation will usually be the introduction of appropriate qualifica-
tions about the generalizability of his findings.
Extending the Area of Application

So far we have discussed ~he Imanner in which the social scientist
can encour~ge the application of his results by agencies that have col-
laporated in the research process. But if the collaborating agency is
small, application solely by that agency means that use of the findings
will be confined to a small group. The social scientist may want to
concern himself with the possibility of extending the appli<:;alion of
his findings bey~nd the situation in which they were obtained.
The problems in attempting to broaden the sphere of application
EXTENDING THE AREA OF APPLICATION 473
beyond the collaborating agency are, in part, problems of appropriate
communication with other potential users of research findings; in part
they are related to the scientific problem of the justification for gen-
eralizing from one situation to another. We have discussed the question
of generalization of research findings in Chapter 11 and shall consider
it in another context in Chapter 14. In this chapter we shall examine
some occasions where this becomes a practical problem.
The communication of research findings to other interested
agencies can be carried out either by the staff of the agency collaborat-
ing in the original study or by the social scientist himself. Social
workers, housing officials, union leaders, production managers, per-
sonnel of community organizations, and any other collaborators who
may have been involved in a research project usually maintain organized
and personal contacts with others in their professional fields. If their
participation in the research project has been at all active, it is likely
that they will spontaneously communicate their experience and
stimulate others in their field to apply the findings of the study. For
example, the Deutsch-Collins study of occupancy pattern in housing
projects was widely discussed among housing officials and may have
influenced policy formulation by housing authorities which had no
firsthand contact with the inquiry. Wider applicatjon of the findings
was also made possible by the publication of a series of articles in a
housing journal. At least in the United States, virtually all professional
groups who are likely to cooperate with social scientists publish periodi-
cals. Such publications, in which problems of the profession are dis-
cussed, are a suitable medium through which the knowledge of
relevant reSearch can be spread and further application encouraged.
Not infrequently~ the social scientist is asked to advise upon the
application of his research in a new situation. In such cases, the
question of whether ~hat he knows can justifiably be transferred to
a situation he has not: investigated becomes a major one. In applying
results beyond the original setting of the research, the social scientist
shifts his role from that of a research worker to that of a consultant. The
two roles differ in several ways. They involve different skills, activities,
motivations, rewards, and responsibilities. If a research scientist tempo-
rarily assumes the role of a consultant, his performance in the new
role will be better if the shift is made deliberately and of his own
volition, rather than accidentally or in response to demands and pres-
sures. It is not within the scope of this text to offer specific suggestions
for the social scientist in his consultant role. 7 However, one cardinal
rule can be presented: He should not make recommendations prior to
full discussion of the features peculiar to the new situation. In con-
ferences with the persons who have sought his advice, he will usually
find it most helpful to recount the principal characteristics of his
investigation and to ask in return for the essential elements of the
." problem being.brought to him. Such conferences bring into the open
both the potentialities and the i?evitable risks involved in application
of the research results to a new situation. In this process, the agency
under whose auspices the research findings are to be applied becomes
a full partner to the venture, and the reponsibilities for the outcome
of the application are shared by both agency and consultant. It is the
consultant's task to specify the conditions under which his original
findings obtained; it is the agency's task to examine with him differ-
ences in conditions in the new situation. To weigh the importance of
such differences is the combined task of both.
Research ana Social Policy8
Both as a consultant and as a researcher, the social scientist may

see his work becoine an influence in' th~ d~termination of social policy.
But he must be prepared for the fact that there are nonscientific con-
siderations that enter legitimately into the decision-making process and
interfere with or promote the application of his results. Much of hi1>
effort will be wasted unless he is prepared for ~his and learns to identify
the realistic possibilities for application before becoming too deeply
involved in encouraging it.
To take an example: An administrator had supported an investiga-
tion designed to diagnose a situation so that appropriate action could
7For a discussion of problems of consultation, see Likert and Lippitt (1953).

8For a more extended discussion of the relation between research and social
Dolicy. see Hyman (1955).
RESEARCH AND SOCIAL POLICY 475
be taken. He sponsored the study in spite of some opposition from
other persons in his department. At the time the study was completed,
the administrator happened to be engaged in a general reorganization
of his department, which aroused the opposition of the same members
who had been hostile to the study. Although he was convinced of the
value of the study, he realized that fighting two battles simultaneously
might interfere with the proper functioning of his department. Since,
at the time, he regarded the reorganization as more urgent than the
action indicated by the study, he shelved the results until a more op-
portune moment.
Unforeseeable circumstances that interefere with application can
always arise. But frequently it is unforeseen rather than unforeseeable
circumstances that present such obstacles. The position of an agency in
the community, its manner of arriving at policy decisions, and its ability
to carry through program activities may all have an important bearing
on the extent to which the findings of research will be applied. The
social scientist can frequently make an appraisal of such matters before
entering into a collaborative arrangement with the agency. If he is
concerned with the immediate application of his findings, he will be
well advised to do so.
This is not the place to discuss in detail the process of decision-
making in social policy, a process about which all too'little is known.
One element in this process, however, must be m~ntioned because it
has on occasion given rise to ~ontroversies among social scientis_ts about
the justification for encouraging application of results. It is sometimes
argued that, since decisions on social policy involve social values as
well as factual considerations, the social scientist should not attempt to
participate in or to infi,uence such decisions. Without doubt social
values enter into the making of decisions about the application of re-
search. It is a value question, for example, whether community efforts
are to be directed toward eliminating or intensifying discrimination
against minority groups. Is the social scientist, by virtue of his status as a
scientist, compelled to i~nore such values, or can he take sides? Leigh-
ton (1949) has answered the question thus:
Within an are/marked off for scientific investigation, the
values of science reign supreme over each step in the process to-
14
RESEARCH AND THEORY
The Function of Theory

Theory as a Basis for Research
/
The Contribution of Research to the Development of Theory

Interrelation of Theory and Research-A Summary
PROPERTY OF I
The Kansas S.ate University of i
Agriculture & Applied Science I
"-_ _ TC M India. _j
Those who refuse to go beyond facts rarely get as far as facts.
T.H.HUXLEY
The Function of Theory
T HROUGHOUT this book we have pointed to the limitations in-

herent in the results of a single study and the urgent need for
syst.ematic knowledge founded on a broader base. Without such
broader knowledge, the insights of social science will necessarily be
limited to the specific settings and problems in which investigations
have been conducted.
From this point of view, the development of scientific laws and
scientific theory has a very practical function. A scientific law is a
summary of available knowledge of the relationship between properties
in more general terms than the emptrical findings on which it is based.
Scientific laws are hypotheses considered to be true.
Braithwaite (1955) defines a theory as consisting of a:
. . . set of hypotheses which form a deductive system; that is,
which is arranged in such a way that from some of the hypotheses
as premisses all the other hypotheses logically follow. The pro-
positions in a deductive system may be considered as being
arranged in an order of levels, the hypotheses at the highest level
being those which occur onl'y as premisses in the system, those
at the lowest level being those which occur only as conclusions
in the system, and those at intermediate levels being those which
occur as conclusions of deductions from higher-level hypotheses
and which serve as premisses for deductions to lower-level hy-
potheses. ,
It is important to distinguish the modern scientific usage of the
word theory from other meanings the word may have. In common
480
THE FUNCTION OF THEORY 481
parlance, theory is frequently identified with speculation; what is
"theoretical" is unrealistic, visionary. Although it is true that in the
early days of a science theories are often the result of armchair specula-
tion and may have meager and weak support in empirical data, theory
and observation become more and more closely connected as a science
develops. In the present state of the social sciences, research and
theory are not always closely linked, and theories are likely to contain
speculative elements that go far beyond the evidence of available data.
Thus, for example, some aspects of psychoanalytic theory have little
confirmation in empirical investigation. Yet it has proved extremely
useful as a basis for clinical work and as a source of new perspectives
in considering human behavior. In general, however, the intention of a
theory in modern science is to summarize existing knowledge, to pro-
vide an explanation for observed events and relationships, and to predict
the occurrence of as yet unobserved events and relationships on the
basis of the explanatory principles embodied in the theory.
One other characteristic of theories in modern science should be
noted: this is their provisional character. In earlier times, a theory
was considered a final explanation. Today, a theory is always held with
some tentativeness, no matter how great the accumulation of findings
consistent with it. It is considered as the most probable or most
efficient way of accounting for those findings in}he light of present
knowledge, but it is always open to revision. It is not a static or a final
formulation.
AN ILLUSTRATION OF THEORETICAL EXPLANATION
In the third century A.D., Tertullian observed a curious and ap-

parently regular concurrence of events: "If the Tiber overflows into
the city, if the Nile does not flow into the countryside, if the heavens
remain unmoved, if.the earth quakes, if there is famine and pestilence,
at once the cry goes up: 'To the lions with the Christians.' "
Seventeen hundred
, years later, several social scientists (Raper,
1933; Hovland and Sears, 1940) reported another observation: If the
per-acre value of cotton in the southeastern section of the United States
is low, the number df lynchings of Negroes in that area is high.
Let us assume that the accuracy of both observations is beyond
482 RESEARCH AND THEORY
1
doubt. The underlying similarity in the two statements-catastrophf..
leads to persecution-is all the more striking because they were made
so far apart in time and referred to entirely different people and
events. What is the explanation for this sequence of events?
According to Gibbon, the Roman populace had an explanation
ready at hand: "The superstitious Pagans were convinced that the
crimes and impiety of the Christians, who were spared by the exces-
sive lenity of the government, had at length provoked Divine justice.'~
Since the Christians were thus responsible for the fury of the gods~
their death would "naturally" placate the gods, who would then order
the Tiber back into its banks.
Let us consider whether this explanation accounts for the later
event. Allowing for the change in superstitions throughout the cen
turies, the argument would run as follows: The Negroes have caused
the fall of cotton prices; if they are killed, the cotton price will rise.
Both explanations hinge upon the assumption that the Christians
and the Negroes were guilty of a crime for which they deserved punish-
ment. However, there are certain differences in the two sequences
which make the explanation unsatisfactory for the second. Although we
do not share the superstitions of the third century, we must admit that
the behavior of the pagans had, at least, internal consistency; they acted
on the assumption that punishment of the Christians would stop the
pestilence. Lynch mobs, on the other hand, do not relate their actions
to the cotton price; rather, they justify them in terms of an alleged or.
actual rape or a threatened or factual breach of the caste system:
Cotton prices do not figure as a conscious motive. Nevertheless, the
correlation exists.2
When two observations of a similar structure cannot be under-
stood in terms of a given eXl?lanation, the social scientil!t is .forced to
one of two alternatives: either to demonstrate that the similarity was
apparent rather than real, or to search for another explanation that
does apply to both events. An explanatory theory is available that can
,
1 Actually, this is not quite the case. The second has been challenged by Mintz
(1946). .
2As poirited out in Chapters 4 and 11, the existence of a correlation between
two variables does not constitute proof that they are causally related. However, it
does make reasonable a search for an explanation of why the two factors vary
together.
account for both our examples. It involves the concepts of frustration,
aggression, inhibition, and displacement. 3
Like all concepts, these cannot be directly observed. They may,
however, be inferred from events that can be observed. Frustration is
defined by Dollard. et al. (1939) as "that condition which exists when
a goal-response suffers interference." Aggression refers to a class of acts
that are designed to injure someone or something. Inhibition refers to
the tendency to restrain acts because of the negative consequences one
anticipates from engaging in them. Displacement refers to the tendency
to engage in acts of aggression that are directed not against the source
of frustration but against another target. These concepts are related to
one another by a system of interlocking hypotheses called "the frustra-
tion-aggression theory" (see Dollard et al., 1939).
The major hypotheses of the theory are:
1. The amount of frustration is a function of three factors:
(a) The strength of instigation to the frustrated goal-re-
sponse;
(b) The degree of interference with the frustrated goal-
response; and
(c) The number of goal-response sequences frustrated.
2. The strength of instigation to aggression varies directly with
the amount of frustration. /
3. The strongest instigation aroused by a frustration is to acts
of aggression directed against the agent perceived to be the
source of the frustration; progressively weaker instigations are
aroused to progressively less direct acts of aggression.
4. The inhibition of any act of aggression varies directly with the
strength of the punishment anticipated for its expression.
Punishme* includes injury to loved objects and being pre-
vented fro~ carrying out a desired act as well as the usual
situations that
I
cause pain.
5. The inhibition of direct acts of aggression is an additional
frustration I that instigates aggression against the agent per-
3 For a discusslol). of the nature of concepts, see Chapter 2, pages 41-42. For a
more detailed discussion of the concepts of frustration, aggression, inhibition, and
displacement, and a Ij10re detailed presentation of the frustration-aggression theory,
see Dollard et a1. (1939). Our presentation of the frustrationaggression theory
differs from theirs inia number of respects.
ceived to be responsible for this inhibition and increases the

instigation to other forms of aggression. There is, consequently,
a strong tendency for inhibited aggression to be displaced to
different objects and expressed in modified forms.
6. The expression of any act of aggression is a catharsis that re-
duces the instigation to all other acts of aggression.
These interrelated hypotheses help to explain many diverse
phenomena, all of which appear to be reflections of the same under-
lying process-namely, that when a person is frustrated and is inhibited
from expressing his aggression directly toward the perceived source of
frustration, he will displace his aggression. 4 According to this theory,
the aggression against Christians by the Romans after the occurrence
of natural catastrophes and the increase in lynchings of Negroes by
southern whites after the fall of cotton prices both are explainable in
the same terms. Frustration, produced by natural catastrophe' or by
fall in cotton prices (and, hence, in income) results in aggression,
but the futility or fear of being directly aggressive against the gods or
against society inhibits the direct expression of the aggression and
results in its being displaced onto groups whose low social status
prevents them from retaliating and hence from also inhibiting the
aggression.
Thus, with the help of the concepts of frustration, aggression,
inhibition, and displacement, an explanation can be offered not only
for the observations about lynching and for the occurrences noted by
TertuHian but also for many other events, such as t~e aggressive action
of a child against ,a younger sibling following his frustration by his
parents. This fulfills one requirement of theoretical explanation: It
reduces a number of different phenomena to underlying general prin-
ciples. The more diverse observations a theory can explain, the greater
confidence we can have in usi~g the general principles it embodies for
the purpose of prediction.
4 We hope the reader will understand that in presenting the frustration-aggres
si9n theory we do not mean to imply that it should be viewed uncritically. We
have presented it for illustrative purposes. As a psychological theory, it has been
subjected to a ~ood deal of criticism (see Lewin et al., 1944). Som~ of the major
criticisms have centered upon the assumptions that frustration always instigates
aggressive acts, that displacement is in a sense blind, and that expression of aggres
sion is necessarily cathartic. Despite its limitations, the frustration-aggression theory
was an important landmark in psychology.
RELATION BETWEEN OBSERVATION AND THEORY
Hempel (1952) has likened a scientific theory to a network, in

which the terms or concepts are represented by the knots, and the
definitions and hypotheses by the threads connecting the knots: 5
The whole system floats, as it were, above the plane of observation
and is anchored to it by rules of interpretation. These might be
viewed as strings which are not part of the network but link
certain points of the latter with specific places in the plane of
observation. By virtue of these interpretive connections, the net-
work can function as a scientific theory: From certain observa-
tional data, we may ascend, via an interpretive string, to some
point in the theoretical network, thence proceed, via definitions
and hypotheses, to other points, from which another interpretive
string permits a descent to the plane of observation.
The advantages of this two-way communication between the
plane of observable phenomena and the plane of theory are obvious,
provided we follow the rules governing traffic on either plane and
between them. Awareness of this particular theoretical formulation
of the relationship between frustration and aggression will suggest
directions for the study of situations involving frustration. It explains
otherwise puzzling situations. Applied to social problems of discrimina-
tion and persecution, it suggests that the attack on the actual cause of
5 The reader may be puzzled by one feature of Hempel's model, the representa-
tion of definitions as well as of hypotheses by threads. Is not the definition merely a
characterization of a knot, a specification of what a particular knot stands for?
In one sense this is, of course, true. But it is easy to miss the fact that every defini-
tion that serves a scientific purpose conceals within itself a theoretical proposition-
namely, that the things, processes, or events that are encompassed by the definition
belong together in the sense that treating them as mutually equivalent under the
definition facilitates the discovery of lawfnl relationships and the organization of
data. We could, for example, define fish as creatures that live in water. By this
definition, both whales; and plankton would be fish. But scientists have found it
more helpful to classifY whales as mamm:lls and to put plankton in still another
category. The precedipg definition of fish obscures- important relationships.
Similarly, if we were to define Negro in terms of degree of pigmentation of the
skin, we would be putting together into one category groups of individuals who
might better be distinguished and excluding others who might better be counted
together with some of those we would have included under the definition (see
K1ineberg, 1935b). Is a. given definition "good" or "correct"? That depends entirely
on what it leads to. At issue is not the word, but the proposition that certain
things are equivalent filr certain purposes and that for these purposes those things
that are excluded by the definition are not equivalent to the things that are
included.
frustration may be a more successful means of preventing persecution

than steps aimed directly at the persecution itself. It offers an alterna-
tive for the racist theory of discrimination, which states that the per-
secution of an out-group is a result of the out-group's attributes rather
than of the needs of the in-group-a theory not dissimilar to the pop-
ular explanation which satisfied the Roman pagans.
However, as Zawadzki (1948) has pointed out, some observations
of prejudice and discrimination raise questions for which the theory
has no immediate answer, and thus reveal shortcomings in it as an
explanatory principle. Why is aggression displaced most frequently
onto the Negro in the American scene, where there are many other
minorities that might serve the same function? Why do some people,
such as the Comanche Indians whom Kardiner described, engage in
aggression against their neighbors even though none of the ordinary
frustrations are discernible either in their patterns of child-rearing or
in their economic situation? What about the evidence presented"by
Freud, that frustration may lead to constructive activity as well as t!)
aggression?
None of these questions is answered satisfactorily by the frustra-
tion-aggression theory without putting a strait jacket on facts or dilut-
ing the definitions of terms and theoretical postulates to such an
extent that they lose their substance. In other words, the theory in its
present form is inadequate because it cannot account for the variety of
relevant phenomena.
This does not mean that the frustration-aggression theory is wrong_
It does mean that it is insufficient, not specific enough and not
adequate to cover all of the relevant observable phenomena, although
it has done good service _for a limited number. The limiting conditi~ns
under which it applies, however, have not yet been defined,6 and this
interferes seriously with its predictive value.
A survey of the state of theories in"the social sciences will reveal
that there are very few that can safely be used for explanation and
prediction. Often, to speak in terms of Hempel's analogy, one discovers
isolated knots with loose threads attached, awaiting the systematic
I effort needed to ,tighten them and to tie them together. Arid, most
6 See G_ W. Allport's comprehensive discussion (1950) of the frustratioil-
aggression theory as it applies to the phenomena of prejudice.
frequently, even knots are not yet available. For example, in surveys
conducted in 1948, 66 per cent of a sample of the American population
agreed with the statement that "most people can be trusted," whereas
in Germany only 6 per cent agreed with it (National Opinion Research
Center, 1948). One feels intuitively that this is an important difference.
But one searches in vain for a well-defined theory of national character
or of the psychological impact of miltiary defeat and turbulent histori-
cal events in which the concept "confidence in other people" has a
systematic place. Under these circumstances, the potentially important
addition to knowledge on which these figures may bear remains hidden
unless a relevant hypothesis is formulated and tested in a variety of
situations.
Research and theory must proceed together toward increases of
knowledge. Each has an important contribution to make to the other.
Any social scientist may take either one as his starting point, but he
has an obligation to consider at some point the bearing of his work
on the interrelation of the two. If he concentrates on empirical re-
search, he must examine its relevance to social theory if its potential
contribution is to be realized. If his major interest is in the develop-
ment of sodal theory, he must take into account ways of testing and
expanding his theory by empirical research it it is to become more than
interesting speculation. / .
Theory as a Basis for Research
Theories-even fragments of partially developed theory-provide

an important guide for the direction of research by pointing to areas
that are likely to be fruitful-that is, in which meaningful relationships
are likely to be found. As Merton (1957) has put it:
. . . if concepts are selected such that no relationships between
them obtain, the research will be sterile, no matter how meticu-
lous the subsequent ob~ervations and inferences. The importance
of this truism lies in its implication that truly trial-and-error
procedures in empiriqal inquiry are likely to be comparatively
unfruitful, since the number of variables which are not signifi-
cantly connected is infinitely large.
Cohen's attempt (1955) to formulate a theory that would ac-

count for the appearance of a "delinquent subculture" in certain sectors
of American communities illustrates this function of theory for research
stimulated by specific social urgencies, just as the frustration-aggression
theory was stimulated by and served to guide clinical and experimental
psychological research. In developing his theory, Cohen drew on
the findings of prior research on delinquency, on his own experience
with delinquent gangs, and on other theoretical formulations in
psychology and sociology. Briefly, the development follows these steps:
1. All human behavior has the purpose of solving problems.
2. Although all individuals have problems, different kinds of
problems are not randomly distributed throughout society;
certain kinds of problems are more likely to be encountered
by members of the working class than by professional people,
by young people more than by older people, by boys more
than by girls, etc.
3. A crucial condition for the emergence of new cultural forms
is the existence, in effective interaction with one another, of a
number of persons with similar problems of adjustment.
4. An important condition for an individual's adjustment is that
he be thought well of by others who are important to him, and
by himself.
5. For many working-class children, this condition is hard to
fulfill. In school, in recreation centers, in all the activities of
the larger community, they are judged in terms of middle-
class standards, which ma~y working-class children, for a v~riety
of reasons, are not well equipped to meet.
6. The delinquent subculture deals with this problem by provid-
ing criteria of status which these children can meet.
7. Since most working-class children have been exposed to, and
have partially internalized, middle-class standards, there is a
conflict between these standards and those of the deiinquent
gang. .
8. In order to eliminate this conflict, the delinquent subculture
explicitly rejects (that is, it does not simply ignore) middle-
class standards, particularly as they refer to the symbols and
THEORY AS A BASIS FOR RESEARCH 489
actuality of achieved status; the criteria for status within the
delinquent gang are the opposite of those held by "respectable"
society.
Cohen points out that this theory suggests a need for data of a
somewhat different sort than those that have usually been collected
in research on delinquency. First, he suggests that more accurate data
on the occurrence of delinquent acts are needed. Present statistics are
based on official (police and court) records, or on quasi-official records
of social agencies, neighborhood centers, etc. There is ample reason
to believe that these statistics do not come anywhere near reflecting
all the delinquent acts that occur. The problem is not simply, or even
primarily, one of underestimating the extent of delinquency; there is no
assurance that these records give an accurate picture of how delin-
quency is distributed in various sectors of society (in other words, there
is n_o assurance that they constitute an unbiased sample of delinquent
behavior). Studies comparing delinquent and nondelinquent children
have typically selected their delinquent subjects from these official or
quasi-official records, and their nondelinquents from among children,
presumably comparable in other respects, who do not appear in such
records. Cohen maintains that, in order to collect accurate statistics
and to find unbiased samples of delinquent and nondelinquent chil-
dren, it is necessary to start by selecting a randbm sample of the
juvenile population of a given area and then to determine, through
careful and sympathetic interviews, the actual occurrence of delin-
quency among this sample.
, One may ask what it is about Cohen's theory that leads to this
particular recommendation. Are not all students of juvenile delin-
quency aware of the limitations of present statistics, and would they
not all agree with tqe desirability of an approach such as Cohen
suggests? Perhaps so. But a test of Cohen's theory requires such data,
since the theory rests explicitly on the assumption that delinquent
gangs are more commbn I
in working-class neighborhoods than in other
parts of the community. Although present official and quasi-official
statistics show much higher incidence of delinquency in working-class
neighborhoods, some ~investigators have suggested that this is a reflec-
tion of the social processes by which children come to the attentioT"
of official agencies rather than a reflection of the true incidence of

delinquency. If it should prove to be true that the existence of gangs
which are delinquent in relation to middle-class standards and symbols
of achieved status is as typica1 of midd1e-dass as of working-class
neighborhoods, Cohen's entire theory would be invalidated.7
Second, Cohen recommends a changed emphasis in the kinds of
information sought in studies of delinquent and nondelinquent chil-
dren. Most such research has been concerned with correlates of delin-
quency-faIPily background, personality, neighborhood characteristics,
etc. Although recognizing that such data are useful, Cohen suggests
that still more useful would be information about the delinquent
behavior itself, and above all on the collective or individual nature of
delinquent activity and on how delinquency_differs in individual and
group situations. Such information, in combination with information
about the background factors that have typically been studied, would
help to provide an understanding of how delinquent activity serves as
a solution of the gang members' problems of adjustment.
'Finally, Cohen's theory suggests a need for more research on
delinquent groups as social systems-that is, on th~ structure, processes,
history and subculture of delinquent gangs-rather than on the delin-
quent individual.
So far, this illustration has been concerned with ways in which a
theory suggests fruitful research approaches to phenomena in the
general area with which the theory is concerned-in this case, juvenile
delinquency. But theQry provides leads for research in still another
way-by suggesting other kinds of phenomena that may-perhaps be
understood in the same general terms. The first four steps in Copen's
theoretical formulation, as -given on page 488, constitute an approach
to the understanding of how any supculture arises. Although Cohen
does not elaborate this point, he suggests that s,:!ch different subcultur:es
as college fraternities and the world of jazz musicians may be under~
stood in the same terms. Research on such groups would concentrate
on discovering the common problems faced by the members an~ the
7 Cohen recognizes the existence of delinquent gangs in middleclass neighbor-
h~ods and suggests conditions, in accord with his theory, which might account for
them. However, the theory rests on the assumption that delinquent gangs occur
primarily in workingclass neighborhoods.
THEORY AS A BASIS FOR RESEARCH 491
ways in which the particular patterns of these subcultures help the
members to deal with them.
Closely related to this function of theory is another contribution
it makes to research: increasing the meaningfulness of the findings of
a given study by making it possible to perceive them not as isolated
bits of empirical information but as a special case of the working out
of a set of more abstract propositions. For example, if more adequate
statistical studies confirm the assumption that delinquent gangs are
more common in working-class than in middle-class neighborhoods,
this fact in itself would not add greatly to our understanding of delin-
quency. But looking at gangs as an instance of a general tendency for
people faced with common problems to work out a common solution
helps to explain the association between working-class neighborhoods
and delinquent gangs.s
This linkage of the specific empirical findings to a more general
concept has still another advantage: It provides a mOre secure ground
for prediction than do the findings by themselves. The prediction may
be concerned with estimating whether a relationship between two
variables which has been observed in the past will continue in the
future, or it may be concerned with estimating whether changes in
certain conditions will lead to changes in the observed relationship. To
go back to the delinquencl-example: A variety of/steps have been
suggested as ways of reducing delinque~-more adequate recrea-
tional facilities, provision of counseling services in the schools, censor-
ship of comic books, fining the parents of delinquents, increasing the
number of police in high-delinquency areas, etc. Cohen does not dis-
cuss 'possible remedial measurcs, pointing out that there is not neces-
sarily a direct link between understanding the "cause" of a phenom-
enon a~d finding a "cure." Nevertheless, his theory would seem to
suggest that a measure :intended to reduce gang delinquency is likely
to be Sllccessful to the 'extent that it either changes the standards by
which working-class children are judged in school and in the com-
munity generally, or helps them to meet those standards.
In summary: Theory increases the fruitfulness of research by
providing significant l~ads for inquiry, by relating seemingly discrete
findings by means of similar underlying processes, and by providing an
S This function of theory has been discussed in more detail in Chapter 2.
explanation of observed relationships. The more research is directed

by systematic theory, the more likely are its results to contribute
directly to the development and further organization of knowledge.
The Contribution ot Research to the Development ot Theory

The contributions between theory and research are not all in one
direction. Theory stimulates research and enhances the meaning of its
findings; empirical research, on the other hand, serves to test existing
theories and to provide a basis for the development of new ones.
From a well-formulated theory, deductions can be drawn about
what will happen in various situations under specified conditions.
Thus, from Hull's reinforcement theory of learning, deductions can
be drawn about such diverse matters as the rate at which different
parts of a task will be learned, the effects of spaced versus massed
practice, the effects of rewarding a given act regularly versus occasion-
ally, etc. These deductions provide hypot]1eses for empirical research.
If a given hypothesis is confirmed by studies planned to test it, the
studies have contributed to verifying the entire theoretical structure
from which the deduction was drawn. Ifr on the other hand, a hy-
pothesis is not confirmed by research, the theory must be re-examined
to consider whether it should be discarded as invalid or whether some
modification would make it consistent with the research findings.
In the latter case, of course, further studies are required to test whether
deductions from the modified t~eory will be supported by empirical
observation.
In the social sciences at the present time few theories are worked
out in sufficient detail to suggest specific studies testing their validity.
Thus research more often has 'the function of contributing to the
development of theory than to its testing ..
Contributions to the development of theory may be consciously
planned for, or they may be more or less accidental. The social sci-
entist who deliberately sets out to develop some aspect of theory is
likely to follow one, or both, of two approaches. He may re-examine
existing studies, or he may plan a program of related studies focused on
the question in which he is interested. In either case, he is not likely tc
RESEARCH AND THE DEVELOPMENT OF THEORY 493
start completely from scratch; on the basis of earlier studies, or other
theoretical formulations, or his own observations, he usually has certain
concepts in mind, certain possibilities of alternative formulations, etc.
His examination of existing studies, or his plan for studies to be carried
out, will be made in the light of these concepts and these tentative
formulations.
The development of theory on the basis of existing studies has
been demonstrated by Merton and Rossi (1957) in "Contributions to
the Theory of Reference Group Behavior." Merton and Rossi use as
their material numerous researches in The American Soldier (Stouffer
et al., I 949a ) bearing upon the ways in which individuals select groups
as reference points for the evaluation of their own status. Although
The American Soldier studies were not planned in terms of the refer-
ence-group concept, a number of the studies concerned with morale
and satisfaction were interpreted in these terms.
That people appraise their own status in comparison with that of
others is not a new notion. Merton and Rossi, in their search for a
theoretical formulation, examine efforts by other social scientists who
have used this or similar concepts. Using all these available "knots"
and "loose threads," the authors discuss the common elements in nine
studies from The American Soldier; in each of these studies the atti-
tudes of the soldiers are explained in terms of the Irelative deprivation
they suffered compared with the reference group they used as a stand-
ard. The authors point out how the selection of a group for comparison
varies from situation to situation. The soldiers evaluate their own situa-
tion sometimes with reference to members of their in-groups, some-
times with reference to another group. They may use as bases of com-
parison groups of the same status as their own, or of higher or lower
status. Merton and Rossi I
see emerging from this re-examination of
research results a ques~ion of:
... central importance to a developing theory of reference
group behavior: u'nder which conditions are associates within
one's own groups taken as a frame of reference for self-evaluation
and attitude-formation,
I
and under which conditions do out-
groups or non-membership groups provide the significant frame
of refpence?
If that question were answered, we would have a reference-group

theory instead of only the first steps toward its development. Such a
theory would be an important tool in predicting and understanding
morale in many situations.
In this example, Merton and Rossi ~ave done what amounts to
an ex post facto interrelation of separate pieces of research through
the use of a single unifying concept-that of reference groups. In so
doing, they have extended and clarified the reference-group concept,
and have pointed out the direction in which further research is needed
if the concept is to take its place in a theory of reference-group behavior.
The Yale Communication Research Program provides an example
of the planning and carrying out of coordinated systematic research
for the purpose of developing theory. The purpose and nature of this
program have been described by Hovland, Janis, and Kelley (1953):
A great deal of descriptive information has accumulated
concerning persuasive communications-such as educational pro-
grams, publicity campaigns, advertising, and propaganda-and #
their effects on behavior and opinion ..Most of this information

comes from studies which focus on practical questions posed by
communicators who make use of mass media. But for purposes
of developing scientific propositions which specify the condi-
tions under which the effectiveness 0{ one or another type of
persuasive communication is increased or decreased, the avail-
able evidence is extremely limited. Although applied research
can be useful in suggesting tentative hypotheses and in posing
theoretical problems for further analysis, the practical emphasis
often results in the neglect of significant and provocative issues
which do not appear to have immediate application. Hence ba~ic
research is greatly needed to supplement the findings derived
from investigations of a practical nature. Such research,- iI1yblv-
ing psychological experiments in a -communication setting, can
contribute to our understanding of the processes of memory,
thought, motivation, and social influence.
The first phase of the program, as reported by Hovlan<;1.and his
colleagues, focused on relatively specific questions about factors and
conditions infl~encing the effectiveness of communicatiqns in pro
ducing opinion change; for example, "How do differences in the credi
bility of the communicator affect 1) the way in which the content
and presentation are perceived and evaluated? 2) the degree to which
attitudes and beliefs are modified?," "Under what conditions does
overt verbal conformity facilitate or interfere with acceptance of the
beliefs or opinions advocated by a communication?," etc. The initial
studies provided at least tentative answers to many of these questions.
At the same time, they suggested new ways of organizing the problem
that seemed likely to be more fruitful in developing a unified theory
of communication and opinion change. These new topics "cut across
the usual categories of communication research," emphasizing the
processes involved in opinion change rather than the conditions under
which it occurs. They included study of internalization processes by
which "outer conformity" is transformed into "inner conformity"; the
relation between conflict and opinion change; and the relations among
perception, judgment, and concept formation.
Research. contributes to the development of theory also in less
phinned-for ways. It clarifies concepts; it initiates, it reformulates, and
it refocuses theory.9
One of the most frequent contributions of empirical research is
to the clarification of the concepts used in theoretical formulations.
This occurs because research cannot proceed on the basis of concepts
phrased in general terms; for research purposes, so~e indicator of the
concept must be found.lO In order to decide on an observable indicator
of a concept, one must be clear about what he means by the concept.
For,example, if an investigator wishes to study the influence of certain
features of community structure on mental health, he must have some
way of estimating the mental health of his subjects. But in order to
J
devise an appropriate measure, he must be clear about what he means:

by "mental health." 1he need for "working definitions" in carrying
out research often poipts up a fuzziness in the concepts used in theo-
retical formulations and forces more precise definitions.
Research may alJo lead in unplanned ways to the initiation of
!) In this section we have drawn extensively on the discussion by Merton
(1957, Chapter 3). I, .
10 This is, of course, the process of establishing "working definitions" discussed
in Chapter 2, pages 42-44.
theory. An investigation, whether it has its origin in some theoretical

formulation or in a more trial-and-error approach, may yield an un-
expected finding that seems surprising because it is incompatible either
with existing theories or with other facts. In his search for an explana-
tion, the investigator may formulate a new hypothesis, which then
becomes the basis for subsequent research. Stouffer (in Merton and
Lazarsfeld, 1950) describes in the following terms the incident which
first led to the introduction of the concept "relative deprivation" in
the interpretation of The American Soldier studies:
In the Research Branch I wen remember our puzzlement,
which went on for months, over the finding that Northern
Negroes in Southern camps, in spite of the fact that they said
they wanted to be stationed in the North and that they resented
discrimination in Southern buses and by Southern police, showed
as favorable or more favorable responses to items reflecting p_er-
sonal adjustment in the Army than did those in Northern camps.
Some of our analysts were almost in despair at this discrepancy.
They actually held up the report on t~eir study for over a month
while they checked and rechecked in the vain hope of finding
errors in the data or analysis to explain the paradox. When, even~ -
tually, it was'suggested that the Northern Negro soldier in the
South had very great advantages over Negro civilians in the South
and that the advantages over Negro civilians in the North were
much less, a clue to the paradox appeared. After a number of such
experiences, it became evident that some concept like "relative
deprivation" might be useful.
To be sure, the notion that people appraise their own status in
comparison with that of others was not entirely new. It had been sng-
gested earlier by George H. Mead, Herbert Hyman, and Muzafer
Sherif. Nevertheless, the formulation made in The American 'Soldier
had certain new elements and considerably extended the concept. It
seldom happens that an unexpected finding leads, at one stroke, to the
development of an entirely new theoretical concept.
On other occasions, research may lead to the reform,ulation or
extension of theory by highlighting hitherto neglected .facts. This
function differs from the preceding one in that, in the present case,
the facts are not inconsistent with existing theories or with other facts.
The need is not to find some new explanation, but to reformulate an
existing theory which is capable of accounting for them.
For example, in the wen-known "Hawthorne studies" (Roeth-
lisberger and Dickson, 1939) the investigators started with the theory
that physical conditions affect work output. This relationship had,
in fact, been well established; the investigators were concerned with
identifying the effects of specific changes, in order to find optimal
conditions. At first they found what they had expected: improvement
in physical conditions led to increased output. But then they found
that changes in the direction of objectively poorer conditions were still
accompanied by increases in output. This led to a re-examination of
the initial theory. The difficulty appeared to be not that the theory
was incorrect, but that it omitted important variables. Physical condi-
tions did affect output, but this effect was overshadowed by the far
greater effect of other changes accompanying the experimental manipu-
lations. The workers in the experimental group knew that they were
taking part in an experiment and were interested in its outcome; they
had a different relationship with their supervisor than that which ob-
tained in the regular departments of the plimt; their being set apart
as a small group led to an increase in cohesiveness among them. These
social and attitudinal factors were so important that they obscured the
effects of changes in physical conditions. The result was a significant
broadening of the theory that output is influenced by factors within
the work situation, to include social as well as physical conditions.
Empirical research may also refocus theory by shifting interest to
new areas. This comes about chiefly through the development of new
research procedures, which leads to a rapid growth of research on
matters that can be investigated by these techniques. As research ac-
cumulates, theoretical interest shifts in the same direction, not because
research findings automatically lead to theories, but because an abun-
dance of empirical observations provides fertile ground for the devel-
I
opment and testing of theoretical concepts. Consider, for example, the

problem of the nature and organization of human abilities. Even be-
fore the days of what we now consider scientific psychology, this prob-
lem had evoked some interest. But the faculty psychologists and the
phrenologists who " were interested in aspects of the problem had no
effective means of, studying it; they were necessarily limited to specu-
lative armchair theorizing and appeals to dramatic instances (or, as

was often the case, dramatic coincidences). With the development of
techniques for the measurement of abilities, reliance on dramatic in-
stances could give way to systematically collected data; and otherwise
sterile speculation could interact with these data, pose new problems
for measurement, call for experimentation with factors affecting the
(mtcomes of measurement, and thus grow into fruitful theory. It is not
surprising, therefore, that Binet's response to the practical need for
some dependable method of diagnosing feeblemindedness marked the
onset of a new era in the history of psychology-an era in which the
successes, failures, and abuses of the intelligence-testing movement
not only led to the critical examination of basic premises and the
growth of psychometric theory but also served as a spur to efforts to
meet the challenge of measurin~ther complex aspects of personality.
And, of course, the extension of m~asurement into the domain of per-
sonality inevitably raised issues of conceptualization of persomliity
variables, etc., etc. .
Or, to take a less familiar example, the repercussions of Which
extended to a much narrower domain than' those of the testing move-
ment: When Zeigarnik decided to test an inference from Lewinian
theory that subjects would remember better tasks which they were not
permitted to complete than tasks which they carried through to com-
pletion, she introduced an experimental technique which lent itself
to the testing of a wide variety of hypotheses. There occurred there-
upon a proliferation of research and theory on the effects upon. tension '
systems of such variables as success, and failure, task versus ego oriel1ta-
tion, carrying out substitute actions that differ from the interrupted
ones in certain systematic respects, etc. l l
Interrelation of Theory and Research-A Summary

The relation of theory and research is one of mutual contribu-
.tions. Theory can point to areas in which research is likely to be ffuitful,
can summarize th,e findings of a number of specific studies, and can
11 For an unusually complete account of the interaction of theory and research
along these lines, see Lewin (1951), "Formalization and Progress in Psychology."
SUMMARY 499
provide a basis for explanation and prediction. Research findings, on
the other hand, can test theories which have been worked out, can
clarify theoretical concepts, and can suggest new theoretical formula-
tions or extend old ones. Moreover, the process of reciprocal contribu-
tion is a continuing one; research stimulated by theoretical considera-
tions may raise new theoretical issues, which in turn lead to further
research, and so on indefinitely. At whatever point in this spiral of
activities a social scientist chooses to work, the significance of his con-
_ tribution will increase with his insight into the processes that link the
results of separate studies to theoretical formulations. To conduct
research without theoretical interpretation or to theorize without re-
search is to ignore the essential function of theory as a tool for achiev-
i~g economy of thought.
APPENDICES
A. Estimating the Time and Personnel

Needed tor a Study
B. An Introduction toSarnpling
C. Questionnaire Construction and Interview Procedure

Appendix A
ESTIMATING THE TIME AND PERSONNEL
NEEDED FOR A STUDY
I N THE VERY earliest stage of planning, the investigator must consider

what resources of time, personnel, and money will be needed to
carry out the study he has in mind. This [s equally important for the
student planning a term paper, the doctoral candidate starting work on
his dissertation, the social scientist applying to a foundation for support
of a project, the consultant proposing a study to help in the solution of a
a client's problem. Under some circumstances it is possible to secure
the time, personnel, and money required by the tentative research plan;
in others, it is necessary to adjust the plan so that it can be carried out
within the available,resources. In either case, it is important to have; a
realistic estimate of what will be re'quired. ,
In this appendix, we shall make a few suggestions for estimating
time and personnel requirements. Translating these requirements into
appropriate salaries provides the basis for estimating that part of the
budget. Additional costs, such as office rent, travel expenses, purchase
or construction of equipment, etc., depend so much on the nature of
the specific study that little can be said in the way of general guidance.
There is, unfortunately, no rule of thumb by which pne can
predict exactly how much time will be needed to carry out, a study.
Perhaps the most efficient way of going about the task is to list the
various phases of the research procedure and estimate how much time
502
ESTIMATING FOR A STUDY 503
and how many people will be required for each, rather than to attempt
to arrive directly at an over-all estimate for the total project. The steps
discussed in the various chapters of this book provide a basic outline of
the steps that must be planned for: formulating the research question,
defining concepts, planning the study design, developing the data-
collection instruments, collecting and analyzing the data, writing the
report. But many practical steps not discussed in detail in this book
must also be considered; for example, the hiring and training of inter-
viewers or observers, negotiations with agency personnel in the case
of a field study, etc.
How does one estimate how long each operation will take? Pri-
marily, one learns through experience-usually sad. It appears to be an
almost universal rule that every operation takes longer than one would
anticipate if everything went smoothly. The estimate must provide
leeway for the snags that inevitably arise: the fact that the pretest.
shows that several interview questions need rewording, or that observers
need additional training in order to use the data-collection instrument
reliably; failure to find respondents at home on the first visit, and the
need for several call-backs; the decision of the director of an action
agency that he must consult his board of directors before giving permis-
sion to carry out research within the agency; the mistake in punching
scores onto calculating machine cards, which leads to errors in several
tables before it is discovered; and so on. One way ~f providing the extra
time required by such contingencies is first to estimate how long each
operation might be expected to take if it went reasonably smoothly,
then to go back over these estimates, increasing them by 50 per cent
o'r even more.
The'amount of time needed for the different stages of the research
po:ocess will depend Ito a considerable extent on the type of study. The
I
following records of time and personnel actually used in three studies
I
will illustrate both tre wide range in total time and personnel that may
be needed and the differences in the relative amount of time devoted
to various operation~.l
1 It should perhaps be noted that each of these budgets underestimates the
total period of time over which the study extended, since it includes only activities
directly concerned witH the study. Almost inevitably, however, there are some other
demands on the time of the research ~taff, with the result that total elapsed time is
greater than indicated by total time spent on the study.
504 ESTIMATING FOR A STUDY
Time Budget 1 is from the Deutsch-Collins study of intergroup

relations in bi-racial housing projects, described in Chapter 1. The pre-
liminary "experience survey" of housing experts, which took six
months, is not included; the budget starts from the point of specific
planning for the study of the effects of occupancy pattern. In this
study, personal interviews were conducted with about four hundred
white housewives (one hundred in each of four projects). The inter-
view schedule contained more than one hu~ed questions;' on the
average, a single i.nterview lasted about an hour and a quarter. In
addition, about one hundred Negro housewives and a number of
white children were interviewed. There were nineteen interviewers;
the time shown in the "Interviewers" column refers to total man-days,
not to actual elapsed time.
As we mentioned in Chapter 1, the study was preceded by.an
"experience survey" of housing experts, and began with rather definite
hypotheses drawn both from this source and from the general body
of literature on intergroup attitudes. Thus_ relatively little time was
required for formulation of the problem and the research design or for
analysis, compared with the large amount of time devoted to conduct-
ing and coding the interviews. The large amount of time needed for
establishment of categories and coding stemmed from the fact that
the interview consisted mainly of open-ended questions. In exploratory
studies aimed at the development of hypotheses, analysis is likely to
require a considerably greater proportion of time, since such studies,
by their very definition, frequently have to proceed on a lime-consum-
ing trial-and-error basis.
Time Budget 2 applies to a small-scale exploratory study. Its pur-
pose was to examine in systematic fashion the impressions left on
twenty participants by a training workshop they had attended ten
months prior to the study. These impressions were elicited in the
course of intensive, relatively unstructured individual interviews. On
the basis of the interviews, the investigator analyzed the probable
effects of the workshop, deriving a number of hypotheses that could
be tested in more rigorously designed subsequent studies. Since the par-
ticipants in the workshop lived at some distance from one another, an
exceptionally high proportion of the project time went into traveling
NNO........-lLl'\
U\ U\
N N
N N
o o
N\ N\
N N NU\ OU\
o o o N\
M
o o OM (V\ 0 0 LI'\ lr\ lr\

M M ...... N
r..
o
;:..,
I-<,
CI)
505
TIME BUDGET 2
AN EXPLORATION OF THE IMPACT OF A WORKSHOP FOR TEACHERS
(The figures indicate working days)
Study
Operation Director Consultant Secretary
Formulation 3 3
Negotiations with workshop organizers 1
Development of interview schedule 3 1 1
20 intensive interviews 15 6
Analysis and writing, draft 8 1 2
Editing and final report 1 5 2
Miscellaneous administrative and clerical work 3
Total working days'" 33 11 12
'"There are about 230 to 250 working days in a year, depending on vacation
arrangements.
Time Budget 3 refers to a study of the acceptability of answers to

anti-minority remarks made in public. In this study, playlets in which
one character made an anti-minority remark and another answered him
in different ways were presented to groups of people. Some of the
groups already existed for other purposes; e>thers were composed of
passers-by recruited on a busy street corner. Altogether, there were
about one hundred groups, compris-~ng some fifteen hundred individ-
uals. Each group session lasted about half an hour. Data were collected
primarily by means of short fixed-alternative questionnaires adminis-
tered during the experimental sessions. In addition, approximately two
hundred and fifty subjects were interviewed, each interview lasting
approximately fifteen to thirty minutes.
The rather unusual nature of the problem and of the experimental
stimulus (the playlets) led to a high proportion of time spent in. the
initial stages of problem formulation, research design, and develop-
ment of the playlets. On the other hand, the fact that data were
collected in groups, by means of short questionnaires and intervi~ws,
resulted in a relatively small proportion of time devoted to this aspect
of the research operation. There is no "Interviewers" column because
the interviewing was done by the staff members.
When research budgets are faulty, they almost invariably err on
the side of underestimation. Often the social scientist simply does not
realize how much work will be involved in his plans. Or, vaguely
~
.....OJ<:II
...
U
0 tr\ N'\ M V 1..1"\ ",0
N ......
0
'"
t--
OJ
en
] ~ 0
'oB OJ
\D
0
\D
~ U
en
...OJ
'" 0 0
'"t:l \D \D
0
U
...,on
III
1>01 '"... .:
OJ
c>: .....0 0
'"
N '"
"" 8
~f>l
U
<:: '"
Ill)
.:
.....: ~
....
~ ~
>- '"
<:II
>..
C1l .....
<:II .:
Ii""2 "0I:l.O ::; 0 I.r\
'" ...... tr\ 00 \D t-- z
0
'" "" '" 0- (3
?S ~=
0 .:
0 <:II
N\ U >
~ 0:;:: .:
""
IJl
t:: .....v 00 N
0
"
Ill)
0 OU"\OOll'\
Q
::::> ~ UC1l <::'"'" ci '"
T r-r. NNNff'..-4
0
T
T
'" "" :.a.:
T
N
~ 0 ;.a Z .:
~ ...<:i
... '"0..
::g ""
'"
c>:
05 U
<:II '"
T
0 0\.1"\001..1"\ 0
'" '" '" '"
'0
~ ~ '"....v OJ
'" ci
ff\ NNNN'\~ T
""
N .....
::: ::l
I:l.O ~
OJ
Z <:II
OJ
'"Z lQ
>.
<: v
,..c::
... /
!:II
~ >..8 0 oS
0 E-< '"t:lU '"
T
""
OV'\lI\Oll'\
NN
"" ......
ON
\D
\D 0 00
>- V5
::l
.~
OJ N
'"
N '">.
<:II
Q Q '"t:l
::::> Ill)
.!. .:
""
tJ:l
:g 2 bD .~ ;.;;;....
<: '"3
8
oS
'0
.:
:.a0
.: OJ
.... .:
0
'"
.~
OJ
0
~
.~
oU. '0 .", U 0
!:II
'0
.:<:II OJ 11"\
'" .: u '0
.: N
.:
'C" '"....
0; <:II
0,c
'"
.:0""
~ '"<:IIbD
<:II
...,0
.: '"
.:
;:: '"'"
0;:: !:II ....
.... '"0 .:c'" 0
=oc:8 .:: ""

0
z ...,
oS 0
00 '"
..........
..8U .b
N
....
t:! .c;; ~
OIl'" '"
OJ ...,'" ~r: o~ :::s
Os.:
OJ
0.. ~OJ :::s <:II
U .... >
<v '" <:II *'">. 0
0" .: OJ ..0
0 "0 """' ~ '0 .~ .- ~ ...c: <:II !:II
"0 "0
;0 o~~oI-J~bl) o~ <:II OJ
....
00 !:II
=~ --~ ~s..c~ ..8~ S63 ~~
.9 815 8t::",'::0~
tn.S '"
::l
:.g
.: OJ
....
<v
.... 8'" u ...... ...c:<:II.:~
0.)
0 ...c:
~ ~~~~8~~ ~ ~ ~ b.O~.:a~~
S "i3;> .- OJ :0 ."E ~ t) oS ~
..::; > .- .s ~ ~ ~ ~.-.~
Q) ~ 15
~ = C/)
Q) CI') '20
~
.
E-<
'""'(1) QJ>C1j~coOQJCI')"'Oo~'''''
ri:o 0 Owci)u~ wu ::E E-<
507
recognizing it, he may still make a minimal estimate because he believes

this will be more acceptable to his sponsors in the situation.
Sometimes the estimate is adequate for the work originally
planned, but unforeseen possibilities for investigation emerge during
the study. The social scientist may be carried away by zeal for collecting
data or for following up these newly emerging leads. Many studies col-
lect an enormous amount of irrelevant data, which are never used. To
strike a happy medium between concern for scientific and financial
economy and a flexibility of mind that permits a continuous develop-
ment and refinement of ideas is an art that few people have mastered. It
would be disastrous for the advancement of knowledge if the rigidity of
the budget should restrict the flexibility of the mind. But it is no light
matter either if flexibility of mind turns into undisciplined imagination
and involves much unforeseen time and expense, thus endangering the
continuity of the work because available resources have been exhausted.
AppendixB
AN INTRODUCTION TO SAMPLING
BY ISlDOR CHEIN
I N THIS APPENDIX we have not attempted to develop a manual of

sampling procedures. Nor have we attempted to review statistical
formulae and procedures appropriate for the handling of data obtained
in sampling studies. We have aspired, without resorting to mathe-
matical analysis and within the limits of available space, to give the
reader some basis for an intelligent appreciation of the considerations
involved in sampling. l
Some Basic Definitions and Concepts

A populatioll' is the aggregate of an of the cases that conform
to some designatea set of specifications. Thus, by the specifications
"people" and "resi'ding in the United States," we define a population
consisting of a11 the people who reside in the United States. We may
similarly define populations consisting of an the shop stewards in a
1 For more comprehensive discussions of sampling and appropriate statistical
treatments, see the chapters by McCarthy in Jahoda, Deutsch and Cook (1952,
Volume 2), by Kishjin Festinger and Katz (1953), and by Cochran in Snedecor
(1956); also see Deming (1950), Cochran (1953), and Hansen, Hurwitz, and
Madow (1953). For a very broad discussion of sampling theory and practice, includ-
ing a review of empirical tests of sampling, see Stephan and McCarthy (1958).
509
510 AN INTRODUCTION TO SAMPLING
factory, a11 the households in a particular nty district, a11 the boys in
a given community under sixteen years of age who are stamp collectors,
all the case records in a file.
By certain specifications, one population may be included in
another. Thus, the population consisting of all the men residing in
the United States is included in the population consisting of all the
people who live in the United States. In such instances, we may refer
to the included population as a sub-population, a population stratum,
or simply as a stratum (pl. strata). A stratum may be defined by one
or more specifications that divide a population into mutually exclusive
segments. For instance, a given population may be subdivided into
strata consisting of males under twenty-one years of age, females under
twenty-one years of age, males from twenty-one through fifty-nine
years, etc. Similarly, we;may specify a stratum of the American popula-
tion consisting of white, male, college graduates who live in New
England and who have passed their seventy-fifthJ>irthday; or we may
have some reason for regarding this group of individuals as a popula-
tion in its own right-that is, without reference to the fact that it is
included in a larger popUlation.
A single member of a population is referred to as a population
element. We often want to know how certain characteristics of the
elements are distributed in a population. For example, we may want
to know the age distribution of the elements or we may want to know
the proportion of the elements who prefer one political candidate to
another. A census is a count of all of the elements in a population
and/or a determination of the distributions of their characteristics,
based on information obtained for each of the elements.
It is generally much more economical in time, effort, and money
to get the desired information for only some of the elements than (or
all of them. When we (select some of the elements with the intention
, I
of finding out something about the population from which they are
taken, we refer to that group of elements as a sample. We hope; of
course, that what we find out about the sample is true of the popula-
tion as a whole. Actually, this mayor may not be the case; how
closely the information we receive corresponds to what- we would
find by a comparable census of the population depends largely on the
way the sample is selected.
SOME BASIC DEFINITIONS AND CONCEPTS 511
For example, we may want to know what proportion of a popula-
tion prefers one candidate to another. We might ask one hundred
people from that population which candidate they prefer. The propor-
tion of the sample preferring Mr. Jones mayor may not be the same
as the corresponding proportion in the population. For that matter,
even the actual distribution of votes in an election may not correctly
represent the distribution of preferences in the population. Unless
there is'a 100 per cent turnout, the actual voters constitute only a
sample of the population of people eligible to vote. A very high propor-
tion of the people who prefer Mr. Smith may be overconfident with
respect to their candidate's chances and neglect to come to the polls; or
they may be living in a rural area and be discouraged from coming to
the polls by a heavy downpour. The election results may properly
dete~mine which candidate will take office, but they will not neces-
sarily indicate which candidate is preferred by a majority of the popu-
lation. 2 Similarly, the early returns in an election may be taken
2 It has been a common practice to predict the outcome of an election on the
basis of a pre-election sample survey which, at best, answered only the question of
preferences. The results have occasionally been disastrous_ The fiascos are by no
means attributable simply to the failure of the samples to represent the distribution
of preferences in the population at the time the polls were taken. In one instance
(the presidential election of 1948), the pre-election surveys showed that a large pro-
portion of people were undecided, and there are clear indications that an unantici-
pated consolidation of opinion in this group helped to confound the predictors. As
already indicated in the text, the fact that different proportions of those who prefer
different candidates may actually vote complicates the translation o.f preference
estimates into election forecasts.
There are also measurement problems involved. Preferences measured one way
m3j or may not correspond to preferences measured another way. Thus, behavior in
the voting qooth does not necessarily correspond to preferences expressed to an
interviewer. The former is generally accepted at face value as the more valid measure,
but we have no certainty that this is the case. A housewife, for instance, may follow
her husband's preference rather than her own, at the last moment, and it is possible
that there may be enough such instances to materially affect the outcome of an
election; similarly, othe~ kinds of subjectively felt pressures or momentary impulses
may take effect in the !,!Iection booth. Practical politicians seem to feel that the
position of their candid~te's name on the ballot affects his chances, as do the names
of other candidates running for other offices on the same ticket; such effects may
have bearing on voting behavior without affecting preferences.
Further complications arise from the gerrymandering of election districts and
other factors (e.g., the. electoral college system), which have the effect of giving
different voters different weights in determining the outcome of an election. Perhaps
the moral of this footn6te will be clear: The usefulness of findings obtained from a
sampl~_may depend in Ilarge measure on factors which are extraneous to the sam-
pling l,sues per se. Nor is it easy to draw a hard-and-fast dividing line between the
as a sample of the population of returns; and, as everyone knows, they

can be thoroughly deceptive.
>
In the case of elections and in the case of early returns in a
national election, there probably is not much we can do to guarantee
that the samples will correctly represent their populations. The out-
come of an election we must, perforce (barring some radical revisions
of election procedures), accept on faith that it does reflect the popular
will. And if we are misled by the early returns with respect to the final
outcome, then, at least, our errors are soon corrected. There are, how-
ever, situations in which we can to some extent control the properties
of the sample. In these situations, the way we go about drawing the
sample can, if not guarantee, then at least increase the likelihood that
the sample returns will not be too far from the true population figures
for our purposes. We can never guarantee that the sample returns do
reflect the population with respect to the characteristics we are st}ldy-
ing unless we have simultaneously conducted a complete comparable
census. We can, however, devise sampling plans which, if properly
executed, can guarantee that, if we were tu repeat a study on a number
of dIfferent samples selected from a given population, our findings
would not differ from the true population figures by more than a
specified amount in more than a speCified proportion of the samples.
For instance, suppose that we frequently want to know what
percentage of a population agrees with certain statements. On each of
these occasions we might put such a statement to a sample, compute
the percentage who agree, and take this result as an estimate of the
proportion of the population who agree. We can devise a n~mber of
sampling plans that will carry the insurance that our estimates will hot
differ from the corresponding true population figures by, say, more than
5 per cent on more than, say" 10 per cent of these occasions; the
estimates will be correct within 5 percentage points (the margin of
error or limit at accuracy) 90 per cent of the time (the probability or
confidence level). We can similarly devise a number of sampling plans
that will produce correct results within 2 percentage points 99per cent
factors which are' extraneous and those which are not. Thus, what is extraneous to
the sampling of one population (e.g., eligible voters) may be intrinsic to the
sampling of another (e.g., actual voters); the ambiguity arises when we sample one
popUlation with the intention of learning something about the other.
of the time; or within any other limits of accuracy and any assigned
probability. In practice, of course, we do not repeat the same study on
an indefinite number of samples drawn from the same population. But
our knowledge of what would happen in repeated studies enables us to
say that, with a given sample, there is say a 90 per cent probability that
our figures are within 5 percentage points of those that would be shown
by a census of the total population using the same measures. Having
set our level of aspiration for accuracy and confidence in the findings,
we would select from the available alternatives the sampling plan
which can be most economi~ally carried through. Needless to say, the
higher the level of aspiration, other conditions being equal, the higher
the cost of the operation.
A sampling plan that carries such insurance may be referred to as
a representative sampling plan. Note that in this usage the word
"representative" does not qualify "sample," but "sampling plan." What
a. representative sampling plan can do is to insure that the odds are
great enough that the selected sample is, for the purposes at hand,
sufficiently representative of the population to justify our running the
risk of taking it as representative.
The use of such a sampling plan is not the only kind of insurance
that can be taken out to decrease the likelihood of misleading sample
findings. Another involves taking steps to guarantee the inclusion in
the sample of diverse elements of the population and to make sure
( either by controlling the proportions of the various types of elements
or by analytical procedures in the handling of the data) that these-
diverse elements are taken account of in the propoItions in which
they occur in the population. We shall consider this type of insurance
at greater length in our discussion of quota sampling and of stratified
random sampling. :
It should perhaps be emphasized that the dependability3 of survey
findings is affected not only by the sampling plan and the faithfulness
with which it is carried out, but also by the measurement procedures
used. This is one reJson why sample surveys of a large population can,
3 Throughout this appendix, the terms "accuracy," "dependability," and
"precision" are used ihterchangeably. Although technical distinctions are some
times made among these words, in most discussions of sampling they are used
as synonyms.
in practice, produce more dependable results on some matters than

can a census. There simply are not enough highly skilled interviewers
available to get anything beyond the most superficial information in
a national census; a survey on a smaller scale puts ,Jess of a drain on
the available supply of interviewers and also more readily p.ermits a
relatively intensive training program. Similarly, a smaller-scale survey
may make it economically feasible to spend more time wjth each
respondent and, hence, make it possible to use measurement devices
that could not be seriously considered (except on a sampling basis4)
I)} connection with a census of a large population. 5
The basic distinction in modern sampling theory is between prob-

ability and nonprobability sampling. The essential characteristic of
probability sampling is that one can specify for each element of the
population the probability that it will be included in the sample. In
the simplest cas'RJeach of the elements has the same probabilify of
---
being included, but this is not a necessary condition .. What is necessary
is that for each element there must be SOlTIe specifia~I~l?Jqbability that
it will be inclu~ed. ~is _point will. be considered more fully in con-
nection with the discussions of simple random samples and stratified
randb'm samples. In non probability sampling,' there is no way of
estimating the prob~ility that each element has of being included in
I
4 It ,is' not uncommon nowadays to collect certain items of information on a

sampling basis in the course of conducting a census for other items of information.
5 There is another reason why sampling surveys may produce more dependable
information than censuses. In practiCe, no census ever reaches all t~e population
elements; in effect, what is supposed to be la census is actually a sample, albeit a
sample which includes a very high proportion of the population elements. If the_.
ul)reacbed elements differ markedly from those that are reached, the result may be
quite different from the true population value even though the unreached elements
may De a relatively small proportion of the population. Not all of the unreached
are unreachable; they vary along a continuum of accessibility, depending on the
amount one is prepared to invest in trying to reach thein. In a relatively small sc~le
survey, one may be able to afford a greater relative investment in trying to reach the
comparatively inaccessible elements. A census may also go after the known un-
reached on a sampling basis (e.g., by having intervie'\Yers return one or more times
to dwellings where no one is at home on the first visit, or by sending more highly
. skilled interviewers to talk with respondents who have refused to _give the d",sired
information), but there is also an issue of the relative proportion of unknown
unreached in the two procedures and this is a function of what one can afford to
invest in exploring the terrain (e.g., discovering dwelling units in unsuspected
places)
the sample, and no assurance that every element has some chance of
being included. 6
Probability sampling is the only approach that makes possible
representative sampling plans. It makes it possible for the investigator
to estimate the extent to which the findings based on his sample are
likely to differ from what he would have found by studying the popula-
tion. Conversely, if he uses probability sampling, he can specify the
size of the sample (or the sizes of various components of complex
samples) that he will need if he wants to have a given degree of
certainty that his sample findings do not differ by more than a specified
'lmount from those that a study of the total population would yield.
The major advantages of nonprobabiltiy sampling are convenience
and economy-advantages that may outweigh the risks involved in not
using probability sampling. Precise comparisons of the relative costs
of the two approaches to sampling are, however, not available (see
Stephan and McCarthy, 1958). Moreover, the comparative costs will
vary depending on the number of surveys that are contemplated. Thus,
if a number of surveys of the same population are to be carried out, the
cost of preparing and maintaining lists from which to sample (generally
a necessary step in probability sampling) can be distributed over all
of them.
Major forms of nonprobability samples are: accidental samples,
quota samples, and purposive samples. Major forms of probability
samples are: simple random samples, stratified random samples, and
warious types of cluster samples. 7
6 If there is a class of elements that have no chance of being included, this

implies a restriction on the definition of the population. If the nature of this
class of elements is unknown, then the precise nature of the population is also
unknown. If there is no ~ssurance that every element has some chance of being
included, this uncertaintYlimplies that there can be no assurance as to the precise
., nature of the population that is being sampled.
7 The reader should' be warned that "accidental sampling" and "random
sampling" are technical terms, as defined in the text. The words "accidental" and
"random" may have quitd different connotations in ordinary everyday usage . .These
meanings should not be confused with those assumed in the technical usage. Thus,
it may be no "accident'l (everyday usage) that a sampler picks the cases he does in
an "accidental sample" 4~echnical usage). In everyday usage a "random sample"
may connote any nonpurposive sample or what is technically defined as an acciden-
tal sample. The justification of the technical usage would take us too far afield and
will not be attemDted here.
s:l6 AN INTRODUCTION TO SAMPLING
Nonprobability Sampling
ACCIDENTAL SAMPLES
In accidental sampling, one simply reaches out and takes the cases
that fall to hand, continuing the process until the sample reaches a
designated size. Thus, one may take the first hundred people one meets
on the street who are willing to be interviewed. Or a college professor,
wanting to make some generalization about college students, studies
the students in his classes. Or a journalist, wanting to know how "the
people" feel about a given issue, interviews conveniently available cab
drivers, barbers, and others who are presumed to reflect public opinion.
There is no known way (other than by doing a parallel study with a
probability sample or with a complete census) of evaluating the biases 8
introduced in such samples. If one uses an accidental sample, one can
only hope that one is not being too grossly misled.
QUOTA SAMPLES
Quota sampling (sometimes misleadingly referred to as "repre

sentative" sampling) adds insurance of the second type referred to
above-provisions to guarantee the inclusion in the sample of diverse
elements of the population and to make sure that these diverse ele-
ments are taken account of in the proportions in which they occur in
the population. Consider an extreme case: Suppose that we are sam-
pling from a population with equal numbers of males and females and
that there is a sharp difference between the two sexes in the character-
istic we wish to measure. If we did not interview any males, the results
of the survey would almost certainly be an extremely misleading picture
of the total population. In anticipation of such possible differences
between subgroups, the quota sampler tries to guarantee the inclusion
in his sample of enough cases from each stratum.
As commonly described, the basic goal of quota sampling is t):le
selection of a sample that is a replica of the populati0n to which one
1 8 Bias refers to the difference between the average of the estimates of a popula
tion value that would be obtained from a very large number of samples selected by a
given procedure and the actual population value, assuming identical measurement
processes.
NONPROBABILITY SAMPLINc' 517
wants to generalize-hence the notion that it "represents" that popula-
tion. If it is known that the population has equal numbers of males
and females, the interviewers are instructed to interview equal numbers
of males and females. If it is known that 10 per cent of the population
lies within a particular age range, assignments are given to the inter-
viewers in such a way that 10 per cent of the sample will fall within that
age range.
The question of the kinds of characteristics that must be taken
into account will be considered in more detail in the course of our
discussion of stratified random sampling. It is enough, for the moment,
to say that in the sampling of preferences, opinions, attitudes, etc.,
experience indicates that it is wise to take into account such bases of
stratification as age, sex, education, geographical region of residence,
socioeconomic status, and ethnic background. Not all these are equally
visible; the usual practice is to set the quotas for the interviewers in
terms of the more manifest traits and to get information in the course
of the interviews on the less manifest ones. The latter information
permits correction of the inadequacies of the sample by adjustments
introduced during the analysis, a procedure that will be illustrated in
the following paragraphs. It also calls attention to omissions, if any
should occur, of important segments of the population.
It often happens, in practice, that the various c9mponents of the
sample turn out not to be in the same proportions as the corresponding
strata are in the population. The interviewers may not have carried out
their instructions exactly; instead of interviewing equal numbers of
males and females, 55 per cent of the people they interviewed may have
been' males. Disproportions between the sample and the population
are most likely to occur, of course, in the less manifest traits which
have not been included 'as part of the specifications for the interviewers'
quotas. Suppose it is k~own that, in a given population, 40 per cent
~have not gone beyond grammar school; suppose, however, that only
20 per cent of the people interviewed fall in this category. The inade-
quacy in the sample can be corrected in the analysis by weighting the
different strata in terms of their proportions in the population. This
may be done by multiplying or dividing the obtained results by the
appropriate figure.
Let us say that the total sample consisted of 1,000 persons, of
whom 800 had attended high school, 200 had not. Suppose we asked
this sample whether they had seen a certain television program, and
the responses were as follows:
No High School High School Total
Yes 20 400 420
No 180 400 580
Total 200 800 1,000
In other words, one-tenth of the people without high-school education
and half of those with such education said they had seen the program.
If we wished simply to report the figures for the educational groups
separately, no adjustment would be needed. But if we wanted to esti-
mate the proportion of the total population that had seen the program,
our sample findings would be misleading. The program had been seen
by 42 per cent of the people in our sample. But our sample un~er
represented people in the lower educational category, overrepresented
those with high-school education. To derive an estimate of the correct
figure for the total population, we must calculate what the responses
would have been if 40 per cent of the people-in the sample had had only
grammar-school education, 60 per cent had attended high school (the
proportions we have assumed for the population). One way of doing
this is to multiply the responses of the no-high-school group by 2 (to
bring the 20 per cent in the sample up to 40 per cent), and of the high-
school group by three-fourths (to reduce the 80 per cent to 60 per
cent). This would give 40 "yes's" in the no-high-school group and 300
in the high-school group, or 340 for the total group; thus we would
estimate that 34 per cent of the population had seen the progr~m,
rather than the 42 per cent we would have estimated if we l;1ad
not weighted the strata in terms of their actual proportions in the
population.
From this example it should he clear-that the critical requirement
in quota sampling is not that the various population strata be sampled
in their correct proportions; but rather that there be enough cases from
each stratum to make possible an estimate of the population stratum
value, and that we know (or can estimate with reasonaBle acctlracy)
~the proportion that each stratum constitutes in the total population.
If these conditions are met, the estimates of the values for the various
NONPROBABILITY SAMPLING 519
strata can be combined to give an estimate of the total population value.
However, despite these precautions in the selection of the sample,
and the corrections in the analysis, quota sampling remains basically
an accidental sampling procedure. The part of the sample in any par-
ticular class constitutes an accidental sample of the corresponding
stratum of the population. The males in the sample are an accidental
sample of the males in the population; the twenty-to-forty-year-olds in
the sample constitute an accidental sample of the twenty-to-forty-year-
olds in the population. If the instructions received by the interviewers
and their execution of these instructions produce correct proportions
of the compound classes (e.g., white males in the twenty-to-forty age
range), the sample cases in these classes are still accidental samples of
the corresponding compound strata in the population. The total sample
is thus an accidental sample.
There is by now, however, enough experience with quota sampling
to make it possible to minimize the risks of at least certain types of
unfortunate accidents. It is known that interviewers, left to .their own
devices, are especially prone to certain pitfalls. They will interview
their friends in excessive proportion. But their friends are likely to be
rather similar in many respects to themselves. Now, consider the pos-
sibility that, in certain matters, people who do interviewing and others
like them are atypical of the population at large. If these matters are
involved in the survey, the sample results are likely to be inaccurate.
Once we are aware of the danger, however, we can take steps to dis-
courage the practice.
If interviewers fill their quotas by stopping passers-by and inviting
them to be interviewed, they will tend to concentrate on areas where
there are large numbers of potential respondents: the entertainment
centers of cities, business districts, railway and air terminals, the en-
trances of large depattment
I
stores. Such samples will overrepresent the
kinds of people who tend to gravitate to these areas. A concentration
I
on many varieties of l such areas will presumably be better than a con

centration on only ore, but, even so, such samples will underrepresent
the kind of people who seldom leave their immediate neighborhoods
and especially thos~ who seldom leave their homes. Often this will
I
make no difference, but it is conceivable that, on some matters at
some times, there may be sharp differences between the overrepre-
sen ted and the underrepresented population segments. When this is

the case, such a sample would, of course, yield misleading results.
Again, to be forewarned is to be forearmed.
If the interviewers fin their quotas by home visits, they will tend
to proceed along lines of convenience and striking appearance. Thus,
concentrating on certain times of the day, they will tend to miss the
kinds of people who are not at home at such times (e.g., working
women during the daytime). Similarly, they will tend to avoid the
upper stories of buildings without elevator service. They will tend to
favor corner buildings and to avoid dilapidated buildings and buildings
situated behind others. Such sampling tends to build in a systematic
socioeconomic bias (i.e., in each residential area, to overrepresent the
wealthier people) and possibly other biases as wen.
The point to be noted about selective factors such as these is that
they are not easily corrected during the analytical treatment of the data.
For many populations we know in advance the true relative proporti~ns
of the two sexes and of the various age groups, and so can correct for
disproportions in the sample, but what true proportion of what defin-
able population is most likely to be found at- a railroad terminal during
the course of a survey? The major control that an investigator has
available to him in connection with such variables is in the sampling
process itself. He can try to assure that important segments of the
population are not entirely unrepresented in his sample, try to benefit
from his experience and sample in such a way that many possibly
relevant variables are not too grossly distorted in his sample, a~d hope
that whatever disproportions remaip will not have any Dearing on the
opinions, preferences, etc., that he is interested in.
PURPOSrvE SAMPLES
The basic assumption behind purposive sampling is that 'with

good judgment and an appropriate strategy,one can hand-pick the cases
to be included in the sample and thus develop samples that are satis-
factory in relation to one's needs .. 'A common strategy of purposive
sampling is to pick cases that ar ju~ged to be !ypical of the population
r
in which one is interested, assuming that errors of judgment in the
NONPROBABILITY SAMPLING 521
selection will tend to counterbalance each other. Experiments on pur-
posive sampling suggest that, without an objective basis for making the
judgments, this is not a dependable assumption. In any case, without
an external check, there is no way of knowing that the "typical" cases
continue to be typical.
Purposive samples selected in terms of assumed typicality have
been used in attempts to forecast national elections. One such approach
is as follows: For each state, select a number of small election districts
whose election returns in previous years have approximated the over-
--all state returns; interview all the eligible voters in these districts on
their voting intentions; and hope that the selected districts are still
typical of their respective states. The trouble with the method is that
when there are no marked changes in the political atmosphere, one
can probably do as wen by forecasting the same returns that obtained
in previous years without doing any interviewing at all; when changes
are occurring, one needs to know how the changes are affecting the
selected districts in comparison with other districts.
Probability Sampling
Probability samples involve the first kind of insurance against mis-
leading results that we discussed earlier-the ability to specify the
chances that the sample findings do not differ by more than a certain
amount from the true population values. They may also include the
~ecpnd kind of insurance-a guarantee that enough cases are selected
from each relevant population stratum to provide an estimate for that
stratum of the population.
i
SIMPLE RANDOM SAMPLES
Simple random sampling is the basic probability sampling design;

it is incorporated in all of the more complex probability sampling
designs. A simple random sample is selected by a process that not only
gives each element in'the population an equal chance of being included
in the sample, but also makes the selection of every possible combina-
tion of the desired number of cases equally likely. Suppose, for example,
that one wants a simple random sample of two cases from a popula-
tion of five cases. Let the five cases in the population be A, B, C, D, and
E. There are ten possible pairs of cases in this population: AB, AC,
AD, AE, BC, BD, BE, CD, CE, and DE. Write each combination on
a disc, put the ten discs in a hat, mix them thoroughly, and have a
blindfolded person pick one. Each of the discs has the same chance
of being selected. 9 The two cases corresponding to the letters on the
selected disc constitute the desired simple random sample.
There are, in the tiny illustrative population of five cases, ten pos-
sible samples of three cases: ABC, ABD, ABE, ACD, ACE, ADE,
BCD, BCE, BDE, and CDE. Using the same method, one can select
a simple random sample of three cases from this population.
In principle, one can use this method for selecting random samples
from populations of any size, but in practice it could easily become a
lifetime occupation merely to list aU the combinations of the desired
number of cases. The same result is obtained by selecting each case
individually, using a list of random numbers such as may be found in
most textbooks of statistics. These are sets of numbers that after care-
ful examination have shown no evidences of systematic order. Before
using the table of random numbers, it is first necessary to number all
the elements in' the population to be studied. The table is then entered
at some random starting point (e.g., with a blind pencil stab at the
page.), ,and the, cases whose numbers come up as one moves from this
point down the column of numbers are taken into the ~ample-until the
"desired number of cases is obtainecL The-selection of any given case
places no limits on what other cases can be selected, thus making
equally possible the selection of anyone of the many possible combina-
9 In t,his illustration, each of the <liscs (i.e., each combination of t:vo cases)
has one, chance in ten of being selected. Each' of.' the individual cases also has the
same chance of being selected-four in ten, since each case appears on four of the
discs. There are, however, very many ways of giving each case the same chance of
being selected without getting a simple random sample. For example, s~ppose we
were arbitrarily to divide an illustrative population of ten cases into five pairs as
follows: AB, CD, EF, GH, IT. If we write the designations for these pairs on five
I discs, blindly pick one of the discs, and take as our sample the two cases designated
on this disc, then every case has one chance in five of being picked but, obviously,
not every possible combination has the same chance of being selected as every other
-in fad, most of the combinations (e.g., AC) have no chance at all, since they
have not been included on the discs.
PROBABILITY SAMPLING 523
tions of cases. This procedure is therefore equivalent to selecting
randomly one of the many possible combinations of cases.lO
Without going into the mathematical argument, it is possible only
to illustrate the underlying principles of probability sampling. Con-
sider, for this purpose, a hypothetical population of ten cases, as
follows:
Case: A BCDEFGHI I
Sex: F F F FFMMMMM
Age: Y OY OYOYOYO
Score: 0 I 2 3456789
The first five cases are females, the last five males; the cases designated
Yare younger and the O's are older. Age and sex will be considered
10 The procedure of selecting a random sample should not be confused with
the procedure of sampling from a list or a file of cases by taking every kth (for
example, every fourteenth or every sixty-third) case. The latter procedure is called
systematic sampling. Systematic samples may be either probability or nonprobability
samples, depending on how the first case is selected. Suppose one wants to select
every sixtieth case. To get a probability sample, the first case has to be selected
randomly from the first sixty, and every sixtieth case thereafter is selected. If the first
case is not selected randomly, the resulting sample is not a probability sample since
most of the cases have a zero probability of being included in the sample. Although
to the uninitiated systematic sampling seems to be the most natural and rational
way to go alJout sampling from a list, it involves complications not present in a
simple random sample. When the first case is drawn raIJdomly, in a systematic
sample, there is in advance no limitation on the chances of any given case to be
included in the sample. If we are selecting a sample of 100 cases from a population
of 6,000, before the first case is selected each case has one chance in sixty (100 in
6,000) of being included in the sample, whether we are using simple random or
systematic sampling. But in a systematic sample, once the first case is selected, the
cl:j.ances of other cases are altered. Suppose the first case drawn is #46. Selecting
every sixtieth case thereafter means that # 106, 166, 226, etc., will be drawn; the
cases between these numbers now have no chance of being included.
This means that a systematic sampling plan does not give all possible combina-
tions of cases the same chance of being included; only combinations of elements 60
cases apart in the list have any chance of being selected for the sample. The results
may be quite deceptive lif the cases in the list are arranged in some cyclical order.
Suppose, for example, that the 6,000 cases are houses 'in a community that was
built according to a systematic plan, and that they are listed in order of streets and
numbers. Corner house~ would then appear at regular intervals throughout the list;
say, the first house and every twentieth house thereafter is a corner dwelling. A
sample consisting of ca~es 1, 61, 121, etc., would be made up entirely of corner
houses; one consisting of cases 2, 62,122, etc., would contain no corner houses. But
corner houses are usua\!y larger and more expensive than those within the block,
and their occupants may accordingly differ systematically in certain characteristics.
Thus any sample made up entirely of corner houses, or entirely lacking in corner
houses, would give misleading results if the study concerned characteristics in which
occupants of the two types of dwellings differ.
524 AN INTRODUCTION TO SAMPI.ING
later, in relation to stratified samp1ing. The score represents some at

tribute of the individual, such as, his performance on a test of me-
chanical aptitude.
The mean score for this population of ten cases is 4.5. Assuming
that this were not known, the problem would be to make an estimate
of the population mean on the basis of the scores of the elements in
the samp1e that is drawn. According to the definition of simple random
sampling, the method of selecting the sample must give equal prob-
ability to every combination of the desired number of cases-in other
TABLE 1
MEAN SCORES OF SAMPLES FROM ILLUSTRATIVE
POPULATION OF TEN CASES WITH POPULATION MEAN SCORE
OF 4.5 (SIMPLE RANDOM SAMPLES)
Number of Samples
Samples of Samples of SampJesof
Sample Means l l 2 cases 4 cases 6 cases
.5 1
1.0 I
1.5-1.75 2 2
2.0-2.67 5 10 2
2.75-3.25 3 25 10
3.33-4.00 8 43 52
4.17-4.83 5 50 82
5.00-5.67 8 43 52
5.75-6.25 3 25 10
6.33-7.0 5 10 2
7.25-7.5 2 2
8.0 I
8.5 I
Total no. of samples 45 210 210
Mean of sample means 4.5 4.5 4.5
% of sample means
greater than 4.00
and less than 5.00 II 24 39
%of sample means
greater than 2.67
and less than 6.33 60 89 98
11 With the small number of different scores in the illustrative population,

there are only a limited number of possible sample means. Thus, fpr samples of
two cases, there is no combination that can yield a mean of 2.25; but there are
three samples of four cases (ABDF, ABCe, ACDE) with a mean of 2.25.
Similarly, a mean of 2.67 is not possible for any sample of two or four cases, but is
possible for one sample of six cases. For convenience of tabulation and in order to
help bring out the characteristics of the sampling distributions, the means of the
samples have been grouped.
words, over the long run, with repeated sampling, every combination
should come up the same number of times. We can, therefore, figure
out what will happen in the long run in our illustrative population by
the simple device of considering all the combinations. That is, we
take every combination of the desired number of cases and compute
a mean for each combination. What results is a distribution of sample
means-known as a sampling distribution. For example, there are 45
possible combinations of two cases in our hypothetical population of
ten cases. One, and only one, combination (cases A and B) will yield
a sample mean of .5; there are five combinations (A and 1, B and I,
C and H, D and G, E and F) that will yield sample means of 4.5; and
so on. Similarly, there are 210 possible samples of four cases. One of
these combinations (A, B, C, and D) will yield a sample mean of 1.5;
one (A, B, C, and E), a sample mean of 1.75; and so on.
Table 1 shows the sampling distributions for sample means based
on simple random samples of two, four, and six cases from our illustra-
tive population.
Notice that for samples of any given size the most likely sample
mean is the population mean;12 the next most likely are figures close to
'the population mean; the more a sample mean deviates from the
population mean, the less likely it is to occur. Also, the larger the
~sample the more likely is it that its mean will be <;lose to the popula-
tion mean.
It is this kind of behavior on the part of probability samples (not
only with respect to means, but also with respect to proportions and
other types of statistics) that makes it possible to estimate not only
the population characteristic (e.g., the mean) but also the likelihood
that the sample figure differs from the true population figure by any
given amount.
One interesting :feature of simple random sampling ought to be
mentioned, even though it is hard for most people to believe it without
mathematical proof. IWhen the population is large compared to the
sample size (say, mor than ten times as large), the variabilities of sam-
pling distributions are influenced much more by the absolute number
of cases in the samples than by the proportion of the population that is
I
12 Ths point is obscured in Table 1, for the case of samples of two, by the
grouping of means. Actually, there are five possible samples of two cases with
means of 4.5: there are four possible samples with mean of 4.0; etc.
included. That is, the magnitude of the errors that are likely depends
more on the absolute size of the sample than on the proportion of the
population that it includes. Thus, the estimation of popular pref-
erences in a national pre-election poll, within the limits of a given
margin of error, would not require a substantially larger sample than
the estimation of the preferences in anyone state where the issue is in
doubt. Conversely, it would take just about as large a sample to estimate
the preferences in one doubtful state with a given degree of accuracy
as it would to estimate the distribution of preferences in the entire
nation. This is true despite the fact that a sample of a few thousand
cases obviously includes a much larger proportion of the voters in one
state than the same-sized sample does of the voters in the nation.13
STRATIFIED RANDOM SAMPLES 14
In stratified random sampling, as in quota sampling, the popula-

tion is first divided into two or more strata. Again, the strata may be
based on a single criterion (e.g., sex, yielding the two strata of male and
female) or on a combination of two or more criteria (e.g., age and sex,
yielding strata such as males qnder 21, males.21 and over, females under
2V females II and .orr). :In stratified random sampling, a simple
13 For the benefit oli' those who may have some knowledge of analytical
statistics but who may nevertheless react with startled incredulity when explicitly
confronted with the principle of the indifference of sample statistics to the sam-
pling fraction (i.e., the proportion of the population included in a sample), it may
be pointed out that the sampling fraction is not even mentioned in the relevant
formulas given in most statistics textbooks. Thus, the familiar formula for the
standard error of the mean is uh/N, where u is the estimated standard deviation of
the population and N is the number of cases in the sample. This formula is derived
from the mathematics of simple random sampling and, as given, omits a term.
Correctly, the formula should be multiplied by y'(l-f), where f designates the
sampling fraction. Obviously, the smaller the value of f, the less difference this
, mu,ltiplier males. In sampling from an infinite population, t equals zero; in sampling
Ifr~:im a fini~e. population, it is never quite zero, but is generally too small to have
J any practical cpnsequences and may hence be disregarded. This is the principle dis-
cussed, in' the'text above. It should be remembered, however, that when a large
proportion of a population is being sampled, taking account of the sampling frac-
tion may considerably reduce the estimate of the probable margin of error of the
sample findings. When the population is small, one must include a large proportion
in the sample in order to achieve a small margin of error.
14 For reasons of simplicity of presentation, some points already made in the
discussion of quota sampling will be repeated here.
random sample is taken from each stratum, and the subsamples are
then joined to form the total sample.
To illustrate how stratified random sampling works, we may return
to the previously described popUlation of ten cases. Consider samples
of four with equal proportions of males and females (i.e., samples made
up by combining subsamples of two males with subsamples of two
females). To satisfy this last condition, many samples of four that were
possible under the conditions of simple random sampling are no longer
possible-for example, samples consisting of cases A, B, C, D or of
cases A, B, C, F or of cases D, F, G, I-because they do not have two
males and two females. In fact, there are now exactly 100 possible
samples as compared to the 210 previously possible. As before, we
have computed the mean score for each of the possible samples and
thereby obtained the sampling distribution of the mean. Table 2 com-
pares the sampling distributions for samples of four obtained on the
basis of simple random sampling, stratified sampling using sex as a
Table 2
MEAN SCORES OF SAMPLES OF FOUR CASES FROM ILLUSTRATNE
POPULATION OF TEN CASES WITH POPULATION
MEAN SCORE OF 4.5 (SIMPLE AND STRATIFIED
RANDOM SAMPLES)
Number of samples
Simple SampleS Samples
Sample Means15 random stratified stratified
samples by sex by age
1.50-1.75 2 1
2.00-2.50 10 7
2.75-3.25 25 3 8
3.50-4.00 43 25 26
4.25-4.75 50 44 16
5.00-5.50 43 25 26
5.75-6.25 25 3 8
6.50-7.00 10 7
7.25-7.50 2 1
Total no. of sa~lples 210 100 100
Mean of sample means 4.5 4.5 4.5
% of sample IIJeans
greater than l4.00
and less than 5.00 24 44 16
% of sample nieans
greater thaw 2.50
and less than 6.50 89 100 , 84
15 Again, the means of the samples have been grouped. See note to Table 1.
criterion for stratification, and stratified sampling using age as a

criterion.
It will be noted that there is a marked improvement over simple
random sampling when the sampling is based on a stratification of au!
hypothetical population by sex; with this kind of stratification we get a
marked increase in the number of samples that give means very close
to the population mean and a marked reduction in the number of
sample means that deviate widely from the population mean. When
the population is stratified by age, however, there is no such marked
improvement in the efficiency of sampling; in fact, the means of in-
dividual samples are somewhat less likely to be very close to the popula-
tion mean.
In general, stratification contributes to the efficiency of sampling
if it succeeds in establishing classes that are internally comparatively
homogeneous with respect to...~he characteristics being studied-that is,
if the differences between ciasses (e.g. between males and females)
are large in comparison with the variation within classes (e.g., among
the males and among the females). In our illustrative population, the
difference in scores between the sex groups is relatively large,
that between age groups relatively small; that is why stratification by
sex is effective in this case, stratification by age ineffective. The general
principle is that, if one has reason to believe that stratifying according
to a particular criterion or set of criteria will result in internally homo-
geneous strata, then it is desirable to stratify. If the process of breaking
/ the population down into strata that are likely to differ ~harply from
one another is costly, then one has to balance this cost against the cost
of a comparable gain in precision obtained by taking a larger ~imple
random sample. The issues involved in the decision whether to stratify
have, basically, nothing to do with trying to make the sample a .replica
of the popula~ion; they only have to do with the"anticipated homo-
geneity of the defined strata with respect to the char.tcteristics being
studied and the comparative costs of different methods of achieving
precision. Both simple and stratified random sampling involve repre-
s~ntative sampling plans.
. Except for a slight saving in arithmetic, there is no reason for
sampling from the different strata in the same proportion. That'is, even
with respect to the criteria selected for stratification, it is not necessary
that the sample reflect the composition of the population. Thus, in
sampling from a population in which the number of males equals the
number of females, it is permissible (and may sometimes be desirable)
to sample nine, or five, or two, or some other number of females to
every male. When this is done, however, it is necessary to make an
adjustment in order to find the mean score (or the proportion of
elements with a given characteristic, or whatever measure is desired)
for the sample which will be the best estimate of the mean score of the
total population of males and females. This is accomplished by "weight-
ing" the figure for each stratum in such a way that it contributes to the
score for the total sample in proportion to its size in the population,
as in the quota-sampling illustration on page 518. When the various
strata are sampled in constant proportion, one is spared this bit of
arithmetic since the various strata are already properly weighted.
There may be several reasons for sampling the various strata in
different proportions. Sometimes it is necessary to increase the propor-
tion sampled from classes having small numbers of cases in order to
guarantee that these classes are sampled at all. For example, if one
were planning a survey of retail sales volume in a given city in a given
month, simple random sampling of retail stores might not lead to an
accurate estimate of the total volume of sales, since a few very large
department stores account for an extremely large proportion of: tne
total sales, and there is no guarantee that any of these large stores wbul!d I'
turn up in a simple random sample. In this case, one would stratify tlle jl .
.population of stores in terms of some measure of their total volume of
sales (for example, the gross value of sales during the preceding year) .
Perhaps only the three largest department stores would be in the top-
most stratum. Th~ investigator would include all three of them in his
sample; in other words, he would take a 100 per cent sample of this
stratum. 16 Any other procedure in such a situation would greatly reduce
,
the accuracy of the estimate, no matter how carefully samples were
16 Note that in such a procedure, the cases in the total population do not all
have the same chanc:e of being included in the sample. Each of the three largest
stores has a 100 per I~ent chance of being included, whereas each of the stores in
anc-ther stratum may have only one chance in ten. But the probability of inclusion
of each case can be' specified, thus meeting the basic requirement for probability
'lampling (see pa~es 514-515).
taken from other strata. Again, of course, the figures from the various
strata would have to be appropriately weighted in estimating the total
volume of sales in the city.
Another reason for taking a larger proportion of cases from one
stratum than from others is that one may want to subdivide the cases
within each stratum for further analysis. Let us say that in our survey
of retail sales we want to be able to examine separately the volume of
sales made by food stores, by clothing stores, etc. Even though these
classifications are not taken into account in selecting the sample (i.e.,
the sample is not stratified on this basis), it is clear that one needs a
reasonable number of cases in each volume-of-sales stratum to make
possible. analysis of different types of stores within each stratum. If a
given stratum has relatively few cases, so that sampling in the propor-
tion used in other strata would not provide enough cases to serve as an
adequate basis for this further analysis, One may take a higher proPQr
tion of cases in this stratum.
One of the major reasons for varying the sampling proportions for
different strata cannot be fully explained without going into the mathe-
matical theory of sampling, but the principle involved can be under-
stood on a more or less intuitive basis. Consider two strata, one of
which is much more homogeneous with respect to the characteristics
being studied than the other. For a given degree of precision, it will
take a smaller number of cases to determine the state of affairs in the
first stratum than in the second. To take an extreme example: suppose
that there is reason to know that every case in a given stratum ha.:; the
same score; one could then determine how to represent that stratum
in the total sample on the basis of a sample of one case. Of COllrs_e, In
such an extreme case one IS not likely to have this information without
also knowing what the common score is. But in less extreme cases one
can often anticipate the relative degrees ofl homogeneity or hetero-
geneity of strata before carrying out the survey. For example, there may
be a great deal of experience to suggest that, with respect to certain
types of opinion questions, men will differ among themselves much
!pore than will wo~en; one would accordingly plan one's sample for
a survey of such opinions so as to provide for sampling a larger ,propor-
tion of men than of women. Because women may be expected to be
more alike than men in these matters, they do not have to be sampled
as thoroughly as do the men for a given degree of precision.
In general terms, one can expect the greatest precision if the
various strata are sampled proportionately to their relative variabilities
with respect to the characteristics under study rather than proportion-
ately to their relative sizes in the population. A special case of this
principle is that, in sampling to determine the proportion of cases
possessing a particular attribute, strata in which one can anticipate that
about half the cases will have the attribute and half will not should be
sampled more thoroughly than strata in which one would expect a
more uneven division. Tl:ms, in planning a stratified sample for predict-
ing a national election, using states as strata, one should not plan to
sample each state in proportion to its eligible population; it would be
wiser to sample most heavily in the most doubtful states.
One final point about stratified sampling: There may be reason to
believe that certain criteria will provide very effective bases for stratifica-
tion (ie., using these criteria, we would get strata which differ markedly
from one another), b':!t, as pointed out in the discussion of quota
sampling, the relevant data may become available only in the course
of the survey. In this case one cannot use the- ;'lteria in the sampling
design, but one can apply the logic of stratified sampling theory in the
analysis of the data. Thus, one can take a simple random sample,
ascertain the information necessary for stratification during the course
of the interviews, and use this information in grouping the cases
according to their respective strata and weighting them appropriately
in the analysis of the data.
For example, suppose that we want to survey the attitudes of the
students in a certain school toward some issue and that we have some
reason to believe that the proportions of "pro's," "anti's," and "un-
decideds" are likeiy to be different among the Negro and the white
students. Suppose,; further, that we have a complete listing of the stu-
dent body but no identification of the race of the individual students,
even though we know that 30 per cent of the students are Negro and
70 per cent are white.
I,
We could draw a simple random sample of the
students and ascertain the race of each respondent while recording his
views on the issu",. The data might then come out as follows:
No. in No. that No. in the Sample Who Are

the Turn up in uPro" "Anti" "Undecided"
School the Sample
Negroes 300 40 30 8 2
Whites 700 160 50 40 70
Total no. 1000 200 80 48 72
Percentage of
sample 40 24 36
Projecting from the sample to the total numbers in the two racial
groups, we would get:
Estimated Number \Vho Are
-------
uPro" "Anti" "Undecided"
Negroes 225 60 15
Whites 219 175 306
Total no. 444 235 321
Estimated percentage of total
student body 44.4 23.5 32.1
These figures are easily arrived at. Thus, three-fourths of the Negro
sample are "pro," and three-fourths of 300 is 225. In this case, the
corrected percentages are not dramatically djfferent from the total-
sample percentages, despite marked differences between the Negro
and white groups. This results from the facts that the disproportion
between the sampling fractions in the two groups is not very great
(13 per cent .of the Negroes are included in the sample as compared to
23 per cent of the whites) and that the Negro group constitutes a
relatively small proportion of this population. Despite the relatively
small differences in the results of the two procedures (i.e., the uncor-
rected and corrected percentage estimf-tes) in this example;the assured
precision is greater with the second.
From the viewpoint of the theory of probability sampling, 'it is
essentially irrelevant whether the stratification is introduced in the
sampling procedure or in the analysis of the data, except insofar as the
former makes it possible to control the size of. the sample obtained
from each stratum and thus to increase the efficiency of the sampling
design. It can be shown that any stratum of a 'simple random sample of
a population is itself a simple random sample of the corresponding
population stratum. lhus, not only is our total sample of the student
body a simple random sample of the total population, but the 40
Negroes in the sample are a simple random sample of all the Negroes in
the school, and the 160 whites are a simple random sample of the white
students. In other words, our procedure :of drawing a simple random
sample and then dividing ,it into strata is equivalent to having drawn a
stratified random sample using, as the sampling fraction within each
stratum, the proportion of that stratum that turned up in our simple
random sample. Thus, even though we were not in a position to
stratify in advance, we can take advantage of the increased efficiency of
stratified sampling.
CLUSTER SAMPLING
Except when dealing with small and spatially concentrated popula-

tions, there are enormous expenses associated with simple and stratified
random sampling-for example, in the preparation of classified lists of
population elements and in sending interviewers to scattered localities.
The more widely scattered the interviews, the greater are the travel
expenses, the greater is the proportion of nonproductive time spent
in traveling, and the more complicated-and hence expensive-are the
tasks of supervising the field staff. There are also other factors that oiten
make it difficult or impossible to satisfy the conditions of random
sampling. For example, it may be easier to get permission to administer
a questionnaire to three or four classes in a school than to administer
the same questionnaire to a much smaller sample selected on a simple
or stratified {pndom' basis; the latter may be much more disruptive of
the scho,?l routines. For such reasons, large-scale survey studies seldom
make use of simple or stratified random samples; instead they make use
,of the methods of cluster sampling.
In cluster sampling, one arrives at the ultimate set of elements to
be included in the sample by first sampling in terms of larger groupings
("clusters"). The ,.clusters are selected by simple or stratified random
sampling methods; and, if not all the elements in these clusters are to
be included in the sample, the ultimate selection from within the
clusters is also carried out on a simple or stratified random sampling
basis. .
Suppose, for example, that one wants to do a survey of seventh-
grade public-schopl children in some state. One may proceed as follows:
Prepare a list of school districts, classified perhaps by size of com-
munity, and select a simple or stratified random sample. For each of the
school districts included in the sample, list the schools and take a
simple or stratified random sample of them .. 1f some or all of the
schools thus selected for the sample have more seventh-grade classes
than can be studied, one may take a sample of these classes in each of
the schools. The survey instruments may then be administered to all
the children in these classes or, if it is desirable and administratively
feasible to do so, to a sample of the children.
Similarly, a survey of urban households may take a sample of
cities; within each city that is selected, a sample of districts; within
each selected district, a sample of households.
Characteristically, the procedure moves through a series of stages-
hence the common term, "multi-stage" sampling-from more inclusive
to less inclusive sampling units until one finally arrives at the popula-
tion elements that constitute the desired sample.
Notice that with this kind of sampling procedure it is no long~r
true that every combination of the desired number of elements in the
population (or in a given stratum) is equally likely to be selected as the
sample of the population (or stratum). Henc~, the kinds of effects we
noticed in our analysis of simple and stratified random sampling of our
hypothetical population of ten cases (the pop-ulation value being the
most probable sample result and larger deviations from the population
value being less probable than smaller ones) cannot develop in quite
the same way. Such effects do, however, occur in a more corriplicated
way,11 provided that each stage of cluster sampling is carried out on a
17 The complication arises from the fact that there are two sourc,es of sampling
error: the sampling of the larger sampling, units and the sampling of population'
elements within the larger units. To illustrate the point that cluster sampling does
have the same kinds of effects as simple and stratified random sampling, let us con-
sider the simple case in which the second source of error is eliminated by studying
all the population elements in the sampled larger units. Each larger unit has its score
(consisting, say, of, the mean score of its elelpents) . But this leaves us with a simple
or stratified random sample of the population of larger, units-no different, in prin
ciple, from a simple or stratified sample of population elements. Hence, it is clear
that the trends we noted in connection with random samples will tend to occur un
this level. Now, if instead of taking 100 per cent sa:mples of the elements in each
larger unit, we were to take a simple or stratified random sample of the elements in
each unit, the larger units become the populations from which these sampl~s are
drawn--and the tendencies we noted will again occur. Thus, these tendencies
to~rd the greatest projJability of achieving a sampling result that is the same as
the population value, and toward progressively larger deviations becoming progres-
sively less probable, will occur with respect to both sources of error that are involved
in cluster sampling.
PROBABILITY SAMPLING
probability-sampling basis. One pays a price, however, in terms of

sampling efficiency. On a per-case basis, effective cluster sampling is
much less efficient in obtaining information than comparably effective
stratified random sampling-that is, for a given numbex' of cases, the
probable margin of error is much larger in the former case than in the
latter. Moreover, the correct statistical handling of the data is apt to
be much more complicated. These handicaps are, however. more than
balanced by the associated economies, which generally permit the
sampling of a sufficiently larger number of cases ata smaller total cost.
The comparison of cluster sampling with simple random sampling is
somewhat more complicated. Stratified sampling principles may be
used to select the clusters and what is lost in efficiency because of the
clustering effects may be regained by this stratification. Depending on
the specific features of the sampling plan in relation to the object of
the survey, cluster sampling may be more or less efficient on a per-case
basis than simple random sampling. But again, even if more cases are
needed for the same level of accuracy, the associated economies gen-
erally favor cluster sampling in large-scale surveys.
Combinations of Probability and N onprobability Sampling

/
If sampling is carried out in a series of "stages," it is, of course,
possible to combine probability and nonprobability sampling in one
design. That is, one or more of the stages can be carried out according
to probability sampling principles and the balance by nonprobability
'principles. We shall consider two examples.
The investigator may select clusters by probability cluster sam
pIing techniques bpt, at the final stage, select the elements as a quota
sample. Thus, it is, possible to select a probability sample of counties
in a state; within e~ch of these counties, a probability sample of neigh-
borhoods; and within each of the selected neighborhoods, a quota
sample controlled tor, say, age and sex.
The advantage of such a design is that the major economies of
quota sampling occurL
in obtaining the particular cases for the sample.
It is relatively inexpensive to select the areas within which the final
stage of sampling will take place by probability sampling, and one
thereby gains the advantages of probability sampling, at least for the

areas. There is some evidence, for instance, that quota samples built up
in selected areas are more successful in controlling for such variables as
socioeconomic status than quota samples in which the control of these
variables depends on the judgments of the interviewers (Stephan and
McCarthy, 1958).
It may be remarked, in passing, that quota samples are, in practice,
nonprobability cluster samples. Our earlier description of quota sam-
pling may have made it appear to be the non probability analogue of
stratified random sampling; that is, it may have seemed that specified
proportions of cases with given characteristics were selected from the
total population. But in practice there are always restrictions on the
geographical areas within which the sampling takes place. Hence, the
traditional national quota sample has typically involved a nonprob-
ability sample of areas as well as a nonprobability sample of elements
within these areas. By using probability sampling to select the areas,
however, one can gain an extra measure of security at relatively little
cost.
The second example of combining probability and non probability
sampling involves the opposite strategy. The investigator takes a prob-
ability sample of elements within a non probability sample of areas. The
areas are selected as a purposive sample. For example, a number of
counties may be selected on the grounds that they have, for years,
tended to produce election results typical of their respective states;
within each of the "typical" counties, the investigator selects a prob-
ability sample of eligible voters. .
One way of looking at this kind of design is to regard the "typiC:J.I"
counties as defining a population. If a probability sample of this popula-
tion is taken, the mathematical theory of probability sampling is com-
pletely applicable, and one can state the probable limits of error in the
relation of the sample results to the true population values. One: can
then generalize the inferences regarding this restricted population..to
the national population, subject to the assumption that the "typicalf ,
counties are still typical of their respective states. So long as' this
ilssumption is val~d, it seems likely that such a sampling plan will
produce the most dependable sampling results at the least cost; but
then, of course, with each application of this sampling des'ign, one
PROBABILITY AND NONPROBABILITY SAMPLING 537
must hope that the assumption is in fact valid. The results of such a
sampling plan can, however, always be stated in a form that makes it
clear where the possibilities of error lie. Thus, a conclusion might read,
"There is a nineteen-to-one likelihood that from 60 to 74 per cent of
the eligible voters in these typical counties prefer Candidate A. If
nothing has happened to make these counties atypical in this election,
these results may be taken as reflecting the national distribution of
preferences. If the actual distribution of votes follows the distribution
of preferences at the time of the survey, it seems likely that Candidate
A will be elected." The election post-mortem can establish how well
the vote was predicted in the "typical" counties and whether they were
indeed still typical.
Special Applications of Nonprobability Sampling

It has already been noted that the major advantages of nonprob-
ability sampling are convenience and economy. It is likely, therefore,
that many future sampling operations will be conducted according to
nonprobability principles as long as researchers are convinced that
these sampling procedures work reasonably well, despite the fact that
they do not provide any basis for estimating how far the sample
results are likely to deviate from the true population figures. Investi-
gators, in other words, will continue to use nonprobability methods
and to justify their use on the grounds of practical experience, even
while conceding the superiority in principle of probability sampling.
Moreover, many practical samplers will argue that, in many cases at
least, this superiority exists only on paper. They will point out that
there is a differen;ce between the sampling plan and its actual execu-
tion; there can b~ many a slip in the carrying out of the plan which
would nullify its theoretical advantages. Interviews, for instance, may
fail to follow thei~ instructions in selecting respondents, or they may
omit some of thel questions in interviewing some of the respondents
(and, thereby, produce samples of somewhat different and not strictly
comparable populations
I,
in relation to the various questions in the
same interview schedule); some of the selected cases may refuse to be
interviewed or no:t be available; compromises may be made by allowing
interviewers to substitute other respondents when those designated fOl

the sample are not found at home;18 etc. The sample actually obtained
may, hence, not be the probability sample it was planned to be.
Moreover, there are circumstances in which probability sampling
is unnecessary or inappropriate. One such circumstance arises from the
fact that one does not necessarily carry out studies of samples only for
the purpose of being able to generalize to the populations that are being
sampled. If one uses samples for other reasons, ability to evaluate the
likelihood of deviations from the population values is irrelevant. For
example, during the discussion of "experience surveys" in Chapter 3,
it was pointed out that the goal is to obtain ideas, good insights, and
experienced critical appraisals. One selects the sample-a purposive
sample-with this in mind and not with the intention of assessing the
status of opinion among practitioners. The situation is almost exactly
analogous to one in which a number of expert consultants are callec;I in
on a difficult medical case. These consultants-also a purposive sample
-are not called in in order to get an average opinion that would corre-
18 It is sometimes claimed that one advantage- of quota over probability sam-

pling is that the former avoids the problem of refusals or unavailability. This is not
correct. To be sure, the procedures of quota sampling 'may-and commonly do-
by-pass the problem by ignoring such cases and allowing the interviewers to make
substitutions. Ignoring a problem is not equivalent, however, to solving it. The
existence of uninterviewable cases implies a restriction on the sampled population.
Probability samplers generally acknowledge the restriction; quota samplers generally
say nothing about it.
In probability sampling, when the restriction is accepted, the sound practice is
either to not replace the drop-outs at all, or to replace them by selecting new cases
by the same procedures that were used in selecting the original sample, rather than
by, say, taking the nearest at-home neighbpr of a not-at-home case. When the sound
practice is followed, the resulting sample is a probability sample of the restric,ted
population. This follows from the mathematically provable principle to which we
have already referred: any segment of a probability sample is itself a probability
sample of the corresponding segment of the population. If the restriction is not
accepted, the sound practice is, of course, to make intensive efforts to recover the
drop-outs.
Another expedient involves the use of data acquired in the course of the
survey to transcend the restriction (see Politz and Simmons, 1949, and Simmons,
1954). Briefly, the logic of the Politz and Simmons procedure may be described
as follows: Drop-outs resulting from the failure to find the selected respondent at
,home are consequential only if the probability of being at home has some, bearing
on the subject matter of the survey. Let us then find out from those who are at home
,what the probability would have been of the interviewer finding them at home if he
:had called at the same time of day during, say, the preceding five days. We can
then relate this probability to the responses to the survey questions and correct the
survey findings on the basis of this information.
APPLICATIONS OF NONPROBABILITY SAMPLING 539
spond to the average opinion of the entire medical profession. They are
called in precisely because of their special experience and competence.
Or the situation may be viewed as analogous to our more or less hap-
hazard sampling of foods from a famous cuisine. We are sampling, not
to estimate some population value, but to get some idea of the variety
of elements available in this population.
Another example of sampling for ideas rather than for the estima-
tion of population values is provided by the field of market research
known as motivation research. The typical problem of motivation
research is to find out something about motives, attitudes, associations,
etc., that are evoked by certain products, brand names, package designs,
etc., but that may not be obvious even to the respondents themselves.
The results of such studies are turned over to advertising agencies which
make use of them in developing advertising campaigns. Characteristi-
cally the motivation researchers are quite happy with accidental sam-
ples, or with purposive samples selected in such a way as to maximize
the likelihood of differences among the elements in the sample. They
are looking for ideas to transmit to the advertising men, not for correct
estimates of population distributions. One might argue that they
would be better off if they could establish, not merely the variety of
motives that are likely to become involved with certain products, but
also the precise distributions of these involvements. At present,
however, it seems to be problematical whether the additional informa-
tion would be worth the extra cost of getting it. At any rate, so long as
these researchers deceive neither themselves nor their clients into
believing that they are getting the second kind of information, no one
can take exception to their application of accidental sampling.
Sometimes there is no alternative to nonprobability sampling. If
one is trying to find out something, for example, about the attitudes
of people on the :other side of the "Iron Curtain," one has no realistic
choice but to rely on informants who have recently spent some time
there (each of whom
I
reports on the accidental sample involved in his
contacts), and on escapees, who are themselves far from typical. The
choice here.is between data that do not permit a statistical assessment
of the likelihoodlof error and no data at all. Similarly, if one is trying to
reconstruct a picture of a dying or recently deceased culture, one has no
choice except to rely on relatively articulate informants for certain
types of information. This does not mean that one is not concerned
with the possibility of error; but one places one's reliance on the in-
ternal consistency of the data and its coherence with other things that
one knows.
Another special case justifying the use of nonprobabiIity samples
arises from the fact that there are many important considerations in
research in addition to the sampling design. It may be necessary to
balance one consideration against another-for example, a better sam-
pling design against a more sensitive method of data collection.
Ackerman and Jahoda (1950), for example, studied the characteristics
of patients in psychoanalytic treatment who had given expression to
anti-Semitic sentiments. With complete protection of the anonymity
of the patients, some forty analysts served as informants. The sample
of psychoanalysts was, of necessity, an accidental one and, conse-
quently, so was the sample of patients. Suppose that the investigators.
could hayes,olved the problem of obtaining a probability sample of all
psychoanalytic patients in a given area, should they have done so?
Assume that this would have required giving l!P the psychoanalysts as
informants and substituting a relatively superficial direct interview.
Similarly, in a study of factors related to th<;: use of narcotics by
boys in juvenile street gangs, Chein and .his associates (see Wilner
et al., 1957) used group workers as informants (also with complete
protection of the anonymity of the individual gang member). These
workers had spent months winning the confidence of the boys, convinc-
ing the latter that they were not confederates of the police, ~ocial
reformers, or other things reprehensible in the eyes of the boys; and
they had been working closely with the gangs for many more months-
in some instances, for several years. Since these informants were avail-
able only for the gangs that were being worked with, the sample 6f
gangs-and hence of gang members-was an_accidental sample. ASsum-
ing that (1) it would have been possible to get a probability sample of
gang members and that (2) the information obtained through the
group workers was much more dependable than w01,lld have been
information obtained through direct interview, what should the in:vesti-
gators have done?
The answer to such a question is not easy. The first thing, to do,
of course, is to assure oneself that the dilemma is real. If convinced that
it is, one must then decide whether the problem is, under the circum-
stances, worthy of investigation at all. If the answer is still in the
affirmative, one must decide, in terms of the research purpose, whether
it would be better to gather more adequate information based on a
not very sound sample or less adequate information based on a sounder
sample.
We corne, finally, to another special and controversial case of non-
probability sampling. Many studies in behavioral science are carried
out on accidental samples of SUbjects. The data are treated, however,
in a manner that is appropriate only to probability samples. For ex-
ample, statistical tests of significance which presuppose random sam-
pling are applied to the data.
One justification of this practice is completely spurious. The in-
vestigators argue that they are interested not in estimating population
values, but in studying relationships among variables. For example, the
question, "What are the effects of variations in routines of memorizing
on the retention of the memorized materials?" does not seem to have
reference to any population. Relationships, however, are subject to
sampling error just as averages and proportions are. If a great many
samples are taken from a given population, certain relationships may
appear among some of the variables in some of the samples, and may
not appear, or may appear in different degree: in others. Hence the
results for a given sample may be-quite misleading. If the samples are
probability samples, we may legitimately estimate the probability of
being in error by more than a specified amount; if they are not, we
cannot legitimately make such estimates. Moreover, the answer to the
question may be quite different for different populations of subjects
(e.g., subjects differing in educational experience), for different popula-
tions of materials ito be memorized (e.g., nonsense syllables vs. mean
ingful poems), and for different populations of associated conditions
I
( e.g., presence or *bsence of distracting activities) . Relationships never

exist in a populatipn vacuum.
A second justification of the practice is more subtle. The investi-
gators, in effect, argue
t,
that they are not concerned with estimating true
population values (of means, proportions, differences under different
experimentll conditions, correlation coefficients. etc.) for any par-
does not. Does this not then point to the possibility of defining a more
inclusive population for which the relationship holds, and would we
not want to know the specifications of this more inclusive population?
Suppose, on the other hand, that it does hold only for the sophomores
at our particular college. Would we not then want to know what is
so unique about our population of sophomores? And would not the
tentative formulation of possible uniquenesses of our population sug-
gest hypotheses that we would want to explore-hypotheses that might
suggest population specifications that cut across our initial population
and that include elements not included in our initial population? In
either case, would we not want to press toward the discovery of the
specification of a population within which the trend that we have dis-
covered in our population to be statistically significant becomes a
virtual certainty?
It should perhaps be added that we are not, in these last few para-
graphs, preaching a paralyzing spirit of agnosticism that would pro-
hibit anyone from coming to any conclusions. The progress of science
and the scientific tenability of conclusions at-any point in time are,
after all, based on the coherence and consistency of many bits of
fallible evidence, the articulation of theory, and the interlocking of the
individually fallible bits of evidence with theory. It has been empha-
sized elsewhere in this book th<J.t science offers no possibilities of
absolute proof. The scientist can, at most, aspire to the soundest con-
clusions that can be reached in the light of the best evidence that can
be brought to bear on any issue. At the same time, science_would only
degenerate into dogma if one would not constantly remain alert to
the sources of ambiguity and fallibility in the available evidence and
the semantic gaps that may lie concealed in the ~eneralizations that are
drawn; if one would not attempt to. weigh the possible alternatives that
may be compatible with the evidence, particularly in the light of tpe
sources of fallibility and ambiguity; if one would not attempt to pin-
point the gaps in knowledge; and if, even though one has dismissed
some alternative on the ground that it is not sufficiently plausible to
merit serious consideration or dismissed some manifest gap in knowl-
edge as not sufficiently germane to merit intensive exploration, one
lVould not be constantly prepared to reopen these issues and remain
sensitive to the possibility of reopening them. In the light of these con-
siderations, what we have attempted to do in these last few paragraphs
is merely to look at a considerable body of contemporary research and
research practice in the perspective of sampling theory. If there is any
preachment implied, it is only another lesson in scientific humility.
AppendixC
QUESTIONNAIRE CONSTRUCTION AND
INTERVIEW PROCEDURE
BY ARTHUR KORNHAUSER AND PAUL B. SHEATSLEY
Outline of Procedures in Questionnaire Construction1
E ARLY IN PLANNING his research the investigator will weigh the

merits of several techniques for collecting the desired data and
decide whether to use a questionnaire2 or some other method. If he
chooses another method, he may still want to supplement it with a
questionnaire. In deciding which part of the research job can best be
handled by a questionnaire, the investigator must first discover the
extent to which the desired data are already available in census_ volumes,
in published or unpublished reports, or in collections of letters, diaries,
etc. Then he must decide wheth~r all or parts of the needed data can
best be obt~ined through a formal questionnaire-or through "depth"
interviews, long-continued case studies, standardized tests, refined
observations and/or experiment.
1 This section is a shortened and somewhat modified version of a discussion
by Arthur Kornhauser which appeared in Volume II of Research Metho,ds in Social
Relations, edited By Marie Jahoda, Morton Deutsch, and Stuart W. Cook (The
Dryden Press, Inc., 1951), pages 423-462.
2 The term questionnaire is used throughout this appendix to refer to stand-
ardized interview schedules as well as forms to be filled out by the respondent.
546
QUESTIONNAIRE CONSTRUCTION, INTERVIEW PROCEDURE 547
Let us assume, now, that the study director has decided to use a
questionnaire. The entire process of its construction can be divided
into the following six steps: deciding what information should be
sought, deciding what type of questionnaire should be used, writing a
first draft, re-examining and revising questions, pretesting, editing the
questionnaire, and specifying procedures for its use.
DECIDING WHAT INFORMATION SHOULD BE SOUGHT
The first step in the research procedure-formulating the precise

problem to be answered-provides the starting point for developing the
questionnaire. Suppose that the purpose of a study is to determine the
attitudes of the .public toward the hydrogen bomb. Preliminary con-
sideration of the problem may indicate the need for inquiry into beliefs
and opinions about war, relations with the Soviet Union, the respond-
ents' optimism or pessimism, their realism, fatalism, etc., as wen as
about the hydrogen bomb. The investigator must decide what aspects
of the problem are to be dealt with in the particular study.
An excellent test of one's performance in this stage of question-
naire construction, and at the same time a valuable aid, is the prepara-
tion of "dummy tables" showing the relationships that are anticipated.
By drawing up such tables in advance, the investjgator forces himse1f
to definite decisions about what data are required and how they will
be used. He can even enter figures representing different possible find-
ings, in order to visualize the bearing they would have on alternative
hypotheses and to see what new hypotheses they suggest.
The research planner may well run through the possibilities repre-
sented by the headings in a classification such as the following, stopping
at each point to decide what specific material his questionnaire should
seek in the light of the
I
specific purpose of his research.
Reports of "facts"
About the rdpondent:
I
Personal-history data, such as age, education, employment
Behavior elata, such as newspaper reading, radio listening,
church attendance,
I,
buying habits, voting behavior
About other persons known to respondent (family, em-
ployees, friends, etc. ) :
548 QUESTIONNAIRE CONSTRUCTION, INTERVIEW PROCEDURE
Personal-history and behavior data as above

About events and conditions known to respondent:
Reports of accidents, home conditions, job surroundings,
political meetings, wages received, etc.
Opinions, teelings, beliefs, etc.
"Reasons" for specified behavior and attitudes:
Objective factors: Influence of other persons, of conditions
and events, of published communications, etc.
Subjective factors: Specific wants, underlying desires and dis-
positions, evaluations, meanings, etc.
DECIDING WHAT TYPE OF QUESTIONNAIRE SHOULD BE USED
The appropriate form of question depends on the mode of admin-

istration, the subject matter, the sample of people to be reached (e_9uca-
tional and social level, etc.), and the kind of analysis and interpretation
intended.
Each class of questionnaire content may suggest two markedly dif-
ferent kinds of item-those that ask explicitly for the information
wanted, and those in which the desired information is interred from
responses directed to other matters. For example, instead of asking
the respondent directly about his own social adjustment, he may be
asked whether most people are hard to get along with. Questions of
"fact" are often asked not to obtain direct information about the
facts, which may already be known, but as indirect measure~ of knowl-
edge or interests. Opinions about an issue may be sought because of
research interest in the issue ("direct" questions) or in order to throw
light on the personalify of the respondent ("indirect" questions).
The investigator must also decide whether to use closed or open-
ended qu~stions. The relative advantages and disadvantages ofthe two
types of question, and the use for which each is most appropriate, have
been discussed in Chapter 7.
The use of follow-up questions or probes is advisable' at many
points in the ordinary interview, especially in connection with free
responses. The' questionnaire should anticipate where these are required
and should provide the appropriate wording. Although sometimes a
:lingle, fixed follow-up question can be specified, usually several altema-
tives are needed, depending upon the preceding response. For example,
if the answer is too general and indefinite, the follow-up may be: "In
what way?," "Just how do you mean?," "Can you give me an example?,"
etc. If the answer is incomplete, the questions may be: "Any other
reasons?," or "Would you tell me a little more about that?" Other
follow-ups ask: "What makes you think this?," or "What was there
about the picture that made you feel that way?," or "Where were you
working at that time?," and so on through an endless variety of ques-
tions needed to clarify or amplify the initial response. To the extent
that the improvisation of the follow-ups is left to the interviewer, to be
adapted to the specific responses, we depart from a standardized ques-
tionnaire or interview and border on a "partially structured" interview.
The questionnaire planner also has to decide whether to use a set
of several questions rather than a single question on particular points to
be covered. Several specific questions covering different aspects of a
topic often obtain more precise and useful information than does a
more general question, even if it is open-ended and accompanied by
follow-ups. (This point is discussed in more detail on pages 553-555).
FIRST DRAFT OF THE QUESTIONNAIRE
Probably the best way to begin is to outline or list the topics for
the questionnaire, consider carefully what is likely to be the best
sequence of topics (not the logical sequence, but the best psychological
sequence from the standpoint of the respondent), and then write the
q1.festions.
In addition to the questions deemed essential, the questionnaire
writer sometimes finds it wise to include a few extra ones aimed at
checking the reliability of responses or measuring the 'influence of
changes in wording.iFor example, two or more roughly equivalent or
closely related questions, well separated in the questionnaire, may be
asked in order to m~asure consistency of answers. The effect of dif-
ferent wording may be determined by constructing two parallel forms
of questionnaire ("split-ballot technique"), to be used with equivalent
samples of the popul~tion. The two forms have some of their questions
in common, but certain other questions are worded in different ways
in order that the effects of these_ differences may be measured.
At this stage of questionnaire construction, all available sugges-

tions should be utilized. Questionnaires that have previously been
drafted on the same or similar problems may prove most helpful. But
the wise questionnaire writer will think the questions over and strive
not only to improve them but to supplement or replace them in novel
ways whenever this can be done to advantage. There are few places in
soda] research where time-consuming, painstaking effort is more re-
warding than in the preparation of questions.
RE-EXAMINATION AND REVISION OF TIlE QUESTIONS
In the process of revision, it is invaluable to supplement one's own

efforts by the critical reactions of individuals who are familiar with
questionnaire methods and with the type of problem at hand. As far
as possible, the experts should represent different approaches and
reflect different social orientations. Few social research questionnaires
will fail to benefit from forthright criticism by persons with different
values and a different social outlook. In addition, the questionnaire
should be scrutinized for technical defects that may exist quite apart
from biases and blind spots due to personal values.
PRETESTING TIlE QUESTIONNAIRE
~'" The pretest is a try-out of the questionnaire to see how it works

, and "whether changes are necessary before the start of the full-scale
~tudy. The pretest provides a means of'catching and solving unforeseen
problems in the administration of the questionnaire, such as the
phrasing and sequence of questions, or its length. It may also indicate
the need for additional questions or the elimination of others.
In general, the pretest should .be in the form of personal interviews.
In mail or telephone surveys, the interviewing pretest can be followed
by a trial of the techniques actually to be used. The latter will detect
any further problems peculiar to the procedure, such as lack of adequate
instructions for filling out answers to questions. Ordinarily, if the pre-
ceding steps have been well performed, a few interviews suffice for the
pretest. However, it is sometimes necessary to do many interviews in
order to make sure that people differing in education, temperament,
and opinion will understand the questions and give complete and
pertinent answers. The people interviewed on the pretest should be
similar in characteristics to those who will be interviewed in the final
study.
Those who do the interviewing on the pretest must have a clear
understanding of the purpose of the study. They must be informed of
the over-all aim and the specific intent of each question, since they must
note whether the question is understood and answered by the respond-
ents in the manner intended. In conducting the interviews, they should
be alert to every reaction and comment of the respondent and should
record these verbatim. They should be instructed to try out alternative
wordings of questions that are not clear.
A valuable part of the pretest interview is discussion of the ques-
tions with respondents after they have answered them. The respondent
may be asked what the question meant to him, what difficulties he
experienced in replying, what further ideas he had that were not
brought out by the question, how he would ask the question, what hi~
feelings were on questions to which he responded, "Don't know," etc.
It is also important to have the interviewer record his own observa-
tions, criticisms, and suggestions. What difficulties did he encounter in
locating respondents and in interviewing them? What points seemed to
cause embarrassment or resistance? Where did he have trouble main-
taining rapport? Did the respondent become bored or impatient? On
what questions did the respondent request further explanation? Was
there enough space for recording answers? And so on.
If substantial changes are necessary, such as adding entirely new
questions, a second pretest should be conducted. Sometimes, in fact,
a series of three or four or even more revisions and pretestings is
required.
EDITING THE QUESTIONNAI~E AND

SPECIFYING PROCEDURES FOR ITS USE
After all the preceding steps have been completed, the question-
naire should be ready for use. All that remains is a final editing by the
research staff to ensure that every element passes inspection: the con-
tent, form, and sequence of questions; th<! :;pacing, arrangement, and

appearance of the material; and the spelling out in detail of procedures
for using the questionnaire.
The editorial job is directed primarily at making the questionnaire
as clear and easy to use as possible. Directions for printing or mimeo-
graphing layout should be given precisely, with emphasis on legibility,
ample space for replies, and the convenience of interviewers and re-
spondents in following questions and writing answers.
The questionnaire itself should contain simple, clear directions
telling the respondent (or interviewer) just what he is supposed to do-
which questions are to be answered only by certain classes of respond-
ents, which ones the interviewer is to explain, how fully and in what
terms the response is to be recorded, where a list is to be shown to the
respondent, and so on.
Guide tor Questionnaire ConstructionS
What follows is essentially a check list of points to consider in

formulating questions. The investigator may, for special reasons, decide
to depart from the rules at various places. But he should move ahead
cautiously ,with full awareness of what he is and is not doing. The most
troublesome errors in questionnaires do not arise from bad judgment
after due consideration of doubtful points; they creep in unwittingly,
even in "obviously simple" questions.
DECISIONS ABOUT QUESTION' CONTENT
Is this ,question necessary? Just how will it be useful?

Does the subject matter require a separate question, or can it be
integrated with other questions? . '
Is the point already sufficiently covered by other questions?
. ( Is the question unnecessarily detailed and specific for the purposes
of the study?
S For a more detailed and highly readable discussion, see Payne (1951).
GUIDE FOR QUESTIONNAIRE CONSTRUCTION 553
EXAMPLES: Instead of asking the age of each child in the family,
it is sometimes sufficient to get the number of children under 16.
Some studies of public opinion dispense with questions regarding
respondents' income, since other information serves well enough for
socioeconomic classification (rental level of district, occupation, etc.).
Are several questions needed on the sub;ect matter

of this question?
Should the question be subdivided?
EXAMPLES: Efforts to cut corners by combining two issues in one
question must be avoided-for example, asking for a single answer
about feelings toward Negroes and Jews; asking views about changes
in wages and hours, instead of separating the issues; about nationality
of "parents," instead of father and mother separately; etc.
Does the question adequately cover the ground intended?

EXAMPLES: If the investigator wants information about total family
income, a question about the respondent's "earnings" may be inade-
quate, since it probably will not lead the respondent to mention "other
income" or the earnings of others in the family.
A question simply asking people whether they are in favor of
having more educational radio programs on the air would not reveal
whether they wanted these for themselves or for the other fellow-
that is, whether they would listen.
Is additional related material needed to interpret the answers?

EXAMPLES: Often replies to questions about specific social reforms
I
can be understood better ~n the light of associated attitudes elicited by
other questions-for example, questions that ascertain broad attitudes
toward security and opportunity for the poor, antagonisms toward
centralized authority, beli~f in need for social change, etc.
A question asking for the respondent's opinions about the char-
acteristics of a particular racial group calls for parallel questions about
other groups, in order to:,determine whether his opinions are specific
to that group or reflect a more general view of out-groups or even of
people in general.
In opinion questions, is further information needed about the

intensity of the respondent's conviction or feeling?
EXAMPLES: In connection with questions on stereotypes, it may be
important to find out not only whether the respondent has certain
stereotypes but also what feeling tone, if any, is attached to these
stereotypes. A person may think that the members of a certain group
are "clannish," but he may feel that this is reprehensible, or admirable,
or he may be indifferent about it. He may consider this trait admirable
in one group and reprehensible in another.
Techniques for ascertaining intensity of opinion include "feeling
thermometers," on which the respondent rates directly that he feels
"very strongly," "fairly strongly," etc.; scales running from one extreme
of intensity to the other (e.g., from "Don't care whether I vote for the
candidate" to "Positively going to vote for him, even if I have to get
out of a sickbed"); asking about the respondent's behavior (e.g., ~how
much he has discussed the question or read about it, whether he has
joined organizations, written to newspapers or congressmen, etc.).
Another procedure is to call attentio!1 to difficulties or sacrifices
entailed in the respondent's position (e.g., higher taxes) to see whether
and how this changes his opinion. Similarly, one may mention that
many people hold a view opposite to his and then ask whether he thinks
these people may be right.
One may ask the respondent to specify what sacrifices or penalties
he would be willing to undergo if he could thus guarantee the outcome
he wants-for example, how much money he is willing to pay a year to
get television programs without advertising.
I
Is further information needed about how important the respond-

ent considers the condition orissue asked ab.out?
EXAMPLES: Free-answer questions may help to determine how
significant the point in question is for the respondent, how much he
emphasizes personal consequences, what concern he manifests over
the social effects, etc. For example: What if there is another world
war? What difference does it make if labor unions have (oriare kept
! from having) a closed shop?
The respondent can be asked to give direct ratings of the relative
importance of the issue in question compared with other issues, or to
indicate which of different possible consequences he considers probable
and whether a given condition "matters much" to him or not. For
example, does he think a Republican victory in the next election will
have important consequences, either good or bad; and what possible
effects, as given in a check list, does he consider most important?
Do respondents have the information necessary

to answer the question?
Is it a matter they can report on adequately?

Does the question call for answers the respondent either cannot
give at all or cannot give reliably?
Is the point within the respondent's experience?
Is it too remote or non vivid or difficult a memory?
Is it unanalyzed or unverbalized experience?
Is it subject to serious errors of observation and/or recall?
Does the question ask for opinions on matters so unfamiliar to
the respondent that the opinion does not mean what it seems to?
Can the necessary background information be given to the re-
spondent in the course of the questionnaire or interview, or should
the question not be asked at all of respondents lacking the information?
I
If the question attempts to supply the needed background infor-
mation, does it give an adequate and unprejudiced statement?
EXAMPLES: Does the question concern childhood behavior and
attitudes, incidental recent experiences of no special interest to the
respondent, or unanalyzed and inarticulate experiences? If so, and if
the purpose is to obtain objective information rather than the respond-
ent's subjective impression, official records, refined observations, and
experiments should b~ used when possible. \\Then questioning is used,
the respondent should be aided in ways suggested on pages 556-557.
An opinion-poll question asking about who is to blame for a list of
specific strikes found.'that from one half to four fifths of the respond-
ents did not know enough about these strikes to express an opinion. In
similar questions th~t did not directly ask whether people had the
information, almost all the respondents expressed opinions regardless
.Jf their probable lack of information.
556 QUESTIONNAIRE CONSTRUCTION, INTERVIEW PROCEDl!RE
Can the desired information be supplied better by specific re-

spondents other than the one first approached?
EXAMPLES: A mother may be able to report what books her
youngster reads, but a child himself must be questioned to find out
how he feels about reading these books.
Among people who were known to have redeemed war bonds,
many denied that bonds had been cashed. Since the interviews were
with either husband or wife, one cannot know whether the discrepancy
was due to suppressed information or to lack of knowledge concerning
the spouse's sale of bonds.
It is important in such instances to arrange to have the appropriate
persons answer each set of questions-by using separate schedules, by
having ip.terviewers see the different individuals, by specifically request-
ing that the individual who receives the questionnaire have his col-
leaguee answer certain of the questions, and so on.
Are alternative questions required on this subject matter to adapt

it to different classes of respondents?
EXAMPLES: In radio research, those who listen to a radio program
"regularly," "occasionally," or "never" have to be asked different sets
of questions. The regular listeners may be queried about their attitudes
toward the program; the occasional listeners may be asked, in addition,
their reasons for not listening more often; the questions to nonlisteners
ask whether they listen to other programs, whether they know of this
program, etc.
Does the question need to be more concrete, specific, anq

closely related to the respondent's personal experience?
Is it asked in too general a form?
Can the information be obtained mqre easily by referring it, more
dosely to the respondent's own behavior?
Does it utilize natural psychological aids to recall, such as. having
the respondent recall experiences in temporal sequence-working back
! from the present, or working up to the present from a specified time
in the past?
EXAMPLES: It is often effective to use behavior indicators instead
of, or along with, subjective expressions of feeling. For example, one
can ask not only, "How well did you like the book?," but also "Have you
recommended it to anyone else? Have you looked for other books by
the same author?," etc.
An interview schedule used in an inquiry into women's morning
radio listening, instead of asking what hours they usually listened and
to what programs, first obtained a list of what they had listened to
today, then the same for yesterday-and only then, as a final step, asked
what programs they usual1y listened to during the morning hours.
Is the question content sufficiently general and tree trom

spurious concreteness and specificity?
Ordinarily, the danger lies in questions that are too general; but
on occasion the reverse is true, and a highly specific question is im-
properly used to tap general attitudes or to ascertain over-all facts.
Is the subject matter such that a specific question may elicit
inaccurate or misleading responses?
EXAMPLE: At times, specific and recent instances may be unrepre-
sentative. Since many activities vary seasonally, questions about what
the respondent did today or last week may elicit a far less accurate
report of his general activity than a question about hts usual or average
behavior during previous months (for example, amount of television
viewing, miles of driving, or kind of breakfast food eaten).
Do the replies express general attitudes and only seem to

be as specific as they sound?
EXAMPLE: This faQIt occurs most often in attitude or opinion ques-
tions. Suppose an investigator concerned with opinions about racial
policies in a school sy~tem asks: "Do you feel that qualified Negro
teachers have just as good a chance as qualified white teachers to be
hired in the schools in this city?" Many people answer such a question
on the basis of an assumption that, in general, Negroes are (or are not)
treated fairly, rather than on the basis of any specific knowledge or
opinion about practicds within the school system. Thus, although the
answers may seem to refer to the particular issue, this may be deceptive
In order to be sure, one must ask either a number of questions referring

to different specific situations, or a general question in addition to the
one concerning the specific situation in which he is interested.
Is the question content biased or loaded in one direction,

without accompanying questions to balance the emphasis?
Is the question unfair in any way? Would the content be accepted
as fair by informed persons with opposite views on the point under
inquiry?
Is it likely to obtain answers that will unduly favor one side of the
issue?
Does the question introduce unwarranted assumptions about the
subject matter?
EXAMPLES: Opinion-poll questions often inquire into negativ~ and
vulnerable aspects of labor unions without accompanying questions
about positive features, and without mentioning parallel negative
aspects of business practices. By dweIIlng ~n what is wrong with unions,
thinking is directed disproportionately to their disapproved features
and an unfair picture is obtained.
Many single questions likewise reflect biased selection of subject
matter; for example, a question about who pays the advertising costs of
consumer goods, with no mention of the possibility that advertising
makes it possible to sell the merchandise at a lower price.
Question content sometimes carries a bias simply beca.use of the
timing of the question. If a community has just experienced a race riot,
questions on ending discrimination will give biased results. The same is
true of questions on labor relations or international relations following
upon some. favorable or unfavorable event. Seasonal influences likewise
may seriously bias answers to questions; for example, asking about
favorite sports or recreational activities during the height of the foot-
ball season.
WiII the ,respondents give the information that is asked for?

Is the material too private, of an embarrassing nature, or other-
wise likely to lead to resistance, evasion or deception?
What objection might a person have to answering?
Does the question "put him on the spot" or make him feel he is
being quizzed?
Can one get the information in a manner which would not offend,
or should it be omitted?
Do any special conditions exist at the time and place of the survey
to augment suspicion or resistance?
EXAMPLES: There are obviously matters which people are reluctant
to disclose; for example, family quarrels, receipt of charity, one's own
limitations and difficulties, antisocial attitudes such as race and reli-
gious prejudices, "inside" information about a political or religious
organi2!ation or labor union, etc. Such topics must either be avoided
in interviews and questionnaires or forms of inquiry must be employed
that will elicit the desired information without unduly disturbing the
-respondent.
One technique is that used to ascertain whether people read pulp
magazines. The interviewer offered to buy any old magazines that
might be lying around the house. This resulted in the report of a con-
siderably larger number of low-prestige magazines than had been
found by direct inquiry.
Much of the desired information, however, is subjective and
essentially personal; it can be obtained only by questioning the re-
spondent. Here the informal, free-answer type of interview sometimes
succeeds where formal questions do not.
Special types of indirect and projective interview questions are
sometimes used. Instead of asking a person how much he donated in a
charity drive, why he gave, and why he did not give more, such ques-
tions might be made to refer to "people you know" or "people like
you." This type of ~uestion assumes that what the person says about
others is likely to be a projection of what is true of himself-a some-
times useful but not!altogether safe basis for interpretation. Sometimes,
a personal question 'will be answered more frankly if it follows a par-
allel indirect question about "most people": "And how do you your-
self feel about this?,~4
Even if direct ~uestions are asked, skilled interviewers encounter
4 See Chapters 7 and 8 for a more detailed discussion of questions of this type.
remarkably few refusals to reply. A great deal depends upon the inter-
viewer's own attitude. If he is embarrassed or feels that the question is
too personal, his doubts are readily transmitted to the respondent. If
he confidently expects a reply, he is likely to get it.
Is the question likely to encounter emotional influence~ and de-

sires that will lead to falsification of answers?
EXAMPLES: Many people answer even factual questions in a way
that tends to exaggerate such matters as their income, education, and
social status; reduces their actual age; ennobles their acts and motives.
One can also expect some replies to be thrown off by emotional
influences that tend to exaggerate or minimize hardships. A pessimistic
tendency has, for example, been noted in farmers' estimates of current
crop damages. Where the content of a study is likely to suffer from such
distortions, effort should be made to check or corroborate the informa-
tion by different approaches to the matter within the questionnaire
and by reference to outside sources when possible. Information about
school and employment, for example, can 'be checked by seeing that
dates and periods of time are consistent, by inquiring into courses of
study and details of jobs, and by similar cross-checking techniques.
In addition, when justifiable, efforts can be made to check with school
authorities, previous employers, etc.
H the question content gives a clue as to who is sponsoring the
study or what the purpose is, respondents often feel disposed to ex-
press appropriately favorable feelings, or at least, in a spirit of polite-
ness, to withhold negative expressidns. Hence the general policy of l_1ot
revealing the sponsor or_the specific purpose of the questions.
. Another especially interesting influence. to be mentioned here is
the inclination of respondents ,at I times to answer "for the record."
Questions about labor relations are likely tb be answered by both in-
dustrial and labor leaders with one eye on the possihle effects of the
opinion study. This "public-relations bias" may also affect th ordi-
nary citizen. For example, when questioned about whether religion is
! declining in influence or whether religious or racial hatred is increasing,
. many respondents hesitate to acknowledge what they consider an evil

trend lest they give comfort and encouragement to the "enemy."
In the examination of prospective questions, such influences have
to be considered and, as far as is feasible, guarded against. The decision
may be to omit the questions or to adopt indirect or free-answer meth-
ods or some other form of inquiry.
DECISIONS ABOUT QUESTION WORDING
Can the question be misunderstood? Does it contain difficult

or unclear phraseology?
Are the words simple enough for the least educated respondent?
Are any terms used in a specialized way and, if so, is the meaning
made clear by pictures or otherwise?
Is the sentence structure short and simple? Is there any looseness
or ambiguity? What else could the question mean to a respondent?
Is the meaning clearly distinguished from other ideas the respond-
ent milY think the question asks-ideas that may seem more natural
or important to him?
Could unintended emphasis on a word or phrase change the ques-
tion meaning?
EXAMPLES: Questions which ask for "nationality," "occupation,"
"marital status," and many other items of personal data often cause
trouble unless spelled out in detail.
sUch terms as "compulsory arbitration," "T.V.A.," "C.I.O.,"
"N.A.M.," "real wages," "closed shop," "advertising medium," and
such academic expressions as "exclusion from employment" are en
countered in questionnaires and interviews even though they conve~
no clear meaning to large sections of the American public.
Simple familiar words are often employed in a vague, amqiguom
manner. For example, the simple question, "What kind of headache
remedy do you usu~IIy
, use?" proved ambiguous, since to some respond-
ents "kind" signified brand whereas to others it meant tablets vs.
powder. Or, what does "Do you usually use pancake make-up?" mean?
Every day? "Usually" whenever you use any make-up? "Usually" for
special occasions? ~tc.
Long or difficult sentences are likely to be misunderstood by many
respondents.
Sometimes, accenting particular words changes the meaning of a

question. An unintended emphasis on the word "should" in the ques-
tion, "Do you think the United States should accept Russian influence
in Eastern Europe or not?" introduces a moral note and might well
cause some respondents to answer in the negative, meaning we should
not (although in the world of practical politics, it is the thing to do).
A slight change in wording clearly reduces the likelihood of the moral-
istic accent on the word "should": "Do you think the United States
should or should not accept Russian influence in Eastern Europe?"
If any suspicion of misunderstanding remains after questions
have been worded as clearly as possible, good practice calls for follow-
up questions by the interviewer to determine just what the respondent
meant. The interviewer is instructed to ask such additional questions
as, "Just how do you mean that? ... Would you tell me a little more
about what you have in mind there? ... Will you give me al1. ex-
ample of what you mean?"
Does the question adequately express the alternatives with

respect to the point?
EXAMPLES:The safe rule is to make each of two alternatives ex-

plicit whenever there is the slightest ambiguity about the second alter-
native. Even when the implication is reasonably clear, a statement of
the second alternative may serve to make it more vivid than it would
otherwise be and hence to place the two possible answers 'on a more
nearly equal footing ..Payne (1951) gives a striking example of the
effect of failing to state alternatives explicitly. To the question, ~'Do
you think most manufacturing companies that layoff workers during
slack periods could arrange things,to avoid layoffs and give steady work
right through the year?," 63 per cent said, companies could avoid lay-
offs, 22 per cent said they could not, and 15 per cent expressed no
opinion. A carefully matched sample of respondents was asked" the
question with an alternative explicitly stated: "Do you think most
I manufacturing c,ompanies that layoff workers in slack periods could
. avoid layoffs and provide steady work right through the year" or do you
think layoffs are unavoidable?'; When the question was asked this way,
35 per cent said companies could avoid layoffs, 41 per cent said layoffs
a:re unavoidable, and 24 per cent expressed no opinion.
Is the question misleading because of unstated assumptions

or unseen implications?
Is the frame of reference clear and uniform for all respondents?
Does the question bring out the basis for the respondent's reply-
the frame of reference within which he is answering?
What consequences of the proposed action does he see?
Does the question distinguish between what he wishes to have
true and what he thinks is true?
EXAMPLES: Many questions can be answered only on the assump-
tion that certain other things are true~ or the respondent may make
certain assumptions in his answer. Unless questions designed to probe
these assumptions are also included, the replies are subject to grave
misinterpretation. The following are illustrations:
"Do you think the government should put a ceiling over wages?"
(1942). Of people who answered in the affirmative, many doubtless
took it for granted that prices were likewise to be controlled; perhaps
also profits. However, since there were no follow-up questions to
ascertain how extensively such assumptions were involved, answers
could not be interpreted.
"To which of these groups do you feel you belong-the white-collar

class, the working class, or some other class?" This question assumes
that the person feels he belongs to a "class" based on occupation.
One $ould first ?etermine whether he feels any class identification
and, if so, how he,thinks of "class."
When questions are asked of the general public-whether on the

veto power in the United Nations, the continued manufacture of the
hydrogen bomb, or, an anti-strike law-the investigator must consider
the degree to whicp the implications of the action are seen, how far
the opinions are offhand and superficial, and how far they reflect crys-
tallized views based on public discussion. It is important to ask sup-
plementary questions to ascertain the considerations the respondent

has in mind in answering as he does. Free-answer questions may be
used which ask for reasons, advantages of the course advocated, and
arguments against it; or questions may ask the respondent to react to
specific arguments for and against the measure; or questions of the
following type may be used: "Are you in favor of (or opposed to) doing
this even if it means so and so?"
Is the wording biased? Is it emotionaIIy loaded or slanted

toward a particular kind of answer?
Does it employ stereotypes? Does it contain prestige-carrying

names? Does it employ superlative terms which push the answer one
way or the other? (If such elements of bias are present, are they there
intentionally-and does the research purpose justify their inclusion?)
Does the question tend to elicit replies that are more biased than
those the respondents would give on the same point if they had an
opportunity to answer freely and fully?
Would the wording be acceptable to persons with opposite views
on the matter?
EXAMPLES: Two parallel opinion questions in 1939 asked whether
the United States was likely to be involved in the war. The results
show how greatly a difference in wording can affect responses when
opinion is confused and undecided. One form asked, "Do you think
the United States will go into the war before it is over?" The replies
were: "Yes," 41 per cent; "No," 33 per cent; "Don't know," 26 per
cent. The:ojher form asked, '.'Do you think the United States will suc-
ceed in staying out of the war?" The replies were: "Yes," 44 per cent;
"No," 30 per cent; "Don't know," 26 per cent. (Cantril, 1944.)
The influence of President Roosevelt's name caused an increase
of about 5 per cent in the affirmative replies to the second of the fol-
lowing two questions:
"Do you like the idea of having Thanksgiving a week earlier this
year?"
\"Do you like Pre~ident Roosevelt's idea of having Thanksgiving
a week earlier this year?"
Is the question wording likely to be ob;ectionable to the
respondent in any way?
(See also pages 559-561.)
EXAMPLES: Instead of asking the respondent directly what the
amount of his income is, a question such as the following makes it
easier to obtain the information:
For the purpose of our survey, we need to have a rough indication
of the income of your family. Would you mind telling me in which of
these classes it falls:
Below $1,000 a year From $3,000 to $4,000
From $1,000 to $2,000 From $4,000 to $5,000
From $2,000 to $3,000 Over $5,000 a year
The same use of lists and classifications may aid in questions about
age, religion, etc.
A change in words may make a question more palatable. A ques-
tion asking whether the respondent has "dealt on the black market"
can be rephrased to ask whether he has "paid over-ceiling prices." In-
stead of "Did you graduate from high school?" the question can ask,
"What grade were you in when you left school?" A good general rule
is to avoid placing the respondent on the defensive; always leave him
an easy "out." /
Would a more personalized or less personalized wording

of the question produce better results?
I
#EXAMPLES: The investigator must judge in each instance whether a
personal or an impersonal question will produce better results for the
purposes of the particular survey. For example, a question may be
asked in such different forms as the following:
"Are working cQnditions satisfactory or not satisfactory in the
plant where you work?"
"Do you feel th~t working conditions are satisfactory or not satis-
factory in the plant where you work?"
"Are you personally satisfied or dissatisfied with working condi-
tions in the plant where you work?"
The most personal form of these questions probably elicits a more
566 QUESTIONNAIRE CONSTRUCTION, INTERVIEW PRODECURE
individual expression of feelings; the most impersonal, a judgment

more tempered by what the respondent supposes other people think,
or what he thinks the objective realities may demand. On the other
hand, the impersonal form may, at times, prevent embarrassment and
lead to franker replies.
Can the question b~ better asked in a more direct or a more

indirect form?
EXAMPLES: Using the "indirect" procedure, questions are asked,
for example, about how radical or conservative certain magazines or
men in public life are. The respondent who calls them all conservative
indirectly reveals his own "radicalism," and conversely with one who
rates them all as radical.
Another type of indirect (or quasi-indirect) questioning is well
illustrated in magazine-audience measurement. When respondents are
asked directly to identify the articles and advertisements they have
seen in a magazine, serious errors arise through false identifications
occasioned by confusio!! and tendencies to exaggerate. Instead of ask-
ing the respondent directly to recall whether he has seen the pages,
he is asked merely to say whether each item "looks interesting." This
'is followed by the question: "Is this the first time you have seen it?"
By first focusing attention on interest, more satisfactory answers are
secured regarding ads previously seen.
Other examples have to do with information and views which the
respondent may be reluctant to give if he is asked about them directly.
Veterans who evade (and resent) direct questions about their battle
experiences sometimes report these memories with genuine interest if
the questioner comes at them obliquely-for example, by asking how
well army equipment held up in the field, whether enlisted men were
sufficiently trained before going into actiori, etc.
No general recommendations are warranted regarding the use of
indirect questions. Their possibilities and limitations have to be ex-
amined in each new setting and with the research purposes clearly in
mind. Both technical and moral issues are involved: Whether the
intended inferences can safely be drawn from indirect evidence;
whether the particular type of indirection biases the response; whether
the questions arouse suspicion of trickery or surreptitious prying-and
whether such suspicion is justified.1I
DECISIONS ABOUT FORM OF RESPONSE TO TIlE QUESTION
Can the question best be asked in a form calling for check

answer (or short answer of a word or two, or a number),
free answer, or check answer with follow-up free answer?
EXAMPLES: When check answers are used, the alternatives are
sometimes included as part of the question; sometimes they follow the
question and serve merely for the convenience and guidance of the
interviewer in recording responses. Ideally, the latter use of check an-
swers should not have any effect on the responses; but in actual practice,.
it does exert a substantial influence. This may come about either-
through the respondent's seeing the printed blank or because the in-
terviewer is influenced by the categories in his follow-up questioning
and in his recording of answers. When the alternatives are contained
in the question itself, they naturally exercise a more decided constrain-
ing influence.
Best results are often achieved by means of a combination of free-
answer and check-answer methods or by compr~mises between them.
Thus, single questions may consist of an initial check-answer part
immediately followed by free-answer parts which inquire into the
meaning of the check response and obtain examples, statements of the
respondent's assumptions, the strength of his feelings on the point.
etc. For example:
"Are people at the head of your company interested in their em-
ployees, or donrt they care?" Interested--; Don't care-.-;
Other answer--;- "In what ways do they show that?"
If a check answer is used, which is the best type for this

question-dichotomous, multiple-choice ("cafeteria"
question), or scale?
EXAMPLES: Tl;le simple yes-no response (and similar dichotomous.
choices such as agree-dis.agree, do or do not) are appropriate for many
II For a more detailed discussion of indirect questions, see Chapter 8.
568 QU~STIONNAIRE CONSTRUCTION, INTERVIEW PRODECURE
questions dealing with points of fact and with issues which are clear-cut
and on which well-crystalli~ed views are held. Even on direct yes-no
questions and simple two-way ct;>mparisons, however, it is usually found
desirable to include an intermediate response of "doubtful," "un-
decided," "same," "no difference," "both," etc. The inclusion of such
responses is sometimes considered inadvisable because it provides too
easy and attractive an escape for respondents who are disinclined to
express a defini~e view. On the other hand, forcing replies into two
extreme categories (particularly if the interviewer is not even supplied
with a space for checking an intermediate answer) is likely to cause
difficulty for many respondents and to yield results that are less real-
istic and more misleading than is true when an intermediate reply is
provided for.
A graded series of response possibilities frequently gives the in-
vestigator additional or more accurate information than a dichotomous
response and presents the question more adequately and acceptably
to the respondent. The most common multiple-choice responses utilize
three, four, or five gradations, but larger numbers are sometimes appro-
priate. A widely used special form of multiple choice (the so-called
"cafeteria" question) asks the respondent to choose from a list of
assorted words or statemtnts one or more that best represent his own
view. The items mayor may not be arranged in order from high to low,
good to bad, favorable to unfavorable, etc., to constitute a crude scale.
Examples of several of these forms follow.
Suppose a person is asked by a ~lose friend to do something that

requires a personal sacrifice. In your country, how strong an obli-
gation would a person of your age feel to help his friend?
- - not strong at all
- - not very strong
- - rather strong
- - very strong
- - extremely strong
Put a "I" in fro,nt of the thing that is most important to have or

do in order to get ahead in the world. Put a "2" before the next
most important, etc.
GUIDE FOR QUESTIONNAIRE COI'fSTRUCTION 569
--pull _-brains
--goodluck _ _ hard work
Of course, you want ALL these things-but which will influence

you most when it comes to choosing your next car? Check three
items.
- - Appearance _ _ Operating Economy
_ _, Comfort _ _ Pick-up
- - Dependability _ _ Safety
- - Ease of Control _ _ Smoothness
- - First Cost _ _ Speed
__ ........ 7
If a check list is used, does it cover adequately all the

signi5cant alternatives without overlapping and in a
defensible order? Is it of reasonable length? Is
the wording of items impartial and balanced?
EXAMPLES: Poor check lists constitute one of the most common

faults in questionnaire construction. The "cafeteria" type of question
seems to offer special temptations to careless listing of miscellaneous
alternatives. /
Many check lists consist of loose and ambiguous qualitative
terms-usuaIly, sometimes, rarely, frequently, occasionaIly, good, fair,
poor. The difficulty with these terms is that respondents have different
'standards in mind, so that two persons may report the same facts under
different categories. When feasible, it is better to use concrete and
objective terms for:the different degrees.
Another freqlfent fault of check lists is that they are incomplete.
As noted in Chap~er 7, several studies have shown that replies may be
seriously changed py the omission of one or more significant alterna-
tives from the che~k list. Many check lists are defective because they
contain alternatives that appear equally true or that are not mutually
exclusive; items that do not fit into the continuum formed by the
I.
other items (the list does not remain in one dimension); items that
are loaded in worqing. ambiguous, or too extreme; items that contaiu
more than one idea; items that influence responses by being over-
specific; items that overrepresent or underrepresent one side of an
issue. The following examples illustrate some of these faults:
Which one of these four statements comes closest to what you

yourself think about advertising on television?
3. I'm in favor of advertising on television because it tells me
about the things I want to buy.
b. I don't particularly mind advertising on television. It doesn't
interfere too much with my enjoyment of the programs.
c. I don't like advertising on television, but I'll put up with it.
d. I think all advertising should be taken off television. .
(What if the respondent is in favor of advertising on television
but not "because it tells me about the things I want to buy"? He
is almost forced by such lists to say things he doesn't mean.) .
Why do you think the government wants to keep prices from

going higher? Which one of these comes closest to your opinion?
a. So some people won't be able to get too much while others
get too little. .
..
b. So people won't worry about prices going higher and start to
hoard.
c. So there won't be any necessity for raising wages and salaries.
d. So there won't be a lot of profiteering.
(Most respondents regarded the alternatives here as equally com-
pelling and were at a loss to choose only one. )
To the question, "To what soCial class do you belong-middle

class, or upper, or lower?" almost nine tenths of the respondents an-
swered "middle class." But when psychologically more realistic alterna-
tives-middle class, lower class, working class, upper class-were offered
in a different 9uestion, only 43 per cent saic1 "middle class"; 51 per
cent chose "working class."
The sequence of items in a check list also deserves attention. The
first and last items tend to be favored; when the list is read t9the
respondent rath-er than shown to him, the last item tends especi;tlly to
receive a disproportionate number of responses. It is good practice to
have the interviewers rotate the order in which items are presented, or
GUIDE FOR QUESTIONNAIRE CONSTRUCTION 57l
to have alternate forms of the questionnaire containing different order-
ing of the items.
Is the form of response easy, definite, uniform, and

adequate tor the purpose?
EXAMPLES: Many ways of making answers easy and definite have
been referred to under preceding points. Among the most important
are the use of check categories instead of asking for precise estimates
(as with income, age, etc.) and the obtaining of raw figures rather than
averages, percentages, or other derived figures. Thus it is better not to
ask what percentage of income is spent for rent but to get the figures
for income and rent separately; to ask not the average distance of recent
automobile trips but the distance of each recent trip. It is also becoming
standard practice to show the respondent a card with the alternative
replies whenever the check list is at all lengthy or difficult to hold in
mind.
Whenever a question schedule contains a long series of blanks for
check answers or numerical entries, it is desirable to have clearly
indicated columns and to use guide lines or extra spaces to reduce the
chances of error in locating responses.
One of the methods that make for easy tabulation
/
is precoding.
The answers are anticipated; the code is set up and included on the
questionnaire; and the interviewer or respondent merely checks or
circles the coded answer. For example:
Col. No. Yes No
1. a. Do you have a radio in working order? 5 1 2
b. Do you usually read a daily newspaper? 6 1 2
c. Do you usually read a weekly newspaper? 7 1 2
d. Do you n:ad any magazine regularly? 8 1 2
This method can be used for all or part of the questionnaire. Its
advantage is that a further coding operation is unnecessary and the
information can be machine-punched immediately.
The disadvantage of this method is that there may be a tendency
to force answers into a code. Some questions are better left as free-
answer questions; precoding them would mean inaccuracy and a pos-
sible sacrifice of information.
.572 QUESTIONNAIRE CONSTRUCTION, INTERVIEW PRODECURE
DECISIONS ABOUT THE PUCE OF THE QUESTION

IN THE SEQUENCE
Is the answer to the question likely to be influenced by

the content of preceding questions?
Do earlier questions create a certain set or expectation that might
influence answers to this question?
Do preceding questions aid the recall of ideas that bear on this
question?
Does this question brcome inappropriate if certain answers were
given previously?
EXAMPLES: When both general and specific questions are to be
asked on a topic, it is usually advisable to have the general ones come
first. Thus, if people are to be asked what improvements they desire in
their working relations and also how they like their foreman, the first
question must be placed before the second; otherwise, better super-
vision will be disproportionately emphasized as a desired improvement
simply because it has been freshly called to mind~
Earlier questions may also create a generally favorable or unfavor-
\
able mental set toward the topic considered-as when a series of ques-
tions about strikes and labor troubles precedes questions about atti-
tudes toward unions and their regulation.
A final example illustrates a simple and fairly common pitfall. The
second of the following questions obviously becomes inappropriate and
awkward if the respondent has given all; affirmative answer to the first:
a. Do you think the government is giving the public as much in-
formation as it should about our military strength as compared
with Russia's? '
b. What aspects would you especially li~e to know more about?
Is the question led up to in a natural way? Is it in correct

psycho_1ogical order?
EXAMPLES: People are often more willing to answer objective ques-
tionk about situations 'and behavior than about attitudes, preferences,
and motives. Moreover, they may find it easier to formulate their atti-
tudes or motives after the situation or behavior has been clearly
specified. Thus it is usually best to start with simple objective questions.
Once the respondent has been drawn into the interview, he may be
more willing to answer questions about his feelings, motives, etc. Thus,
an interview that was designed to ascertain how people felt about living
in a trailer camp began by asking when the respondent had moved to
this camp, where he had lived before, how he happened to come to
this place, and then how he liked it in the camp, etc.
Questioning "along the time line" may be helpful. The respondent
can more readily recall objective facts and report reasons and attitudes
if he is helped to remember them in their original sequence. FOI
example, in a study of the development of occupational interests, rather
than asking simply "When did you first decide you wanted to be a
- - - ? " and "Why did you choose that .occupation?" one might
start by asking the respondent if he remembers whether as a small child
he had any ideas about what he wanted to do when he grew up, what
they were, why the occupations he favored at that time seemed attrac-
tive, and so on, specifying various periods in his life.
People often find it easier to answer if questions proceed from the
more familiar to the less so, from the more important aspects to those
of minor importance, from the relatively specific or immediate to the
more general or remote. For example, an inquiry jnto points of satis-
faction and dissatisfaction in work can move smoothly from such
matters as the activities involved in the job to surrounding conditions,
fellow workers, employment terms (hours, wages, etc.), and on to life
outside
, the plant and to economic and social conditions.
Does the question come too early or too late from the point of
view of arousil1g interest and receiving sufficient attention,
I avoiding resistance, etc.?
EXAMPLES: Some suspicion may be aroused if a home interview
with workers opens ~bruptly with the question: "Where do you work?
In what department?"
An opening qurstion such as, "Do you think the government is
giving the public as much information as it should about the hydrogen
bomb?" is likely to arouse some resistance, since the respondent may
hesitate to criticize the government to a stranger. The same question

would probably be answered much more readily at a later point in the
interview.
An opening question that was found rather uninteresting to many
respondents read: "Would you say that housing conditions for defense
workers around here are satisfactory, only fair, or poor? How about
transportation conditions (bus service, etc.) for defense workers?" An
opener such as this is to be contrasted with simple, definite, interesting
ones such as: "What was the last movie you saw?" "How many times
have you moved since Pearl Harbor?" "Have you ever been up in an
airplane?" "Do you have a radio in working order?"
It is usual to place questions asking for personal and identifying
data about the respondent at the end of the question form. He is
ordinarily more willing to tell his age, education, marital status, income
group, etc., by this time-all:d in case he is not, at least the replies to the
previous questions have not been interfered with by the suspicion or
resentment that personal questions occasionally arouse.
The Art of Interviewint

This section, like the preceding one, is concerned with stand-
ardized interviews, containing either closed or open questions. A few
of the points apply also to unstructured and partially structured inter-
vij:!ws, but for the most part such interviews require greater skill and a
quite different interviewing procedure (see Chapter 7, pages 263-268).
The quality of interviewing depends first upon proper study de-
sign. Even the most skilled interviewers will not be abie to collect
valid and useful data if the schedule of questions is inadequate to the
survey's objectives or has been put t9gether clumsily. On the other
hand, if they are. properly selected and trained, a staff of ordinary men
and women using a well-designed standardized questionnaire can elicit
the required information.
.
Within the limits of survey design, however, there is ample room
This section is tal)en, with slight modifications, from a discussion by Paul
. 6
B. Sheatsley which appeared in Volume II of Research Methods in Social Relations,
edited by Marie Jahoda, Morton Deutsch, and Stuart W. Cook (The Dryden
Press, 1951), pages 463-492.
THE ART OF INTERVIEWING 575
for "the art of interviewing" to come into play. The interviewer's art
consists in creating a situation wherein the respondent's answers will
be reliable and valid. The ideal usually sought is a permissive situation
in which the respondent is encouraged to voice his frank opinions
without fearing that his attitudes will be revealed to others and with-
out the expression of any surprise or value judgment by the interviewer.
The first requisite for successful interviewing, therefore, is to
create a friendly atmosphere and to put the respondent at his ease.
With a pleasant, confident approach and a questionnaire that starts off
easily, this is usually not difficult to achieve. From then on, the inter-
viewer's art consists in asking the questions properly and intelligibly, i1)
obtaining a valid and meaningful response, and in recording the re-
sponse accurately and completely.
CREATING A FRIENDLY ATMOSPHERE
The interviewer's introduction should be brief, casual, and posi-

tive. The study's interest lies in the actual questions, and the inter-
viewer should get into them as quickly as possible. Lengthy intro-
auctions or explanations only arouse the respondent's curiosity or
suspicion. The best approach is: "Good morning. I'm working on a
local public-opinion survey and would like to get a few of your ideas.
For instance . .. ." -and read the first question. Frequently the re-
spondent will answer that question and go right ahead with the entire
interview with only the most cursory inquiries about the objectives of
the survey.
I The interviewer's aim should be to interview everyone eligible
for the sample. A small proportion of respondents will be suspicious or
hostile, and a larger number may require a little encouragement or
persuasion; but the good interviewer will find that hardly one person in
twenty actually turns' him down. Many people are flattered to be
singled out for an interview. The interviewer should answer any
legitimate questions the respondent has and should, if necessary,
produce his credentials and explain that names are not recorded, that
the interview is not a }est (there are no "right" or "wrong" answers),
and that in a democracy it is important to find out how people feel
:}bout important issues-and the only way to find out is to ask them.
The interviewer's manner should be friendly, courteous, conversa-

tional, and unbiased. He should be neither too grim nor too effusive;
neither too talkative nor too timid. The idea should be to put the
respondent at ease, so that he will talk freely and fully. A brief remark
about the weather, the family pets, flowers, or children will often serve
to break the ice. Above all, an informal, conversational interview is
dependent upon a thorough mastery by the interviewer of the actual
questions in the schedule. He should be familiar enough with them to
ask them conversationally, rather than read them stimy; and he should
know what questions are coming next, so there will be no awkward
pauses while he studies the questionnaire.
The interviewer's job is fundamentally that of a reporter, not an
evangelist, a curiosity-seeker, or a debater. He should take all opInions
in stride and never show surprise or disapproval of a respondent's
answer. He should assume an interested manner toward his respond-
ent's opinions and never divulge his own. If he should be asked for his
views, he should laugh off the request with the remark that his job at
the moment is to get opinions, not to have them.
The interviewer must keep the direction of the interview in his
own hands, discouraging irrelevant conversation and endeavoring to
keep the respondent on the point. Fortunately, he will usually find
that the rambling, talkative respondents are~the very ones. who least
resent a firm insistence on attention to the actual business of the
interview.
AsKrnG THE QUESTIONS
Unless the interview is unstructured or only partially strlictured,

interviewers must be impressed with the importance of asking each
question exactly as it is worded. Each question has been carefully
pretested to express the precise meaning desired in as simple a manner
as possible. Interviewers must understand that even a slight rewording
of the question can so change the stimulus as to provoke answers in a
different frame of reference or bias the response.
Kny impromptu explanation_of questions is similarly taboo. Such
~ explanation ag~in may change the frame of reference or bias the
response, and it is easy to see that if each interviewer were permitted
to vary the questions as seemed best to him, the survey director would
have no assurance at all that responses were in comparable terms. If
any respondent gives evidence of failing to understand a particular
question, the interviewer can only repeat it slowly and with proper
emphasis, offering only such explanation as may be specifically author
ized in his instructions and, if understanding is still lacking, note thill
fact on the schedule.
For similar reasons, the questions must be asked in the same order
as they appear on the questionnaire. Each question sets up a frame of
reference for succeeding questions, and it is assumed that each respond
ent will be exposed to the same stimulus. Frequently the answer to a
later question will be influenced by facts called to mind in an earlier
one; to ask the later question first, even though to the interviewer there
seems sound reason for doing so, will destroy the comparability of the
interviews.
The interviewer, finally, must ask every question, unless the direc-
tions on the questionnaire specifically direct him to skip certain ones.
It may sometimes seem that the respondent has already, in answering
a prior question, given his opinion on a subsequent one, but the inter-
viewer must nevertheless ask the later question in order to be sure,
perhaps prefacing his inquiry with some such phrase as "Now you may
already have touched on this, but . . ." Similarly, ~ven if the question
seems foolish or inapplicable, the interviewer must never omit asking
it or take the answer for granted. Again, he may preface the inquiry
with some such remark as, "Now I have to ask. "
OBTAINING THE RESPONSE
It might be thought a simple matter to ask a respondent the

required questions and to record his replies, but interviewers will soon
find that obtaining a, specific, complete response is perhapS the most
difficult part of their job. People often qualify or hedge their opinions;
they answer "Don't know" in order to avoid thinking about the ques-
tion; they misinterpret the meaning of the question; they launch oft
on an irrelevant di~cussion; they contradict themselves-and in all
these cases, the interViewer usually has to probe.
Alertness to incomplete or nonspec~fic answers is perhaps the
critical test of a good interviewer, and since no cne can foresee all the
possible replies which may call for probes, each interviewer must
understand fully the over-all objective of each question, the precise
thing it is trying to measure. Both the written instructions and the oral
training should emphasize the purpose of the question and should give
examples of inadequate replies which were commonly encountered
during the pretest. By the time he is actually out interviewing, the
interviewer should have formed the automatic habit of asking himself,
after each reply the respondent gives him: "Does that completely
answer the question I just asked?"
When the first reply ,is inadequate, a simple repetition of the
question, with proper emphasis, will usually suffice to get a response
in satisfactory terms. This is particularly effective when the respondent
has seemingly misunderstood the question, or has answered it irrele-
vantly, or has responded to only a portion of it. If the respondent's
answer is vague or too general or incomplete, an effective probe is:
"That's interesting. Could you explain that a little more?" or "Let's
see, you said. . . . Just how do you mean that?"
Throughout, the interviewer must be "extremely careful not to
suggest a possible reply. People sometimes find the questions difficult,
and sometimes they are not deeply interested, in them. In either case,
they will welcome any least hint from the interviewer which will enable
them to give a creditable response. Interviewers must be thoroughly
impressed with the harm which results from a "leading probe," from
any remark which "puts words in their mouth." To be safe, the inter-
viewer should always content hims,elf with mere repetition of all or
part of the actual question, or with such innocuous nondirective probes
as are suggested in the preceding paragraph.
The "Don't know" reply is another problem for the interviewer.
Sometimes that response represents a gen.uine lack of opinion; but at
other times it may hide a host of other attitudes: fear to speak qne's
mind, reluctance to focus on the issue, vague opinions never yet ex-
pressed, a stalling for time while thoughts are marshaled, a lack of com-
prehension of the question, etc. It is the interviewer's job to distinguish
among all these types of "Don't know" response and, when appropriate,
to repeat the question with suitable assurances. In one case, for ex-
ample, he might say, "Perhaps I didn't make that too clear. Let me
read it again"; in another, he might say, "Well, lots of people have
never thought about that before, but I'd like to have your ideas on it,
just the way it seems to you." Or, again, he might point out, "Well, I
just want your own opinion on it. Actually, nobody really knows the
answers to many of these questions."
Qualified answers to questions that have been precoded in terms
of "Yes-No," "Approve-Disapprove" or similar dichotomies are an
interviewing problem which is actually in the domain of the study
director. As far as possible, the most frequent qualifications of opinion
should be anticipated in the actual wording of the question. If very
many people find it impossible to answer because of unspecified con-
tingencies, the question is a poor one. Most qualifications can be
foreseen as a result of the pretest, and those that are not taken care of
by revisions of the wording should be mentioned in the instructions to
interviewers, with directions on how to handle such answers. In some
cases, special codes may be provided for the most frequent qualifica-
tions; in other cases the interviewer may be instructed to record them as
"Don't know" or "Undecided." In avoiding many qualifications in-
herent in the response to almost any opinion question, the interviewer
may find it helpful to use phrases such as, "Well, in general, what
would you say?" or "Taking everything into consideration," or "On
the basis of the way things look to you now." /
REPORTING THE RESPONSE
There are two chief means of recording opinions during the inter-
view. If the question is precoded, the interviewer need only check a
box or circle a code~ or otherwise indicate which code comes closest
to the respondent's opinion. If the question has not been precoded,
the interviewer is expected to record the response verbatim.
On precoded questionnaires, errors and omissions in recording are
a frequent source of!interviewer error. In the midst of trying to pin the
respondent down to a specific answer, keep his attention from Bagging,
remember which qyestion comes next, and the many other problems
that engage the interviewer's attention in the field, it is not surprising
that he will sometimes neglect to indicate the respondent's reply to one
of the items, overlook some particular question, check the wrong code
on another, or ask some other question when it should be skipped.
The better the interviewer, the fewer the mistakes he will make,
but even the best interviewers will occasionally be guilty. The unfor-
givable sin is to turn in the interview as complete when it contains such
errors and omissions. The only certain way for the interviewer to avoid
this is to make an automatic habit of inspecting each interview, im-
mediately after its completion, before he goes on to another respond-
ent, to make sure that it has been filled in accurately and completely.
If he is lacking any information, he can go back and ask the respondent
for it; if his questionnaire contains any errors or omissions, he can
correct them on the spot; if his handwriting is illegible in places, or if
he has recorded verbatim replies only sketchily, he can correct the
weakness right there. If he waits until later in the day, or until he
returns home at night, he will have forgotten many of the cir~um
stances of the interview, or perhaps the prospect of editing the whole
day's work will seem so forbidding that he will skip the matter com-
pletely. .
The importance of clerical errors and omissions can be impressed
upon the interviewer during training by pointing out that the question-
naire is designed as an integral whole, and that the omission or inac-
curate reporting of a single answer can make the entir~ interview
worthless. Thus, if for each question the responses of persons with
different amounts of education are to be shown separately, and the
interviewer neglects to record the amount of schooling the respondent
has had, that whole interview must be discarded in that part of the
analysis.
In reporting responses to free-answer questions, interviewers
should be aware of the importance of complete, verbatim reporting.
It will often be difficult to get down everything the responde-nt says
in reply, bOut aside from obvious irrelevancies and repetitions, this
should be the goal. Interviewers should be given some idea of the
coding process, so that they can see the dangers of summarizing, abbre-
viating, or paraphrasing responses. Unless the coder can view the whole
answer, just as the respondent said it, he is likely to classify it im-
properly or lose some important distinctions that should be made.
Interviewers should be instructed to quote the respondent directly,
just as if they were news reporters taking down the statement of an
important official. Paraphrasing the reply, summarizing it in the inter-
viewer's own words, or "polishing up" any slang, cursing, or 'bad
grammar not only risks distorting the respondent's meaning and
emphasis, but also loses the color of his reply. Frequently the verbatim
responses of individuals are useful in the final report as illustrations of
the nuances of attitudes, and they should not be abbreviated or dis-
torted.
Although it is frequently difficult to record responses verbatim
without using shorthand, 7 a few simple techniques can greatly increase
the interviewer's speed and the extent to which he succeeds in the
verbatim recording of responses. It is perfectly permissible to ask the
respondent to wait until the interviewer gets down "that last thought
(that's pretty interesting)," but in order not to slow up the interview,
the following devices will be found helpful for speedy recording. First,
an interviewer should be prepared to write as soon as he has asked
a question and to write while the respondent talks, not waiting until
the entire response is completed. (Experienced interviewers often
finish their recording of the prior response while they ask the next
question and the respondent is considering his reply.) Second, the
interviewer should use common abbreviations. Third, he should not
bother to erase, but should cross out instead. Fo'urth, he may depart
from the ideal of verbatim recording to the extent of using a telegraphic
style; omission of "a", "the," and such parenthetical expressions as
"well," "you know," "let's see," will ordinarily not lead to loss or
distortion of meaning. But the interviewer should not speed up his
recording' by merely jotting down key words here and there. The
connecting words and phrases are easily forgotten, and the recorded
answer, even if it means something to the interviewer, may prove in-
comprehensible to the coders.
It is generally helpful if, on precoded questions, the interviewer
reports verbatim anything the respondent says to explain or qualify
his coded response; but he should not solicit such comments. The
volunteered remark~ of respondents often help the study director later
7 Shorthand recording, although it has the advantage of more easily achieving
a verbatim report, has the disadvantage of requiring later transcrjption, which may
be very time-consuming!and thus expensive.
in evaluating the meaning of the results and warn him of any com
monly held qualifications or differences in intensity of opinion.
SAMPLING
Sampling is an essential part of the interviewer's job. No mattel

how precise and detailed the original sampling design, its execution
will depend upon the training and competence of the interviewers who
carry it out. Although the interviewer's responsibility is much greater
under quota-sampling conditions, in which he himself selects the
respondents to be interviewed, even under probability sampling, where
he has no freedom of choice, he must be careful to avoid error and bias. 8
If the sample is predesignated by name, for example, the inter-
viewer should be given advice on how best to make contact with the
assigned individuals and how to overcome any hostility he may en-
counter. If a system of substitution is provided for cases in which the
originally designated respondent cannot be interviewed, the circum-
stances in which substitutions are allowed should be described care-
fully so that the sample will not be biased by too free an exercise of
this provision.
If the sample is of an area type-that is, if it involves selection of
dwelling units within a given area according to some prearranged plan-
interviewers must be thoroughly trained in its execution. It has been
found, for example, that biasing errors may easily creep into the listing
of dwelling units, and into the supposedly random selection of house-
holds and of individual respondents within those households.
As pointed out in Appendix B, biases are particularly likelY'under
quota sampling, in which the interviewer selects the subjects. It is
especially likely to occur when quotas are assigned in terms of economic
levels, the definition of which is largely subjective. Unless some re-
straint is exercised upon the interviewers, they will generally tend to
pass up persons who look unpleasant, uninterested, or inarticulate, and
to seek out individuals they think will give them "good" answers. There
i~ also the danger t,hat unless area controls are introduced, too many
interviews will be concentrated in one neighborhood, with consequent
8 Different types of samples are discussed In Appendix B.
overrepresentation of particular religious, occupational, or national
groups.
It is generally helpful, when a quota-type sample is used, to give
interviewers informal quotas in terms of education and to keep a
check on this factor as the interviews are returned. Such an additional
informal control will ensure some effort on the part of interviewers
using quota samples to avoid the usual tendency of including too few
respondents in the lowest educational and socioeconomic groups, and
will permit the study director to caution any members of the staff
who seem to be guilty of sampling bias of this type.
BIASING FACTORS INTRODUCED BY THE INTERVIEWER
Interviewer "bias"-that is, systematic differences from inter-

viewer to interviewer or, occasionally, systematic errors on the part of
many or even all interviewers-may enter not only in the selection of
the sample, but also in the asking of questions and the eliciting and
recording of responses. Interviewer bias is not simply a matter of prej-
udiced or untrained interviewers exerting influence on their respond-
ents and deliberately or carelessly distorting the answers they receive.
The dangers of bias cannot be overcome simply by hiring "impartial"
interviewers. The fact that an interviewer has strong opinions on the
subject under survey does not necessarily mean that his work will be
biased, nor does the fact that he has no strong stand of his own neces-
sarily make his work free from bias.
I Much Of what we call interviewer bias can more correctly be
described as interviewer diHerences which are inherent in the fact that
interviewers are human beings and not machines and that they do not
all work identically or infallibly. The fact that respondents, too, are
human beings, with differing perceptions, judgments, and personalities,
simply compounds th\'! differences that would occur even if the inter-
viewers were engaged in evaluating physical instead of human mate-
rials. It is not to be expected, therefore, that interviewers will unfail-
ingly bring back complete, comparable, and valid reports. Although a
large number of the !pore obvious types of error and bias can be over-
come by appropriate methods of interviewer selection and training,
some are bound to remain. Fortunately, however, it is easier for the
study director to become aware of the biases of interviewers, and thus

to discount their effects in his interpretation of the data, than it is
for the clinician, the experimenter, or the participant observer to
detect his own bias when he himself collects the data.
Assuming an unbiased selection of respondents, bias in the inter-
view situation appears to come about through (1) the respondent's
perception of the interviewer, and (2) the interviewer's perception of
the respondent. We use the term perception here in the broad sense,
which emphasizes the manner in which the relation between inter-
viewer and respondent is influenced and modified by their wishes,
expectations, and personality structure.
There is an abundance of experimental evidence to prove that bias
may result, under certain conditions, regardless of anything the inter-
viewer may do to eliminate it. In one study, 50 per cent of a sample
of non-Jewish respondents told non-Jewish interviewers that they
thought Jews had too much influence in the business world, whereas
only 22 per cent of an equivalent sample voiced that opinion to Jewish
interviewers. Similar experiments have shown that Negroes will fre-
quently answer differently when interviewea by white people, and
that working-class respondents are less likely to talk freely to middle-
class interviewers. Such effects can occur no matter how conscientiously
the interviewer attempts to be "unbiased."
The magnitude of these effects naturally varies With the way in
which the respondent perceives the situation. Thus, in one study, it
was demonstrated that Negroes spoke more frankly with white inter-
viewers in New York than they did in Memphis, Tennessee. The
interviewing situation was "objectively" the same in both cities, but
respondents perceived it differently. By altering the respondent's per-
ception of the situation (for example, by assuring him that his name
will not be recorded), these biasing .effects can often be reduced, but
they can seldom be eliminated. -
The study director should keep these matters in mind when he
selects his interviewers, and the staff should be warned of the dangers.
It is for reasons of this type that interviewers are usually instructed, for
~ample, to dress inconspicuously so that their clothes and appearance
will not influence lower-class respondents; to interview the respondent
privately so that his opinions will not be affected by the presence of
some third person; and to adopt an informal, conversational mannel
in an effort to achieve the best possible rapport.
Not all interviewer effects operate through the respondent's per-
ception of the interviewer, however. Indeed, some respondents appear
to be totally immune to even the most flagrant biasing characteristics of
the interviewer. Fully as important a source of bias are the interviewer's
perceptions of his respondent. No matter how standardized the ques-
tionnaire may be and no matter how rigidly the interviewer may be
instructed, he still has much opportunity to exercise freedom of choice
during the actual interview, and it is often his perception of the re-
spondent that determines the manner in which he asks the question,
the way in which he probes, his classification of equivocal responses to
precoded questions, and his recording of verbatim answers.
Interviewers do not approach each new respondent in an unstruc-
tured fashion: indeed, they often have strong expectations and stereo-
types, which are more and more likely to come into playas they
continue interviewing. On the basis of their past judgments, or of prior
answers received from other respondents, they may, for example, quite
unconsciously come to associate lack of education with ethnic or reli-
gious prejudice; or they may come to anticipate a large number of
"No opinion" responses from the Negroes they interview. Such ex-
peGtations will almost inevitably affect their performance.
Thus, given the same "No opinion" respons~ from a wealthy
businessman and from a Negro housewife, they may probe the former's
reply, in the belief that an opinion must be lurking there somewhere,
whereas they will routinely accept the latter's reply without probing
and'go on to the next question. An experimental study has shown that
when the same equivocal answer regarding aid to Europe was em-
bedded first in an "isolationist" context of previous responses and then
in an "internationalist" context, only 20 per cent of the interviewers
classified it in internationalist terms in the first context, but 75 per
cent of the same interviewers classified it as internationalist in the
second context. Experiments on verbatim recording have also shown
that interviewers tend to select from long answers those parts that
most nearly conform ~o their own expectations or opinions and to dis-
card the rest.
A final source of' bias arises from the interviewer's perception of
the situation. If he sees the results of the survey as a possible threat to

his interests or beliefs, for example, he is likely to introduce bias. Or if
he regards his assignment as impossible, he is almost bound to intro-
duce bias. Such difficulties can best be overcome by proper motivation
and supervision.
Since interviewers are human beings, such biasing factors can
never be overcome completely, but their effects can be reduced by
standardizing the interview, so that the interviewer has as little free
choice as possible. Thus, the use of a standard wording in survey ques-
tions aims to prevent the bias that would result if each interviewer
worded the question in his own fashion. Similarly, if interviewers are
given standard instructions on probing procedure, on the classification
of doubtful answers, and so on, their biases will have less chance to
operate.
It should be noted, however, that as the interviewer's freedom is
restricted, the opportunities for effective use of his insight are cor-
respondingly restricted. Conversely, the more responsibilities the inter-
viewer is given for probing and evaluating his respondent's opinion, the
more bias is likely to result. A compromise must generally be made.
In a study whose results are to be analyzed statistically and quantita-
tively, and in which large numbers of inexperienced or hastily trained
interviewers are relied upon, it is wise to reduce the interviewer's free-
dom of choice to a minimum by standardizing, so far as possible, every
aspect of the interview situation.
Since bias, in the sense that different interviewers will not always
bring back the same answers from equivalent respondents; can never
I
be entirely eliminated, the study director's main responsibilities are.
so to select, train, and supervise his staff that any net effect of bias
will be at a minimum, and to be aware of the possibilities of bias at
various points so that he can discount their effects in his analysis.
Many critics tend to exaggerate the significance of "interview
bias"-overlooking the fact that social scientists are universally depend-
ent upon data that have been collected by means of oral or written
reports, and that these reports, no matter how collected, are invariably
subject to essentially the same sources of error and bias as are those
collected by survey interviewers. The clinician and, frequently, the ex-
perimenter depend upon oral reports of feelings, perceptions, behavior,
etc.; and they, as well as the sophisticated "participant observer" in
another type of investigation, are just as likely to bias their subjects'
responses as are the interviewers participating in an attitude survey.
The major difference is that when the social scientist has to depend
upon the reports of interviewers whom he selects and trains, he be-
comes more aware of the dangers and difficulties involved.
BIBLIOGRAPHY
NBT, L. E. and L. BELLAK, 1950. Projective psychology. Knopf.

ACKERMAN, N. W., and M. JAHODA, 1950. Antisemitism and emotional dis-
order: a psychoanalytic interpretation. Harper.
ACKOFF, R. L., 1953. The design of social research. University of Chicago.
ADORNO, T. W., E. FRENKEL-BRUNSWIK, D. J. LEVINSON, and R. N.
SANFORD, 1950. The authoritarian personality. Harper.
ALLPORT, G. W., 1942. The use of personal documents in psychological
science. Bulletin 49, Social Science Research Council.
ALLPORT, G. W., 1950. Prejudice: a problem of psychological and social
causation. J. Social Issues, Supplement Series No.4.
AMERICAN PSYCHOLOGICAL AsSOCIATION, 1953. Ethical standards of psy-
I chologists.
AMERICAN PSYCHOLOGICAL ASSOCIATION, 1957. Publication manual (re-
vised).
ANDERSON, H. H., and b. L. ANDERSON, 1951. An introduction to projective
techniques and oj:ber devices for understanding the dynamics of hu-
man behavior. Prentice-Hall.
BAEHNE, G. W., 1935, Practical applications of the punched card method
in colleges and un,iversities. Columbia University.
'BALDWIN, A. L., 1950. Statistical problem~ in the treatment of case his-
tories. J. Clinical Psychology, 6,6-12.
BALES, R. F., 1950. Interaction process analysis. Addison-Wesley.
BALES, R. F., and H. GERBllANDS, J 948. The "interaction-recorder."
Human Relation~, 1,456-463.
589
BIBLIOGRAPHY
BARKER, R., T. DEMBO, and K. LEWIN, 1941. Frustration and regression:

an experiment with young children. Studies in topological and vector
psychology II. University of Iowa Studies in Child Welfare, 18, No. I.
BARTLETT, F. C., 1932. Remembering. Cambridge University.
BEERS, C. W., 1935. A mind that found itself: an autobiography (8th ed.,
1953, Doubleday). National Committee for Mental Hygiene.
BELL, J. E., 1948. Projective techniques: a dynamic approach to the study
of the personality. Longmans Green.
BENNE, K. D., and G. E. SWANSON, eds., 1950. Values and the social sci-
entist. J. Social Issues, 6, No.4.
BERELSON, B., 1952. Content analysis in communication research. Free
Press.
BERKOWITZ, L. and H. GUETZKOW, 1949. Manual for overall observers.
Naval Conference Research, University of Michigan. Mimeographed.
BETTELHEIM, B., and M. JANOWITZ, 1950. Dynamics of prejudice. Harper.
BEVERIDGE, W. I. B., 1950. The art of scientific investigation. Norton.
BEYLE, H. C., 1932. A scale for the measurement of attitude toward
candidates for elective governmental office. American Political Science
Review, 26, 527-544.
BLANKENSHIP, A. B., 1946. How to conduct consumer and opinion research.
Harper.
BLANKENSHIP, A. B., et al., 1947. Survey on problems of interviewer cheat-
ing. International J. Opinion and Attitude Research, 1,93-107.
BOMRDUS, E. S., 1925. Measuring social distances. J. Applied Sociology,
9,299-308.
BOGARDUS, E. S., 1928. Immigration and race attitudes. Heath.
BOGARDUS, E. S., 1933. A social distance scale. Sociology and Social Re-
search, 17,265-271.
BRAITHWAITE, R. B., 1955. Scientific explanation: a study of the function
of theory, probability and law in science. Cambridge University.
BROWN, J. F., 1947. A modification of the Rosenzweig picture-frustratio~
1
test to study hostile interracial attitudes. J. Psychology, 24, 247-272.

BRUNER, J. S., 1941. The dimensions of propaganda: German shortwave
broadcasts t9 America. J. Abnormal and Social Psychology, 36, 311-
337. .
BRUNSWIK, E., .} 955. Representative design and probabilistic theory in
functional psychology. Psychological Review, 62, 193-217.
BRUNSWIK, E., 1956. Perception and the representative design of psycho-
logical experiments (2d ed.). University of California.
BUHLER, C., 1934. Drei Generationen im Jugendtagebuch. Quellen und
I Studien zur Jugendkunde, 11.
BURWEN, L. S., and D. T. CAMPBELL, 1957. The generality of attitudes to-
ward authority and nonauthority figures. J. Abnormal and Social Psy-
chology, 54, 24-31.
BIBLIOGRAPHY 591
CAMPBELL, D. T., 1950. The indirect assessment of social attitudes. Psy-
chological Bulletin, 47, 15-38.
CAMPBELL, D. T., 1957. Factors relevant to the validity of experiments in
social settings. Psychological Bulletin, 54, 297-312.
CAMPBELL, D. T., and D. W. FISKE, 1959. Convergent and discriminant
validation by the multitrait-multiruethod matrix. Psychological Bul-
letin, 56.
CANTER, R. R., JR., 1951. An experimental study of a human relations
training program. J. Applied Psychology, 35,38-45.
CANTRIL, H. A., JR., A. AMES, JR., A. H. HASTORF, and W. H. ITTELSON,
1949. Psychology and scientific research. Science, 110,461-464,491-
497,517-522.
CARTER, H., 1945. Recent American studies in attitudes toward war: a
summary and evaluation. American Sociological Review, 10, 343-352.
CATTELL, R. B., 1952. Factor analysis: an introduction and manual for the
psychologist and social scientist. Harper.
CATTELL, R. B., A. B. HEIST, P. A. HEIST, and R. G. STEWART, 1950. The
objective measurement of dynamic traits. Educational and Psycho-
logical Measurement, 10,224-248.
CATTELL, R. B., and L. B. LUBORSKY, 1950. T-technique demonstrated as
a new clinical method for determining personality and symptom struc-
ture. J. General Psychology, 42, 3-24.
CATTELL, R. B., E. F. MAXWELL, B. H. LIGHT, and M. P. UNGER, 1949.
The objective measurement of attitudes. British J. Psychology, 40,
81-90.
CHAPIN, F. S., 1947. Experimental designs in sociological research. Harper.
CHAPPLE, E. D., 1949. The interaction chronograph: its evaluation and
present application. Personnel, 25,295-307.
CHEIN, I., 1956. Narcotics use among juveniles. Social Work, 1, 50-60.
CHEIN, I., A. F. CITRON, M. H. WORMSER, and W. SCHWARTZ, 1952.
Basic issues in Jewish education and group work. American Jewish
Congress (New York). Mimeographed.
CHURCHMAN, C. W., 1918. Theory of experimental inference. Macmillan.
CITRON, A. F., I. CHEINi and J. HARDING, 1950. Anti-minority remarks: a
problem for action research. J. Abnormal and Social Psychology, 45,
99-126.
CLARK, K. B., and M. P.:CLARK, 1939. The development of consciousness
of self and the emergence of racial identification in Negro pre-school
children. J. Social Psychology, 10, 591-599.
CLARK, K. B., and M. P. CLARK, 1950. Emotional factors in racial identifica-
tion and preference in Negro children. J. Negro Education, 19, 341-
350.
COCHRAN, W. G., 1953. Sampling techniques. Wiley.
592 BIBLIOGRAPHY
COCHRAN, W. G., 1956. Design and analysis of samplings. In Statistical

methods, by G. W. Snedecor, 5th ed. Iowa State College.
COCHRAN, W. G., and G. M. Cox, 1957. Experimental designs (2d ed.) .
Wiley.
COHEN, A. K., 1955. Delinquent boys: the culture of the gang. Free Press.
COHEN, M. R., and E. NAGEL, 1934. An introduction to logic and scientific
method. Harcourt.
COLLIER, J., JR., 1957. Photography in anthropology: a report on two ex-
periments. American Anthropologist, 59, 843-859.
Confessions of St. Augustine, 1948. (F. J. Sheed, trans.) Sheed &Ward.
COOK, L. A., 1950. Intergroup relations in teacher education. College Study
in Intergroup Relations, Vol. II. American Council on Education
(Washington, D.C.)
COOK, S. W., 1949. The role of social values in social-psychological research.
Mimeographed. Research Center for Human Relations, New York
University.
COOK, S. W., J. HAVEL, and J. R. CHRIST, 1957. The eHects of an orienta-
tion program for foreign students. Mimeographed. Research Center
for Human Relations, New York University.
COOMBS, C. H., 1948. Some hypotheses for the analysis of qualitative
variables. Psychological Review, 55, 167-174._
COOMBS, C. H., 1950. Psychological scaling without a unit of measurement.
Psychological Review, 57,145-158.
COOMBS, C. H., 1953. Theory and methods of social measurement. In
Research methods in the behavioral sciences, ed. by L. Festinger and
D. Katz. Dryden.
COTTRELL, L. S., JR., ed., 1957. Public reaction to the atomic bomb and
world aHairs. Cornell University.
CRISWELL, J. H., 1937. Racial cleavages in Negro-white groups. Sociometry,
1,87-89.
CRISWELL, J. H., 1939. Social structure I revealed in a sociometric retest.
Sociometry, 2, 69-75~
CRONBACH, L. J., 1949. Essentials of psychological testing. Harper.
CRONBACH, L. J., 1951. Coefficient alpha and the internal,structure of tests.
Psychometrika, 16, 297-334. I
CRONBACH, L. J., 1953. Correlations between persons as a resf'a.rch tool. In

Psychotherapy: theory and research, ed. by O. H. Mowrer. Ronald
Press.
CRONBACH, L. J., and P. E. MEEHL, 1955. Constrllct validity in psycho-
logical tests. Psychological Bulletin, 52,28]-302.
CRqTCHFIELD, R., and'D. H, GORDON, 1947. Variations in respondents'
interpretations of an opinion-poll question. International J. Opinion
and Attitude Research, 1,22-31.
BIBLIOGRAPHY 593
DAVIDSON, D., P. SUPPES, and S. SIEGEL, 1957. Decision making: an ex-
perimental approach. Stanford University.
DAVISON, W. P., 1947. An analysis of the Soviet-controlled Berlin press.
Public Opinion Quarterly, 11, 40-57.
DEAN, J. P., 1954. Participant observation and interviewing. In An intro-
duction to social research, ed. by J. T. Doby. Stackpole.
DEMING, W. E., 1950. Some theory of sampling. Wiley.
DERI, S., D. DINNERSTEIN, J. HARDING, and A. D. PEPITONE, 1948. Tech-
niques for the diagnosis and measurement of intergroup attitudes and
behavior. Psychological Bulletin, 45, 248-271.
DE SALES, R. R., ed., 1941. My new order. Reynal.
DEUTSCH, M., 1949. An experimental study of the effects of cooperation
and competition upon group process. Human Relations, 2, 199-232.
DEUTSCH, M., and M. E. COLLINS, 1951. Interracial housing: a psycho-
logical evaluation of a social experiment. University of Minnesota.
DIXON, W. J., and F. J. MASSEY, JR., 1957. Introduction to statitical
analysis. (2d ed.). McGraw-Hill.
DOLLARD, J., L. W. DOOB, N. E. MILLER, O. H. MOWRER, and R. H.
SEARS, 1939. Frustration and aggression. Yale University.
Du BOIS, C., 1944. The people of Alor: a social-psychological study of an
East Indian island. University of Minnesota.
DUDYCHA, G. J., 1943. A critical examination of the measurement of atti-
tude toward war.,. Social Psychology, 39, 846-860.
DURKHEIM, E., 1951. Suicide. (J. A. Spaulding and G. Simpson, trans.)
Free Press.
ECKERT, J. W., 1940. Punched card methods in scientific computation. The
Thomas J. Watson Astronomical Computing Bureau, Columbia
U niversi ty.
EDWARDS, A. L., 1950. Experimental design in psychological research.
Rhinehart.
ED,,\ARDS, A:L., 1957a. Techniques of attitude scale construction. Apple-
ton-Century-Crofts.
EDWARDS, A. L., 1957b. The social desirability variable in personality assess-
ment and research. Dryden.
EDWARDS, A. L., and K. C. KENNEY, 1946. A comparison of the Thurstone
and Likert techniques of attitude scale construction. J. Applied Psy-
chology, 30, 72-83.
EDWARDS, A. L., and F. P. KILPATRICK, 1948. A technique for construction
of attitude scales.l-:Applied Psychology, 32,374-384.
EYSENCK, H. J., and S. CROWN, 1949. An experimental study in opinion-
attitude methodology. International J. Opinion and Attitude Research,
3,47-86. I
FERGUSON, L. W., 1935. The influence of individual attitudes on construc-

tion of an attitude scale. J. Social Psychology, 6, 115-117.
594 BIBLIOGRAPHY
FERGUSON, L. W., 1957. More measurement than validation. Contempo-

rary Psychology, 2, 237-238.
FESTINGER, L., 1947. The treatment of qualitative data by scale analysis.
Psychological Bulletin, 44, 149-161.
FESTINGER, L., and D. KATZ, eds., 1953. Research methods in the behav- .
ioral sciences. Dryden.
FESTINGER, L., H. W. RIECKEN, and S. SCHACHTER, 1956. When prophecy
fails. University of Minnesota.
FESTINGER, L., S. SCHACHTER, and K W. BACK, 1950. Social pressures in
informal groups. Harper.
FISHER, R. A., 1951. The design of experiments (6th ed.). Oliver and
Boyd. I
FLESCH, R. F., 1949. The art of readable writing. Harper.

FLESCH, R. F., 1954. How to make sense. Harper.
FLESCH, R. F., and A. H. LASS, 1955. The way to write (2nd ed.). McGraw-
Hill.
FRANK, L. K, 1939. Projective methods for the study of personality. J.
Psychology, 8, 389-413.
FRANZ, J. G., 1940. The place of the psychodrama in research. Sociometry,
3,49-61.
FREEMAN, F. S., 1955. Theory and practice of psychological testing (rev.
ed.). Holt.
FREUD, S., 1939. Moses and monotheism. Knopf.
FROMME, A., 1941. On the use of certain qualitative methods of attitude
research: a study of opinions on the methods of preventing war. J.
Social Psychology, 13,425-459.
GARDNER, B. B., and S. J. LEVY, 1955. The product and the brand. Harvard
Business Review, 33,33-39.
GETZELS, J. W., 1951. The assessment of personality and prejudice by the
method of paired direct and projeCtive questions. Unpublished doc-
toral dissertation, Harvard University.
GOLDS EN, J. M., 1947. Analyzing the contents of mass communication:- a
step toward inter-group harmony. International J. Opinion and Atti~
tude Research, 1, 81-92.
GOOD, C. V., 1941. Effective reporting of research. Phi Delta Kappan, 24,
178-184.
GOODE, W. J., and P. K HATT, 1952. Methods in social research. McGraw-
Hill.
GOODMAN, M. E., 1952. Race awareness in young children. Addison-
I Wesley.
GOTTSCHALK, L., C. KLUCKHOHN, and R. ANGELL, 1945. The use of per-
sonal documents in history, anthropology and sociology. Bulletin 53,
Social Science Research Council.
BIBLIOGRAPHY 595
GRANNEBERG, R. T., 1955. The influence of individual attitude and atti-
tude-intelligence interaction upon scale values of attitude items.
American Psychologist, 10,330-331. Abstract.
GREEN, B. F., 1954. Attitude measurement. In Handbook ot social psy-
chology, Vol. I, ed. by G. Lindzey. Addison-Wesley.
GREENWOOD, E., 1945. Experimental sociology. King's Crown.
GROSS, N., W. S. MASON, and A. W. McEACHERN, 1958. Explorations in
role analysis: studies ot the school superintendency role. Wiley.
GUETZKOW, H., 1950. Unitizing and categorizing problems in coding
qualitative material. J. Clinical Psychology, 6, 47-58.
GUILFORD, J. P., 1954. Psychometric methods (2d ed.). McGraw-Hill.
GULLIKSEN, H., 1950. Theory ot mental tests. Wiley.
HAIRE, M., 1950. Projectiv{. techniques in marketing research. J. Market-
ing, 14,649-656.
HALBWACHS, M., 1930. Les causes du suicide. Felix Alcan (Paris).
HAMMOND, K. R., 1948. Measuring attitudes by error-choice: an indirect
method. J. Abnormal and Social Psychology, 43, 38-48.
HAMMOND, K. R., 1955. Probabilistic functioning and the clinical method.
Psychological Review, 62, 255-262.
HANSEN, M. H., W. N. HURWITZ, and W. G. MADOW, 1953. Sample
survey methods and theory. Wiley. 2 vols.
HARDING, J., and R. HOGREFE, 1952. Attitudes of white department store
employees toward Negro co-workers. J. Social Issues, 8,1,18-28.
HARTLEY, E. L., 1946. Problems in prejudice. King's Crown.
HARTLEY, E. L., and S. SCHWARTZ, 1948. A pictorial doll play approach
tor the study ot children's intergroup attitudes. American Jewish
Committee (New York). Mimeographed.
HEBB, D.O., and W. R. THOMPSON, 1954. The social significance of
animal studies. In Handbook of social psychology, Vol. I, ed. by G.
Lindzey. Addison--Wesley.
HELGERSON, E., 1943. The relative significance of race, sex, and facial
expression in choice of playmate by the pre-school child. J. Negro
Education, 12, 617...:622.
HEMPEL, C. G., 1952. Fundamentals of concept formation in empirical
science. International Encyclopedia of Unified Science, Vol. II, No.7.
University of Chicago.
HENRY, W. E., and H. ]GUETZKOW, 1951. Group projection sketches for
the study of small groups. J. Social Psychology, 33, 77-102.
HINCKLEY, E. D., 1932 .. The influence of individual opinion on construc-
tion of an attitude stale. J. Social Psychology, 3, 283-296.
HOMANs, G. C., 1950. The human group. Harcourt.
HOROWITZ, E. L., 1936. The development of attitude toward the Negro
Archives Psychology, No. 194.
596 BIBLIOGRAPHY
HOROWITZ, E. L., and R. E. HOROWITZ, 1938. Development of social atti

tudes in children. Sociometry, 1, 301-338.
HOROWITZ, R. E., 1939. Racial aspects of self-identification in nursery
school children. J. Psychology, 7,91-99.
HORWITZ, M., and D. CARTWRIGHT, 1953. A projective method for the
diagnosis of groups. Human Relations, 6,397-410.
HOVLANI:, C. I., I. K. JANIS, and H. H. KELLEY, 1953. Communication
and persuasion: psychological studies of opinion change. Yale Uni-
versity.
HOVLAND, C. I., A. A. LUMSDAINE, and F. D. SHEFFIELD, 1949. Experi-
ments on mass communication. Studies in Social Psychology in
World War II, Vol. III. Princeton University.
HovLANn, C. I., and R. R. SEARS, 1940. Minor studies in aggression: VI.
Correlations of lynchings with economic indices. J. Psychology, 9,
301-310.
HOVLAND, C. I., and M. SHERIF, 1952. Judgmental phenomena and scales
of attitude measurement: item displacement in Thurstone scales. J.
Abnormal and Social Psychology, 47,822-832.
HUNT, J. McV., 1941. The effects of infant feeding-frustration upon adult
hoarding in the albino rat. J. Abnormal and Social Psychology, 36,
338-360.
HYMAN, H. H., 1955. Survey design and analysis: principles, cases, aHd
procedures. Free Press.
HYMAN, H. H., et al., 1954. Interviewing in social research. University of
Chicago.
INFORMATION AND EDUCATION DIVISION, U. S. War Department, 1947.
Opinions about Negro infantry platoons in white companies of seven
divisions. In Readings in social psychology, ed. by T. M. Newcomb
and E. L. Hartley. Holt.
JAHODA, M., and S. W. COOK, 1952. S~curity measures and freedom of
thought. Yale Law Journal, 61, 297-333.
JAHODA-LAZARSFELD, M., and H. ZEISL, 1932. Die Arbeitslosen von Mari-
enthal. Hirzel (Leipzig).
JANIS, I. L., 1943. Meaning and the study of symbolic behavior. Psychiatry,
6,425-439. ' .
JASPERS, K., 1950. Is science evil? Commentary, 9, 229-233.
JENNINGS, H. H., 1943. Leadership and isolation (2d ed. 1950). Longmans.
JONES, E., 1953-1957. The Life and work of Sigmund Freud. Basic Books.
3 vols.
JONES, E. E., and R. KOHLER, 1958. The effects of plausibility on the learn
'ing of controversial statements. J. Abnormal and Social Psychology,
57,315-320.
KAPLAN, A., 1943a. Content analysis and the theory of signs. Philosophy of
Science, 10, 230-247.
BIBLIOGRAPHY 597
KAPLAN, A., 1943b. Reliability of "favorable" and "unfavorable" symbol
classification in newspaper headlines. Document No. 42, Library of
Congress, Experimental Division for the Study of War Time Com-
munications.
KAPLAN, A., and J. M. GOLDSEN, 1943. Reliability of certain categories for
classifying newspaper headlines. Document No. 40, Library of Con-
gress, Experimental Division for the Study of War Time Communica-
tions.
KATZ, D., 1951. Social psychology and group process. Annual Review of
Psychology, ed. by C. P. Stone. Annual Reviews, Inc. (Stanford,
Calif.) .
KATZ, D., and H. H. HYMAN, 1947. Morale in war industries. In Readings
in social psychology, ed. by T. M. Newcomb and E. L. Hartley. Holt.
KATZ, I., J. GOLDSTON, and L. BENJAMIN, 1958. Behavior and productivity
in bi-racial work groups. Human Relations, 11,123-141. .
KELLER, H. A., 1903. The story of my life. Doubleday, Page.
KELLEY, H. H., C. I. HOVLAND, M. SCHWARTZ, and R. P. ABELSON, 1955.
The influence of judges' attitudes in three methods of attitude scaling.------
J. Social Psychology, 42,147-158.
KELLY, E. L., and D. W. FISKE, 1950. The prediction of success in the VA
training program in clinical psychology. American Psychologist, 5,
395-406.
KENDALL, P. L., and P. F. LAZARSFELD, 1950. Problems of survey analysis.
In Continuities in social research, ed. by R. K. Merton and P. F.
Lazarsfeld. Free Press.
KERR, M., 1943. An experimental investigation of national stereotypes.
Sociological Review, 35,37-43.
KINSEY, A. C., et al., 1948. Sexual behavior in the human male. Saunders.
KISH, L., 1953. Selection of the sample. In Research methods in the be-
havioral sciences, ed. by L. Festinger and D. Katz. Dryden.
KLINEBERG, 0., 1935a. Negro intelligence and selective migration. Colum-
bia University.
KLINEBERG, 0., 193 5b. Race differences. Harper.
KORNHAUSER, A.; 195L Constructing questionnaires and interview sched-
ules. In Research methods in social relations, Vol. II, first ed., ed. by
M. Jahoda, M. Deutsch, and S. W. Cook. Dryden.
KRAMER, B. M., 1949. Dimensions of prejudice. J. Psychology, 27, 389-
451.
KRECH, D., and R. S. CRUTCHFIELD, 1948. Theory and problems of social
psychology. McGraw-Hill.
LAMBERT, R. D., and M. BRESSLER, 1957. Indian students on an Amer-
ican campus. University of Minnesota.
LANDIS, K. M., 1948. Segregation in Washington. National Committee on
Segregation in the Nation's Capital (Chicago).
598 BIBLIOGRAPHY
LAPIERE, R. T., 1934. Attitudes vs. actions. Social Forces, 14, 230-237.
LASSWELL, H. D., 1942a. Analyzing the content of mass communication:
a brief introduction. Document No. 11, Library of Congress, Experi-
mental Division for Study of War Time Communications.
LASSWELL, H. D., 1942b. The politically significant content of the press:
coding procedures. Journalism Quarterly, 19, 12-23.
LASSWELL, H. D., 1946. Describing the contents of communications. In
Propaganda, communisation, and public opinion, by B. L. Smith,
H. D. Lasswell and R. D. Casey. Princeton University.
LASSWELL, H. D., N. LEITES, and associates, 1949. Language of politics.
George W. Stewart.
LAZARSFELD, P. F., 1935. The art of asking why. National Marketing
Review, 1, 26-38.
LAZARSFELD, P. F., 1944. The controversy over detailed interviews-an
offer for negotiation. Public Opinion Quarterly, 8, 38-60.
LAZARSFELD, P. F., 1957. Latent structure analysis. Bureau of Applied
Social Research, Columbia University. Mimeographed.
LEAHY, A., 1931. Punching psychological and sociological data on Hol-
lerith cards. J. Applied Psychology, 15, 199-207.
LEE, A. M., 1949. A sociological discussion of consistency and incon-
sistency in intergroup relations. J. Social Issues,S, 12-18.
LEE, A. M., and N. D. HUMPHREY, 1943. Race riot. Dryden.
LEE, R. S., 1957. The family of the addict: a comparison of the family
experiences of male juvenile heroin addicts and,controls. Unpublished
doctoral dissertation, New York University, Graduate School of Arts
and Science.
LEIGHTON, A. H., 1949. Human relations in a changing world. Dutton.
LEVINE, J. M., and G. MURPHY, 1943. The learning and forgetting of con-
troversial material. J. Abnormal and Social Psychology, 38,,507-517.
LEVY, D. M., 194~. Maternal overprotection. Columbia University.
LEWIN, K., 1951. Formalization and progress in psychology. In Field
theory in social science: selected theoretical papers, ed. by D. Cart-
wright. Harper.
LEWIN, K., C. E. MEYERS, J. KALHORN, M. L. FARBER, and J. R. P.
FRENCH, JR., 1944. Authority and frustration. Studies in topological
and vector psychology III. University of Iowa Studies in Child Welfare
20.
LEWIS, H. B., 1941. Studies in the principles of judgments and attitud<:!s:
IV. The operation of "prestige suggestion." J. Social Psychology, 14,
1229-256.
I LIKERT, R., 1932. A technique for the measurement of attitudes. Archives
Psychology, No. 140.
LIKERT, R., and R. LIPPITT, 1953. The utilization of social science. In
BIBLIOGRAPHY 599
Research methods in the behavioral sciences, ed. by L. Festinger and
D. Katz. Dryden.
LINDGREN, E. J., 1935. Field work in social psychology. British J. Psychol-
ogy, 26, Part 2,174-182.
LINDQUIST, E. F., 1953. Design and analysis of experiments in psychology
and education. Houghton Miffiin.
LINDZEY, G., and E. F. BORGATTA, 1954. Sociometric measurement. In
Handbook of social psychology, Vol. I, ed. by G. Lindzey. Addison-
Wesley.
LOEVINGER, J., 1947. A systematic approach to the construction and eval-
uation of tests of ability. Psychological Monographs, 61, No.4.
LOEVINGER, J., 1948. The technic of homogeneous tests compared with
some aspects of "scale analysis" and factor analysis. Psychological
Bulletin,45,507-529.
LOWENTHAL, L., 1943. Biographies in popular magazines. In Radio re-
search 1942-43 ed. by P. F. Lazarsfeld and F. N. Stanton. Duell,
Sloan.
LTJNDBERG, G. A., M. KOMAROVSKY, and M. A. McINERY, 1934. Leisure:
a suburban study. Columbia University.
MCCARTHY, P. J., 1951. Sample design. In Research methods in social
relations, Vol. II, first ed., ed. by M. Jahoda, M. Deutsch, and S. W.
Cook. Dryden.
MCCLELLAND, D. C., 1951. Personality. Dryden.
MACCOBY, E. E., and N. MACCOBY, 1954. The interview: a tool of social
science. In Handbook of social psychology, Vol. 14 ed. by G. Lindzey.
Addison-Wesley.
MACCRONE, I. D., 1937. Race attitudes in South Africa. Oxford.
MCGRANAHAN, D. V., 1951. Content analysis of the mass media of com-
munication. In Research methods in social relations, Vol. II, first ed.,
, ed. by M. Jahoda, M. Deutsch, and S. W. Cook. Dryden.
MCGREGOR, D., 1935, Scientific measurement and psychology. Psycho-
logical Review, 42,246-266.
McNEMAR, Q., 1940.: Sampling in psychological research. Psychological
Bulletin, 37,331-365.
MARGENAU, H., 19 50~ The nature of physical reality: a philosophy of
modern physics. McGraw-HilI.
MEAD, M., 1946. Rese:arch on primitive children. In Manual of child psy-
chology, ed. by L. Carmichael. Wiley.
MEREI, F., 1949. Group leadership and institutionalization. Human Rela-
tions, 2, 23-29.
MERTON, R. K., 194 Y. Selected problems of field work in the planned
community. American Sociological Review, 12, 304-312.
MERTON, R. K., 1957. Social theory and social structure, (rev. ed.). Free
Press.
600 BIBLIOGRAPHY
MERTON, R. K, M. FISKE, and A. CURTIS, 1946. Mass persuasion. Harper.

MERTON, R. K, M. FISKE, and P. L. KENDALL, 1956. The focused inter-
view. Free Press.
MERTON, R. K., and P. F. LAZARSFELD, eds., 1950. Continuities in social
research. Free Press.
MERTON, R. K, and A. S. ROSSI, 1957. Contributions to the theory of
reference group behavior. In R. K Merton, Social theory and social
structure, (rev. ed.) . Free Press.
MERTON, R. K, P. S. WEST, and M. JAHODA, forthcoming. Patterns of
social life: explorations in the social psychology and sociology of hous-
ing. Part of this work, entitled Social facts and social fictions: the
dynamics of race relations in Hilltown, is available in hectographed
form from the Bureau of Applied Social Research, Columbia Uni-
versity.
MINTZ, A., 1946. A re-examination of correlations between lynchings and
economic indices. J. Abnormal and Social Psychology, 41, 154-160.
MORENO, J. L., 1934. Who shall survive? (rev. ed. 1953, Beacon House)
Nervous and Mental Disease Publishing Co., Series No. 58. Wash-
ington, D. C.
MORGAN, J. J. B., 1945. Attitudes of students toward the Japanese. J. Social
Psychology, 21, 219-227.
MORRIS, R. T., and O. M. DAVIDSEN, forthcoming (probably 1959). The
two-way mirror: a study in cross-cultural education (tentative title).
University of Minnesota.
MOWRER, O. H., 1953. uQ technique"-description, history, and critique.
In Psychotherapy: theory and research, ed. by O. H. Mowrer. Ronald
Press.
MUNSTERBERG, H., 1908, 1923. On the witness stand: essays on psychol-
ogy and crime. Clark Boardman.
MURPHY, G., and R. LIKERT, 1938. PU'bJic opinion and the individual. ,
Harper.
MURPHY, G., L. B. MURPHY, and T. M. NEWCOMB, 1937. Experimental
social psychology. Harper.
MURRAY, H. A., et al., 1938. Exploratio~s in personality. Oxford.
MURRAY, H. A., and C. D. MORGAN, 1945. A clinical study of sentiments.
Genetic Psychology Monographs, 32, 3-149; 153-311.
MUSSEN, P. H., 1950. The reliability and validity of the Horowitz faces
test. J. Abnormal and Social Psychology, 45,504-506.
NATIONAL OPINION R~SEARCH CENTER. Opinion News, Aug. 1, 1948.
NEWCOMB, T. M., 1931. An experiment designed to test the validity of a
rating technique.,. Educational Psychology, 22, 279-289.
NEWCOMB, T. M., 1943. Personality and social change: attitude formation
in a student community. Dryden.
BIBLIOGRAPHY 601
NEWCOMB, T. M., 1946. The influence Of attitude climate upon some
determinants of information. J. Abnormal and Social Psychology, 41,
291-302.
NEWCOMB, T. M., 1947. Some patterned consequences of membership in
a college community. In Readings in social psychology, ed. by T. M.
Newcomb and E. L. Hartley. Holt.
NORTHROP, F. S. C., 1947. The logic of the sciences and the humanities.
Macmillan.
OGBURN, W. F., 1947. On scientific writing. American J. Sociology, 52,
383-389.
OSGOOD, C. K, G. J. SUC!, and P. H. TANNENBAUM, 1957. The measure-
ment of meaning. University of Illinois.
PACE, C. R., 1939. A situations test to measure social-political-economic
attitudes. J. Social Psychology, 10, 331-344.
PACKARD, V., 1957. The hidden persuaders. McKay.
PARSONS, T., 1937. The structure of social action. McGraw-Hill.
PARTEN, M. B., 1950. Surveys, polIs, and samples. Harper.
PATON, M. R., 1935. Selection of tabulation method, machine or manual.
J. Marketing, 6, 229-235.
PAUL, E., and C. PAUL (trans.), 1921. A young girl's diary. Preface with a
letter by Sigmund Freud. Thomas Seitzer.
PAYNE, S. L., 1951. The art of asking questions. Princeton University.
PEAK, H., 1953. Problems of objective observation. In Research methods
in the behavioral sciences, ed. by L. Festinger and D. Katz. Dryden.
PEASE, K., 1949. Machine computation of elementary statistics. Chartwell
House.
PINTNER, R., and G. FORLANO, 1937. The influence of attitude upon scaling
of attitude items. J. Social Psychology, 8, 39-45.
POFFENBERGER, A. T., 1942. Principles of applied psychology. Appleton-
, Century.
POLITZ, A., ind W. SIMMONS, 1949. An attempt to get the "not at homes"
into the sample without call-backs. ,. American Statistical Associa-
tion, 44, 9-31.
PROC~OR, C. H., and C. P. LOOMIS, 1951. Analysis of sociometric data. In
Research methods in social relations, Vol. II, first ed., ed. by M.
Jahoda, M. Deutsch, and S. W. Cook. Dryden.
PROSHANSKY, H. M., 1943. A projective method for the study of attitudes.
J. Abnormal and Social Psychology, 38, 393-395.
PROSHANSKY, H. M., 1950. Projective techniques in action research. In
Projective psycho'logy, ed. by L. K Abt and L. Benak. Knopf.
RAPER, A. F., 1933. The tragedy of lynching. University of North Carolina.
RAPKIN, C., E. GRIER, and G. GRIER, 1957. Group relations in Newark-
1957. Urban Research (New York).
602 BIBLIOGRAPHY
REMMERS, H. H., ed., 1934. Studies in attitudes. Purdue University studies

in higher education, 26. Bulletin of Purdue University, 35, No.4.
ROBINSON, "V. S., 1957. The statistical measure of agreement. American
Sociological Review, 22, 17-25.
ROETHLISBERGER, F. J., and vV. J. DICKSON, 1939. Management and the
work.er. Harvard University.
ROGERS, C. R., 1945. The non-directive method as a technique for social
research. American J. Sociology, 50, 279-283.
ROGERS, C. R., and R. F. DYMOND, 1954. Psychotherapy and personality
change. University of Chicago.
ROGERS, C. R., and B. F. SKINNER, 1956. Some issues concerning the con-
trol of human behavior: a symposium. Science, 124, 1057-1066.
ROHDE, A. R., 1946. Explorations in personality by the sentence comple-
tion method. J. Applied Psychology, 30,169-181.
ROSENBERG, M., W. THIELENS, and P. F. LAZARSFELD, 1951. The panel
study. In Research methods in social relations, Vol. II, first ed., ed.
by M. Jahoda, M. Deutsch, and S. W. Cook. Dryden.
ROSENFELD, E., 1958. The American social scientist in Israel: a case study
in role conflict. American ,. Orthopsychiatry, 28, 563-571.
ROTTER, J. B., 1951. Word association and sentence completion methods.
In Projective techniques, ed. by H. H. Anderson and G. L. Anderson.
Prentice-Hall.
ROTTER, J. B., and B. WILLERMAN, 1947. The incomplete sentence test
as a method of studying personality. J. Consulting Psychology, 11,
43-48.
ROWNTREE, B. S., 1941. Poverty and progress. Longmans.
SACKS, J. M., 1949. The relative effect upon projective responses of stimuli
referring to the subject and of stimuli referring to other persons.
J. Consulting Psychology, 13, 12-21. _
SAENGER, G., and E. GILBERT, 1950. Customer reactions to the integra-
tion of Negro sales personnel. International J. Opinion and Attitude'
Research, 4,57-76.
SAENGER, G., and H. PROSHANSKY, 1950. Projective techniques in the serv-
ice of attitude research. Personality, Symposium No.2, 23-34.
SAFFIR, M. A., 1937. A comparative study of scales constructed by three
psychophysical methods. Psychometrika, 2,179-198.
SANFORD, F. H., 1950. Authoritarianism and leadership: a study ot the
follower's orientation to authority. Institute for Research in Human
Relations (Philadelphia).
SANfORD, F. H., and L ROSENSTOCK, 1952. Projective techniques on the
doorstep. J. Abnormal and Social Psychology, 47,3-16.
SARGENT, S. S., and G. SAENGER, 1947. Analyzing the content of mass
media. J. Social Issues, 3,33-38.
BIBLIOGRAPHY 603
SAll.TRE, J. P., 1948. Portrait of the anti-Semite. Seeker and Warburg (Lon-
don).
SAYLES, L. R., 1954. Field use of projective methods. Sociology and Social
Research, 38,168-173.
SCHUTZ, W. C., 1952. Reliability, ambiguity, and content analysis. Psy-
chological Review, 59,119-129.
SCHWARTZ, W., 1948. Action research in a group work setting: a record of
a co-operative experience. Unpublished master's thesis at the New
York School of Social "Vork, Columbia University.
SCOTT, W. A., 1955. Reliability of content analysis: the case of nominal
scale coding. Public Opinion Quarterly, 19, 321-325.
SEELEMAN, V., 1940-1941. The influence of attitude upon the remem-
bering of pictorial material. Archives Psychology, 36, No. 258.
SEEMAN, M., 1947. Moral judgment: a study in racial frames of references.
American Sociological Review, 12,404-411.
SELLTIZ, C., 1955. The use of survey methods in a citizens campaign
against discrimination. Human Orgapization, 14, 19-25.
SELLTIZ, C., A. L. HOPSON, and S. W. COOK, 1956. The effects of situa-
tional factors on personal interaction between foreign students and
Americans J. Social Issues, 12, 33-44.
SELLTIZ, C., and associates, 1950. The acceptability of answers to anti
Semitic remarks. International J. Opinion and Attitude Research, 4,
353-390.
SHEATS LEY, P, B., 1951. The art of interviewing and a guide to interviewer
selection and training. In Research methods in sQcial relations, Vol.
II, first ed., ed. by M. Jahoda, M. Deutsch, and S. W. Cook. Dryden.
SHERIF, M., and C. SHERIF, 1953. Groups in harmony and tension. Harper.
SIEGEL, S., 1956. Nonparametric statistics for the behavioral sciences.
McGraw-Hill.
SIM~ONS, W. R., 1954. A plan to account for "not-at-homes" by combin-
ing weighting and call-backs. J. Marketing, 11,42-54.
SMITH, G. H., 1954. Motivation research in advertising and marketing.
McGraw-Hill.
SMITH, J. G., and A. J. DUNCAN, 1945. Sampling statistics and applications.
McGraw-Hill. I
SNEDECOR, G. W., 1956. Statistical methods. 5th ed. Iowa State College.
SOLOMON, R. L., 1949~ Extension of control group design. Psychological
Bulletin, 46, 137-~ 50.
SPITZ, R. A., and K. M. WOLF, 1946. Anaclitic depression: an inquiry
into the genesis of psychiatric conditions in early childhood, II. The
Psychoanalytic Sttidy of the Child, Vol. II.
STAR, S. A., and H. M. HUGHES, 1950. Report on an educational campaign:
the Cincinnati pl:m for the United Nations. American J. Sociology,
55,355-361.
604 BIBLIOGRAPHY
STEIN, M. L, 1947. The use of a sentence completion test for the diagnosis
of personality. J. Clinical Psychology, 3, 47-56.
STEINZOR, R, 1949. The development and evaluation of a measure of social
interaction. Human Relations, 2,103-122.
STEPHAN, F. F., and P. J. MCCARTHY, 1958. Sampling opinions: an anal-
ysis of survey procedure. Wiley.
STEPHENSON, W., 1953. The study of behavior: Q-technique and its meth-
odology. University of Chicago.
STEVENS, S. S., 1946. On the theory of scales of measurement. Science, 103,
677-680.
STEVENS, S. S., 1951. Mathematics, measurement, and psychophysics. In
Handbook of experimental psychology, ed. by S. S. Stevens. Wiley.
STEVENS, S. S., 1957. On the psychophysical law. Psychological Review,
64,153-181.
STOUFFER, S. A., 1949. An analysis of conflicting social norms. American
Sociological Review, 14, 707-717.
STOUFFER, S. A, et al., 1949a. The American soldier: ad;ustment during
army life. Studies in Social Psychology in World War II, Vol. I.
Princeton University"
STOUFFER, S. A, et a1., 1949b. The American soldier: combat and its after-
math. Studies in Social Psychology in Wqrld War II, Vol. II. Prince-
ton University.
STOUFFER, S. A, et al., 1950. Measurement and prediction. Studies in
Social Psychology in World War II, Vol. IV. Princeton University.
Technical recommendations for psychological tests and diagnostic tech-
niques, 1954. Prepared by a joint committee of the American Psycho-
logical Association, American Educational Research Association, and
National Council on Measurements Used in Education. Supplement
to Psychological Bulletin, 51, No.2, Pt. II.
THELEN, H., and J. WITHALL, 1949. T,hree frames of reference: a de-
scription of climate. Human Relations, 2,159-176.
THOMAS, W. I., and F. ZNANIECKI, 1918. The Polish peasant in Europe
and America. Badger.
THOMSON, G. H., 1946. The factorial analysis of human ability (2nd ed.).
University of London. .
THORNDIKE, R. L., 1949. Personnel selection test and measurement tech-
niques. Wiley.
THURSTONE, L. L., 1927. The method of paired comparisons for social
values. J. Abnormal and Social Psychology, 21,384-400.
THURSTONE, L. L., 1928. An experimental study of nationality preferences.
J. Genetic Psychology, 1,405-425.
I THURSTONE, L. L., 1929. Theory of attitude measurement. Psychological
Bulletin, 36,222-241.
BIBLIOGRAPHY 605
THURSTONE, L. L., 1931. The measurement of social attitudes. J. Ab-
normal and Social Psychology, 26,249-269.
THURSTONE, L. L., and E. J. CHAVE, 1929. The measurement of attitude.
University of Chicago.
TORGERSON, "V. S., 1958. Theory and methods of scaling. Wiley.
TRYON, R. C., 1955. Identification of social areas by cluster analysis. Uni-
versity of California Publications in Psychology, 8, No. 1.
TRYON, R. C., 1957a. Communality of a variable: formulation by cluster
analysis. Psychometrika, 22,241-260.
TRYON, R. C., 1957b. Reliability and behavior domain validity: reformu
lation and historical critique. Psychological Bulletin, 54, 229-249.
UNDERWOOD, B. J., 1957. Psychological research. Appleton-Century-Crofts
UNIVERSITY OF CHICAGO PRESS, 1949. A manual of style (II th ed.) .
WALKER, H. M., and J. LEV, 1953. Statisticalinference. Holt.
WARNER, W. L., and associates, 1941-1947. Yankee City series. Yale Uni.
versity. 4 vols.
WATSON, G. B., 1925. The measurement of fair-mindedness. Teachers
College, Columbia University Contributions to Education No. 176.
Teachers College, Columbia University.
WEITZ, J., and R. C. NUCKOLS, 1953. The validity of direct and indirect
questions in measuring job satisfaction. Personnel Psychology, 6, 487-
494.
WESCHLER, 1. R., 1950. An investigation of attitudes toward labor and
management by means of the error-choice method. J. Social Psychol-
ogy, 32,51-67.
WEHL, H., 1949. Philosophy ot mathematics and the natural sciences.
Princeton University.
WHITE, B. W., and E. SALTZ, 1957. Measurement of reproducibility. Psy-
chological Bulletin, 54, 81-99.
WHITE, R. K., 1949. Hitler, Roosevelt, and the nature of war propaganda.
, J. Abnormal and Social Psychology, 44,157-174.
WHITING, J. W. M., and I. L. CHILD, 1953. Child training and personality:
a cross-cultural study. Yale University.
WHYTE, W. F., 1951. Observational field-work methods. In Research
methods in social relations, Vol. II, first ed., ed. by M. Jahoda, M.
Deutsch, and S. W. Cook. Dryden.
WHYTE, W. F., 1957. On asking indirect questions. Human Organiza-
tion, 15, 21-23.
WILNER, D. M., E. ROSENFELD, R. S. LEE, D. L. GERARD, and I. CHEIN,
1957. Heroin use and street gangs. J. Criminal Law, Criminology and
Police Science, 48" 399-409.
WILNER, D. M., R. P. WALKLEY, and S. W. COOK, 1955. Human relations
in interracial housing: a study ot the contact hypothesis. University
of Minnesota.
606 BIBLIOGRAPHY
WOLFENSTEIN, M., and N. LEITES, 1950. Two social scientists view "No
Way Out"-the unconscious vs. the "message" in an anti-bias film.
Commentary, 10,388-391.
WOOLLEY, E. C., and F. W. SCOTT, 1944. College handbook of composi-
tion (4th ed.). Heath.
WORMSER, M. H., and C. SELLTIZ, 1951a. Community self-surveys. In
WORMSER, M. H., and C. SELLTIZ, 1951b. How to conduct a community
self-survey of civil rights. Association Press.
WRIGHT, Q., and C. J. NELSON, 1939. American attitudes toward Japa!!
and China, 1937-38. Public Opinion Quarterly, 3, 46-62.
WRIGLEY, C., 1957. Electronic computers and psychological research.
American Psychologist, 12, 501-509.
ZANDER, A., 1951. Systematic observation of small face-to-face groups. In
ZAWADZKI, B., 1948. Limitations of the scapegoat theory of prejudice. ,.
Abnormal and Social Psychology, 43,127-141.
ZEISEL, H., 1957. Say it with figures (4th ed.) . Harper.
ZEVIN, B. D., ed., 1946. Nothing to fear (Addresses. of Franklin D. Roose
velt). Houghton. .
ZOBER, M., 1956. Some projective techniques in marketing research. ,.
Marketing, 20,262-268.
INDEX
INDEX
Abstractions, high-level, 41 Anderson, G. L., 281n.

Abt, L E., 281n_ Anderson, H. H., 281n.
Accidental sampling, 515n., 516, 541- Angell, R., 323n.
544 Animals, social behavior of, 202
Ackerman, N. W., 62, 540 Anomie, 319
Act, in structured observation, 226 Anonymity, in questionnaire, 240
Action agency, 457 Anthropologists, observation by, 202,
final report to, 469-471 203-204
interim report to, 466-469 Anti-Defamation League of B'nai B'rith,
Action-oriented research, 457-466 ~4
audiences for, 471-472 Anti,-Semitism, measures of, 184-185,
presentation of, 466-472 357-358
and social policy, 474-477 Application, practical, of research, 457-
Action personnel, 457 466
contributions of, to research, 458- Applied research, 4-5
460 Argument-completion test, 290
resistance of, to research, 461-464, Army, integrated units in, 138
470-471 Attitudes
Adorno, T. W., 184,284,295,357 definition of, 146
Advertising, motivation research and, of foreign students, 33-35, 37,41-42
539 toward minority,groups, 295, 297
"After-only" experiments, 108-112 scales, 357-383
Age, as variable, 136 semantic differential as test of, 381
Agency, action, 457, 466-471 social, projective methods for study
Aggression, 483, 486 of,285-299
and prejudice, 62 social, structured disguised tests of,
Allp<;>rt, G. W., 235, 323n., 324, 327, 299-310
486 test of, and social pressures, 151
Alor, 284-285 Audio-introspectometer, 229
Ambiguity, in check lists, 569 Aussage test of perception and memory,
American Documentation Institute, 450 304
American Jewish Congress,. 58n. Authoritarian Personality, The, 184-
American Psychological Association, 185,295, 312
218n.,453 Authority figures, and family experi-
American Sociological Soci~ty, 54 ences, 161-162
American Soldier, The, 493 Average deviation, 412-413
Ames, A., Jr., 28 '
Analysis
anticipatory, 387-390
factor, 379n. I Baehne, G. W., 408n.
raw data as aid to, 432-440 Baldwin, A. L., 409n.
symbol,340 Barker, R., 115, 347
of variance, 124-125 Battle of Britain, Ill, 118-119
(See also Categories) "Before-after" design, offset, 116
609
610 INDEX
-'Before-after" experiment, llO, 112- complex, 396-398
122 establishment of, 391-396
with interchangeable groups, ll6- raw data as illustrations of, 434-435
117 for recording group interaction, 225
with one control group, 117-119 for unstructured material, 399-401
with three control groups, 121-122 (See also Coding)
with two control groups, 119-121 Category set, 392-393
Behavior, "eg<tneed," 231 Catharsis, 484
Beliefs,?o1festions to discover, 246-247 Cattell, R. B., 304, 379n., 409n.
Bell, J.ir, 281n. Causal inference
Bellak, L., 281n. (See Inference of causal relationship)
Benjam1n, L., 222 Causality, concept of, 80--83
Benne, K. D., 218n. Ceiling effect, in experiments, 113
Bennington College, 133, 134 Census data, 316, 321, 510
Berelson, B., 334 Central tendency, measures of, 410--
Berkoviitz, L., 23 412,414
Bettelheim, B., 438 Chapple, E. D., 229
Beveridge, W. 1. B., 200n., 464 Check-answers, to questionnaires, 567-
Beyle, H. C., 364 571
Bias, 50,67,69, 516n., 517, 520 Chein, 1., 82, 298,314,400,509-545
in closed questions, 261 Child, and maternal deprivation, 202
introduced by interviewer, 583-587 Child Development Abstracts and Bib-
public relations, 560 liography, 54
in questionnaires, 558 Children's Bureau, U.S. Department of
sampling, in interviewing, 582-583 Health, Education, and Welfare,
Binet, A., 498 54
Biographies, in mass media, 330-331 Christ, J. R., 128
Blankenship, A. B., 74n., 239, 251 Churchman, C . W., 80n.
Bogardus, E. S., 371, 372 Cincinnati, United Nations publicity
Borgatta, E. F., 268n. campaign, in, 117
Boyle's law, 85-86 Clark, K. D., 273, 274
Braithwaite, R. B., 480 Clark, M. P., 273, 274
Breakdown, 407 Class, social, and juvenile delinquency,
Bressler, M., 400 488-491
Brown, J. F., 296 Classification, for unstructured material,
Bruner, J. S., 335n. 399-401 '.
Brunswik, E., 125 Clinical interview, 266
Biihler, C., 328 Closed questions, 256-263
Burwen, L. S., 161 Cluster analysis, 321
Cluster sampling, 533-535
Cochran, W. G., 124n., 509n .
Cafeteria question, 568-569 .Coders, training of, 75,405
Campbell, D. T., 98n., 116n., 161, 300, Coding, 401-402
301 editing for, 403 .
Canter, R. R., Jr., 120-121 reliability in, 75-76, 363-365, 402-
Cantril, H. A., Jr., 28 406
Carter, H., 373 Coefficient alpha, 175
Cartoons, as visual aids in research, 45- Coefficient of correlation, 183-184
46,296 Coefficient of equivalence, 172-176
CartWright, D., 285 Coefficient of stability, 167-172
Case-study approach, 60 , Cohen, A., 488-499
(See also Insight-stimulating cases) Cohen, M. R., 31, 35, 80n.
Categories, 404-405 College Study in Intergroup Relations,
of absent data, 394, 396 296
INDEX 611
Collier, J., Jr., 275 Cross-tabulation, 407
Collier's, 330-331 Crown, S., 364
Collins, M. K, 10, 40, 129, 131, 138, Crutchfield, R., 247n., 262, 345n., 347
346 Cry the Beloved Country, 54-55
Comanche Indians, 486 Culture, and personality, 284-285
Commission on Community Interrela- Cumulative scales, 359, 370-378
tions, 58n. Current Sociological Research, 54
Communication, persuasive, 494-495
Communications, mass (See Mass com-
munications)
Concepts, 483
clarification of, 495 Darwin, Charles, 199
definition of, 41-42, 47 Data
Concomitant variation, 83-85, 87-88, census, 316,321
90,94-95 mass-communication documents a~,
Concurrent validity, 157 330-335
Condition nonquantified, 432 -440
alternative, 82-83 personal documents as sources of,
contingent, 82, 84 323-330
contributory, 82, 84,87 raw, in analysis and interpretation,.
determining, 80,95-96 432-440
necessary, 87 statistical records as sources of, 316.-
sufficient, 81 323
Confessions, of St. Augustine, 324, 325, Data collection
327 in descriptive studies, 66-67, 69-71
Conformity, 62,86,495 dummy tables for, 547
and standards of action, 249-250 sources, 546
Consistency (See Reliability) techniques of, 147-148
Constant errors, 154, 155 (See also Questionnaires, Interviews,
Construct validation, 156, 158-164 Observation, Protective methods)
Constructiveness, of play, 347-348 Davidsen, O. M.,)3, 34, 37, 38,41,43,
Constructs, 41 44
Consultant, researcher as, 473-474 Davidson, D., 333, 383
Content analysis, 335-342 Davison, W. P., 340
Contrast error, in ratings, 352 Dean, J. P., 59n., 212n.
Control group, 94, 96,123 Definitions
in "after-only" experiment, III operational,42n.
in "before-after" study, 114-115 in scientific theory, 485n.
one, in "before-after" study, 117-119 working, 42-44,147,495
selection of, 98-108 Delinquent subculture, 488-491
size of, 99n. Dembo, T., 115, 347
three, in "before-after" study, 121- Deming, W. K, 509n.
122 Democratic ideology, 358
two, in "before-after" study, 119-121 Dependent variables, 20
Cook, L. A., 296 Deri, S., 282n.
Cook, S. W., 22, 29, 55n., 128, 137, De Sales, R. R., 333
310, 386, 433, 447,1 509n., 546n., Descriptive statistics, 50, 65-78, 142-
574n. 143,410-414
Coombs, C. H., 187, 189n., 195n., Descriptive study
349n., 377, 383 analyzing results in, 75-77
Cottrell, L. S., Jr., 349n. I collecting and checking data in, 74--
Cox, C. M., 124n. 75
Criswell, J. H., 271 Design
Cronbach, L. ]., 156n., 159, 168n., experimental,94-125
175n., 181, 345, 378n. nonexperimental, 127-142
612 INDEX
Design- ( cont' d) Type I and Type II, 417-419
representative, 125-127 in use of rating scales, 351-352
research, 9,50,88-94 Ethical Standards of Psychologists,
Deutsch, M., 10, 40, 106, 129, 131, 2l8n.
138, 233, 234, 346, 546n., 574n. Evidence, and adequacy of research de.
Deviant cases, use of in exploratory signs, 88-94
studies, 62 Experience survey, 55-59
Diaries, as data sources, 325-326, 328 analysis of examples in, 59-65
Dickson, W. J., 268n., 319-320,497 Experimental design, 89-92
Differential scales, 359-365 types of, 108-122
Direct questions, 290-292, 548 Experimental group
Discrimination
selection of, 98-108
against Chinese, study of, 250-251
against Negroes, in restaurants, 67- size of, 99n.
71 Experiments, 422-423
(See also Prejudice) causal inference from, 94-98
Displacement, 483 ceiling effect in, 113
Dissertation Abstracts, 54 with two or more variables, 122-125
Distribution Exploratory study, 50, 51-65, 78
frequency control, 107-108 observation in, 204-205
of individual scores, 412-414 survey of literature in, 53-55
of sample means, 525 Eysenck, H. J., 364
Distributionfree statistics, 415n.
Dixon, W. J., 418n. 1 '.
Doctoral Dissertations Accepted by Face validity, 164-166
American Universities, 54 "Faces" test,- 272-273
Donard, J., 483 Factor analysis, 379n.
Dolls, as aids in interviewing, 273-274 Facts, questions to discover, 244-246
in projective techniques, 283, 298 Fantasy, in Marienthal study, 436
Dropouts, in sampling, 538n. "Feeling thermometers," 554
Drug addiction, 82, 266-267, 540 Feelings, questions to discover, 247-248
DuBois, Cora, 284 Ferguson, L. W., 364, 383
Dudycha, G. J., 363 Festinger, L., 62, 219, 270, 373n., 509n.
Duncan, A. J., 413n. Fiction, insights in, 54-55
Durkheim, E., 318 Figure-drawing test, 283
Dymond, R. F., 379 Film
anti-prejudice, 331-333
used in "after-only" design, 111-112 .
Eckert, J. W., 408n. Fisher, R. A., 79, 100, 124n.
Editing, for coding, 403 Fiske, D. W., 161, 355, 356n.
Edwards, A. L., 106n., 124n., 151,240, Fiske, M., 264,
357n., 362, 368n., 375 Flesch, R., 452n.
"Ego need" behavior, 231 Focused interview, 264-266
Einstein, A., 25 Foreign students, study of, 33-35, 37,
Element, population, 510 41-42,386-389,427
Equivalence, of measurements, 172- Forlano, G., 364
177, 186 Formulative study (See Exploratory
'lS transitive and symmetrical, 190n. study)
"Error, generosity," 352, 355 Forster, E. M., 55
Errors, , Foundations, and research grants, 30
constant, 154, 155 Frank, L. K., 280
random, 154, 155 Franz, J. G., 299n.
sampling, 534n. Freeman, F. S., 345-347
INDEX 613
Frequency distribution control, 104, Hammond, K. R., 126, 301-302, 312-
107-108 313
Freud, S., 35, 59, 64, 486 Hansen, M. H., 509n.
Friendships, Westgate study of, 270- Harding, J., 58n., 376
271 Hartley, E. L., 298-299
Fromme, A., 296 Hastorf, A. H., 28
Frustration, 483 Havel,J., 128
and play of young children, ll5, 347- Hawthorne studies, of worker produc
348 tivity, 319-320,497
relation of, to prejudice, 438-439, Hebb, D.O., 202
481-486 Helgerson, E., 273
Frustration-aggression theory, 481-486 Hempel, C. G., 485
F-scale, as substitute measure, 310-311 Henry, W. E., 285
Fundamental measurement, 194-195 Hinckley, E. D., 364, 365
Funds, for research, 30 Hitler, A., 333-334, 341
Hogrefe, R., 55n., 58n., 298, 314, 377
Homans, G., 86
Gangs, street, 490, 491,540 Hopson, A. L., 386,447
Gardner, B_ B., 294 Horowitz, E. L., 272, 273, 304
Generalization, about population, 414- Horowitz, R. E., 273, 304
422 Horwitz, M., 285
Generosity error in rating~, 352, 355 Housing studies, 10-23, 39-40, 129-
Gerbrands, H., 229 132, 137-139, 397-398, 434-435,
Getzels, J. W., 290 439-440,476
Gibbon, E., 482 Hovland, C. I., lll, 113, 308, 364,481,
Gilbert, E., 309 494
Goldsen, J- M., 335n_ Hughes, H. M., 117
Goldston, J., 222 Human Relations Test, 296
Good, C. V., 443n. Hunt, J. MeV., 28
Goodman, M. E., 273 Hurwitz, W. N., 509n.
Gordon, D. H., 262 Huxley, T. H., 480
Gottschalk, L., 323n., 326 Hyman, H. H., 67n., 127n., 213n.,
Granneberg, R. T., 365 423n., 474n.
Graphic scale, in rating, 353 Hypothesis
Graphs, in research report, 453 as basis for establishing categories,
Green, B. F., 195n., 357n. 399-400
Greenwood, E., 93, 104n. causal, 36, 51
Grier, E., 245 definition of, 35
Grier, G., 245 formulation of, 37-40
Gross, N., 258 function of, 35-36,47
Group and nonexperimental studies, 127-
interaction, recording, 225-230 142
interviews, 239n. null,415-418
Guetzkow, H., 231, 285, 406n. and theory, 480, 485
Guilford, J. P., 166n., 175n., 183n.,
345-347, 355 .
Gulliksen, H., 166n., 183n., 195n.
Guttman, L., 373-374, 375 Immediate Memory Test, 304
Implementation of results (See Appli-
cation)
Indexing systems, for observation reo
Haire, M., 293,314 ords, 211-212
Halbwachs, M., 323 Indirect questions, 290-292, 548, 566-
Halo I';!if-er.t, in ratings, 351-352 567
614 INDEX
rnfeld, L., 25 structured,255-263

Inference structured, partially, 263-268, 549
of causal relationships, 83-98, 127- unstructured, 267-268
128,142-143,422-424 (See also Questionnaires)
in indirect tests, 311, 312 Ipsative interpretation, 4090.
Information and Education Division, Isolates, use of in exploratory studies,
War Department, 111, 138 62
Information test, 301-302 Ittelson, W. H., 28
Inhibition, 483
Insight-stimulating cases, 59-65
Interaction chronograph, 229 Jahoda, M., 62, 219, 433, 509n., 540,
Interaction-potential, 388,427-429 546n., 574n.
Interaction recorder, 229-230 Jahoda-Lazarsfeld, M., 436
Interaction, social, of Negroes and Janis, I. L., 494
whites, 222-223 Janowitz, M., 438
Interim reports, of action-oriented re- Japanese, attitudes toward, 303, 340.
search, 466-469 341
Internal consistency, of measure, 184- Jaspers, K., 4
186 Jennings, H. H., 270
"Internationalism Scale," 366 Jews, and non-Jewish interviewers, 584
Interpretation, anticipatory, 389-391 Jones, E., 64, 305 ~
(See also Categories) Judgments, as indicator of attitude,
Interracial housing (See Housing 304-310
studies) lung, C., 288
Interval scale, 189, 19l-194, 196, 198, Juvenile_delinquency, 488-491
359
Jllterviewer
~ -biasing factors introduced by, 583- Kaplan, A., 33 5n.
587 Kardiner, A., 284, 486
introduction by, 575-576 Katz, D.,'237n., 320, 509n.
Interviewing Katz, I., 222
art of, 574-587 Keller, H., 327-328
askl'ng questions during, 576-579 Kelley, H., 365,494
creating friendly atmosphere in, 575- Kelley-Remmers Scale, 362
576 Kelly, E. L., 355, 356n.
omissions during, 580-581 Kendall, P. L., 264, 423n.
reporting response during, 213, 579- , Kenney, K. E., 368n.
582 Kerr, M., 289
sampling during, 582-583 Kilpatrick, F. P., 375
visual aids in, 272-276 Kinsey, A. C., 243n.
Interviews Klineberg, 0., 134, 135, 136, 485n.
advantages of; 241-243 I Kluckhohn, C., 323n.
clinical, 266 Kohler, R., 305
focused, 264-266 Komarovsky, M., 66
group, 239n. Kornhauser, A., 546-587
nondirective, 267-268 Kramer, B. M., 248Il., 249
open-ended, 348-349 Krech, D., 247n., 345n., 347
personal history, 266 Kurtosis, 413n.
pilot, 18 Kymograph, 224
question content in, 243-255
with respondents in experience survey,
56-59 Lambert, R. D., 400
standardized,255-263 Landis, K. M., 317
stress, 268n. LaPiere, R. T., 250-251
INDEX 615
Lass, A. H., 452n. McNemar, Q., 99n.
Lasswell, H. D., 335n., 336, 340, 406n. Madow, W. G., 509n.
Law, scientific, 480 Magazines, as data source, 330-335
Lazarsfeld, P. F., 132n., 219, 253, 262, Mail questionnaire, returns to, 241-242
320, 423n., 496 Manual of Style, A, 453
Leadership Manual tabulation, 407-408
among children, 228 Margenau, H., 164
sociometric study of, 270 Marginal individuals, use of in explora.
study of, 125 tory studies, 61
Leahy, A., 408n. Marginals,407
Learning, as indicator of attitude, 305- Marienthal unemployment study, 219,
306 320,436
Lee, A. M., 62, 249 Market research, 291, 293, 296, 297-
Lee, R. S., 266 298,539
Leighton, A. H., 475-476 Mason, W., 258
Leisure, study of, 66-67 Mass communication
Leites, N., 331, 335n., 406n. analysis, purpose of, 334-335
Letters, as data source, 328-329 content analysis of, 335-342
Lev,J.,418n. Mass media documents, as data source,
Levine, J. M., 305 330-335
Levy, D. M., 62-63, 294 Massey, F. J., Jr., 418n.
Lewin, K., 115, 347, 484n., 498n. Matching, 102-108
Lewis, H. B., 308n.
procedures, 99
Likert, R., 274, 276, 306, 366, 464n.,
of subgroups, ~38-1?9 . .
471n.,474n. Matching techmque, III projective test
Likert-type scale, 366-369
ing, 294
Lindgren, K J., 211 Maternal deprivation, study of, 202
Lindquist, K F., 124 Maternal overprotection, 63
Lindzey, G., 268n. Mean, arithmetic, 410, 411
Lippitt, R., 229, 464n., 471n., 474n. Mean scores, sampling distribution of,
Lists, check, ambiguity in, 569
524,527/
Literature, survey of, 53-55
Measurement
Loevinger, J., 373n., 383
definition of, 187
Logical error, in ratings, 351n.
Loomis, C. P., 270n. fundamen tal, 194-195
Lowenthal, Leo, 330-331 reliabilityof,166-186
~yalty probes, study of, 433 status of, in social sciences, 195-197
Luborsky, 1. B., 409n. validity of, 154-163
Lumsdaine, A. A., Ill, 113 (See also Scales)
Lundberg, G., 66 Measurement procedures, 147
Lynching and cottoil price, relation Measuring instruments, variations in
between, 481-482 scores on, 149-154
Median, 411
Meehl, P., 156n., 159
McCarthy, P. J., 509n., 536 Memory, as indicator of attitude, 304-
McClelland, D. C., 411 310
Maccoby, K, 291,3541 Merei, F., 228
Maccoby, N., 291, 354 Merton, R. K., 39, 219n., 264, 309, 397,
MacCrone, I. D., 360-361, 364 434-437, 439, 487, 493-496
McEachern, A., 258 Mintz, A., 482n.
McGranahan, D. V., hOn. Misperception method, of attitude test
McGregor, D., 187 ing, 304-305
Machines, for tabulation, 408-409 Mobility, status, 38-39
McInerv. M. A., 66 Mode. 411-412
616 INDEX
Mohs' scale of hardness, 192 No Way Out, 331-333
Moreno, J. L., 268n., 271 Nominal scales, 189-190, 195, 198
Morgan, J. J. B., 289, 290-291, 303 Nonconformity, 162
Morris. R. T., 33, 34, 37, 38,41,43,44 Nondirective interview, 267-268
Moses and Monotheism, 36 N onparametric statistics, 415 n.
Motivation research, 539, 560 Nonprobability sampling, 515-521
Movies with probability sampling, 535-537
antiprejudice, 331-333 population problems in, 541-544
used in "afteronly" design, 111-112 special applications of, 537-545
Mowrer, O. H., 378n. (See also Probability sampling)
Multiple choice questions, 557-569 Nonquantified data, use of, 432-440
Multistage sampling, 534 Normative interpretation, 409n.
Miinsterberg, H., 206 Norms, social, 249-251
Murphy, G., 274, 276, 305, 306, 354, Northrop, F. S. C., 52-53
366 Note-taking, during unstructured ob-
Murphy, L. B., 354 servation, 210-213
Murray, H. A., 289, 290-291, 352, 354 Nothing to Fear, 333
Mussen, P. H., 273 Novels, insights in, 54-55
My New Order, 333 Nuckols, R. C., 292
Null hypothesis, 415-418
Numbers, random, 99, 522
Nagel, E., 31, 35, 80n. Numerical scales, 347-349
Narcotics, addiction to, 82, 266-267,
540
National Opinion Research Center, 329, Observation
487 increasingllccuracy of, 212-217
National status, 42 limitations of, 203-204
index of, 43 purposes of, 204-206
Native Son, 55 as scientific technique, 200-203
Negroes structured,221-234
attitudes toward, 137-139, 309, 377, and theory, 481-487
in interracial project, 346, tests of, unstructured, 207-221, note taking
304, 306-307 during, 210-213
discrimination by restaurants, 67-71 Observer, 115
I.Q.'s of, study of, 134-135 note-taking by, 210-212
lynching of, and cotton prices, 481- relation to observed, 217-221, 233-
482 234 ~
in Northern army camps, 496 training of, 232
prejudice toward, 486 (See also Participant observer)
sociometric analysis of children, 271 Ogburn, W. F., 443n.
and white interviewers, 384 Open-ended interviews, 348-349
and whites, in social interaction, 222- Operational definitions, 42n.
223 tSee Working definitions)
(See also Housing,studies) "Ordered metric" scale, 189n.
Nelson, C. J., 340 Ordinal scales, 189, 191-193, 196, 198
New York interracial housing projects, Osgood, C. E., 380-382
16,21 Overprotection, maternal, 63
Newark, N.r, interracial housing proj-
ects, 16,19,21
New~omb, T. M., 133-l?4, 137, 300, Pace, C. R., 251
:35ln., 354 Packard, Y., 289n.
, Newspapers Paired comparisons, in scale coqstruc-
content analysis of, 333, 340-341 tion,359
sampling of, 338 Panel studies, I32n.
INDEX 617
Parsons, T., 318 Precision control, 104-105
Parten, M. B., 67n, 239, 242n., 403n., Precoding, 571
408n., 443n. Prediction, as goal of research, 4, 420n.
"Partially ordered scale," 189n. Predictive validity, 157
Participant observer, 204-205, 207, Prejudice, 62, 271, 331-333, 486
212-215,217 relation of, to frustration, 438-439
in relation to observed, 217-221 (See also Discrimination)
Passage to India, A, 55 Pretest
Paton, A., 54 of datacollection procedures, 70-71
Paton, M. R., 408n. of questionnaire, 550-551
Patterns of Social Life, 439-440 Probability, 136,414
Paul, E. and C., 326n. Probability sampling, 514-515, 521-
Payne, S. L., 552n., 562 535
Peak, H., 357n. and nonprobability sampling, 535-
Pease, K, 408n. 537
Perception, as indicator of attitude, drop-outs in, 538n.
304-310 (See also Nonprobability sampling)
Perception-Span Test, 304 Probes, in interviewing, 257, 548-549
Personal history interview, 266 Proctor, C. H., 270n.
Personality Productivity, Hawthorne study of, 321-
projective techniques in study of, 322
281-285 Projective methods
Personnel advantages of, 287-288
estimate of, for research project, 502, definition cif, 280
505-506 direct and indirect questions in, 290-
(See also Action personnel ) 292
Photographs in study of personality, 218-285
(See Pictorial techniques) in study of social attitudes, 285-299
Piaget, J., 27 validation of, 311-;314
Pictorial Recognition Test, 304 Propaganda techniques, 333-334
Pictorial techniques Proshansky, H. M., 282n., 294, 309,
in interviewing, 272-276 313,349
in projective tests, 294-298 Psychoanalysis, 62
Pilot interviews, 18 Psychodrama, 299
(See also Pretest) Psychological Abstracts, 54
Pintner, R., 364 Public relations, 560
Play Publication Manual, 450n., 452n., 453
constructiveness of, 347-348 Purposive sampling, 520-521
techniques, as projective tests, 298-
299
Poffenberger, A. T., 354 Qsort, 378-380
Policy, social, relation of research to, Qualitative data (See Nonquantified
474-477 data)
Polish Peasant in Europe and America, Quantification
The, 327 in mass communications analysis,
Politics, participation in, 39-40 334-337
Politz, A., 538n. of observation, 203-204
Population, 73, 509-510 Quartile deviation, 412-413
generalization about, 4 ~ 4-422 Question content, in questionnaires,
in nonprobability sampling, 541-544 243-255,552-561
Portrait of the Anti-Semite, 55 Questionnaires
Poverty, study in York, England, 71 advantages of using, 238-241
Pragmatic validity, 156-158, 163-164, anonymity in, 240
166,179,311 bias in, 558
618 INDEX
Questionnaires- ( cont' d) Recording problems, in unstructured

check answers to, 567-571 observation, 210-213
editing, 551-552 Recording, tape, use of, 230, 232
first draft of, 549-550 Reference group theory, 493-494
guide for construction of, 552-573 Relationship, spurious, 424-427, 429-
information sought in, 546-548 432
mail, returns to, 241-242 Relative deprivation, 493
personal data in, 574 Reliability, 50, 277
pretest of, 550-551 in coding, 405-406
question content in, 243-255, 552- in content analysis, 341-342
561 in distinguishing among individuals
revision of, 550 and among groups, 180-182
sociometric, 269-270 and errors, 154
standardized,255-263 of measurements, 166-186
types of, 548-549 methods of increasing, 182-186
Questions in rating scales, 353, 354
closed, 256-263, 567-570 in relationship between variables,
direct, 548 179-180
fixedalternative, 256-263 and validity, 178-179
followup, 548-549 Remmers, H. H., 362
form of response to, 567-571 Replication, of research, 46-47
indirect, 548 within study, 122
openended,257-263 Report, research
sequence of, 572-574 content of, 443-450
wording of, 561-567 short, modification for, 449-451
Quota sampling, 516-520, 535-536, writing style of, 451-454
538n., 582-583 Representative design, 125-127
Representative sampling plan, 513, 528
Racial identity, in children, 271 Research
Random errors, 154, 155 actionoriented, 457-472, 474-477
Random numbers, 99, 522 administrative details in, 9, 502-503
Random sample, 113 advertising and, 539
simple, 521-526 applied,4-5
stratified,526-533 basic, 4-5
Random sampling, 5I5n. contribution of, to theory, 492-498
Randomization, 98-102,104,106,113 design, 50, 88-94 ~
substitutes for, 136-142 exploratory study in, 10-12
Range, 412 formulating problem for, 30-?$
Rank, and conformity, 86
market, 291, 293, 296, 297-298, 5~9
Raper, A. F., 481
Rapkin, C., 245, 246 personnel needed for, 502, 505-506
Raters, 364-365 popularization of reports, 472
qualifications of, 355 practical application of, 457-466
Rating scales . prediction, as goal of, 4, 420n.
comparative, 349-350 projective techniques in, 280-299
construction and use of, 351-356 purpose of, 2
graphic, 345-346 questions asked in, 2-4
itemized,347-349 replication of, 46-47
numerical,347-349 report, 442-454, graphs in, 453
in structured observation, 227 resistance to, by action personnel,
\ validity of, 355-356 . 461-464, methods of reducing,
Rating, by self, 350-351 464-466, 470-471
Ratio scale, 189, 194-195, 196, 198, selecting topic for, 26-3(1
345-356 and social policy, 474-477
INDEX 619
Research- ( cont' d) Samples
survey, 126 accidental, 515n., 516, 541-544
theory as basis for, 487-492, 498- cluster, 533-536
499 quota, 516-520, 535-536, 538n.,
time budget for, 502-507 582-583
topic, 9, 13-14,26-35,47 random, 113, 515n., simple, 521-526
values and, 3 restriction of, 139-141
Research, action-oriented, 457-466 selection of, 71-75
presentation of, 466-472 self-selection in, 101-102
and social policy, 474-477 size of, 422n.
Research desi~ stratified random, 104, 526-533
""l::!!CIequacy 0; as source of evidence, systematic, 73, 74n., 523n.
_8&.Q4 weighting of, 517-518, 529
definition of, 50 Sampling
(See also Design) definitions and concepts of, 509-515
Research Relating to Children, 54 distribution, 525
Research Reports, of the Anti-Defama- drop-outs in, 538n_
tion League of B'nai B'rith, 54 error, 534n_
Researcher, as consultant, 473-474 fraction, 526n.
Respondents by interviewer, 582-583
accuracy of, 245-246 from mass communications, 337-339
for experience survey, 55-56 multi-stage, 534
obtaining response from, 577-579 nonprobability, 515-521, 537-545
perception of interviewer by, 583- plan, 513
587 probability, 514-515, 521-535
(See also Interviewing) probability with nonprobability, 535-
Response, to questionnaires 537
obtaining, 577-579 purposive, 520-521
reporting, 579-582 quota, 516-520, 535-536, 538n.,
Restaurants, and Negro discrimination, 582-583
67-71 random, 5-15n., 521-526
Returns, to mail questionnaires, 241- random, stratified, 526-533
242 statistics, 410, 414-422
Riecken, H. W., 219 systematic, 523n.
Robinson, W. S., 406n. Sanford, F_ H., 297 -
Roethlisberger, F. J., 268n., 319-320, Sargent, S. S., 335n.
i 497 Sartre, J.-P., 55
Rogers, C. R., 218n., 268n., 379 Saturday Evening Post, The, 330-331
Rohde, A. R., 289n. Sayles, L. R., 294-295
Roosevelt, F. D., 333-334 Scale analysis, 373-374
Rorschach Test, 282 Scale-discrimination technique, 375-377
Rosenberg, M., 132n. Scales
Rosenfeld, E., 27 cumulative, 370-378
Rosenstock, J., 297 differential, 359-365
Rosenzweig Picture-Frustration Test, interval, 189, 193-194, 196, 198
296-297 Likert-type, 336-369
Rossi, A. H., 494 of measurement, 186-189
Rotter, J. B., 289n. nominal, 189-190, 195, 198
Rowntree, B. S., 71 numerical,347-349
"ordered metric," 189n.
Sacks, J. M., 290 ordinal, 189, 191-193, 196, 198
Saenger, G., 309, 310, 335n. "partially ordered," 189n.
Saflir, M. A., 359 . rank-order, 350
St_ Augustine, 324, 325, 327,442 rating, 227, 345-356
620 INDEX
Scales-( confd) Social stratification theory, 38

ratiu.189.194-195. 196. 198 Sociodrama, 299
summated. 359, 366-370 Sociological Abstracts, 54
Thurstone-type, 359-365 Sociometric techniques. 268-271
unidimensional,373-378 Solomon, R. L., 119, 120, 121
Scalogram method. 373-375 Spearman-Brown formula, 183-184
Schachter. S., 62,219,270 Specific category scales, 347-349
Schutz. W. C., 406n. Spitz, R. A., 202
Schwartz, W . 298-299, 462 Split-half method of computing relia-
Science bility, 174-175
chance and discoveries in, 200n. Spurious relationship, between variables,
and theory, 480 424-427
as unfinished process. 4 criteria for detecting, 429-432
Scott, F. W., 453 Srole, L., 317
Scott, W. A., 406n. Stability coefficient, 167-168
Sears, R. R., 481 Stability of results, in measurement.
Seeleman, V., 304 168-172
Seeman, M., 306-307,313 Standard deviation, 412-413
Segregation, study of, 317 Standards of action, questions to dis-
(See also Housing studies) cover, 249-252
Segregation in Washington, 317 Star, S., 117
Self-ratings, 350-351 Statistical records
Self-report, limitations of, 237-238 availability of, 316-317
Self-selection pitfalls in using, 322-323
in interracial housing project, 138 principles for using, 317-322
in samples, 101-102 Statistics -
Self-survey, 460 descriptive, 50, 65-78,142-143,410-
Selltiz, C., 55n., 67, 252,386,447 414
Semantic differential, 380-383 nonparamet~ic, 415n.
Semmelweis, 1. P., 464 sampling, 414-422
Sentence-completion test, 283, 289-290 Status, 38 .
Sheatsley, P. B., 546-587 gain and loss, 42
Sheffield, F., Ill, 113 national, 42, 43
Sherif, C., 219 Stein, M. I., 289n.
Sherif, M., 219,308,364 Steinzor, B., 224
Shorthand recording, of questionnaire Stephan, F. F., 509n., 536 ,
responses, 581 n. Stephenson, W., 378n.
Siegel. S., 193n., 383, 415n. Stereotypes, 554
Significance, statistical, 72-73,421-422 Stevens, S. S., 145, 187, 189n., 195,~197
test of, 101, 136,415,416 Story-completion test, 290
Simmons, W. R., 538n. Stouffer, S. A" 132, 250, 373n., 493,
Skewness, 413n. I 496
Skinner, B. F., 218n. Stratified random samples, 526-533
Small groups, Thematic Apperception Stratum, population, 510
Test for study of, 285 Street Ceirner Society, 215
(See also Structured observation) Stress interviews, 268n.
Smith, G. H., 290, 291, 296, 297, 298 Structured disguised tests of social atti-
Smith, J. G., 413n. tudes, 299-311
Snedecor, G. W., 509n. Structured interview (See Interviews)
Social change, 62,133-134 Structured observation, 221-234
Social cohesion, and suicide, 319 increasing reliability of, 230-233
Social distance scale, 371-372 recording during, 228-230
Social research Students, foreign, study of, 33-35, 37,
(See Research) 41-42,386-389,427
INDEX 621
Sub-population, 510 Time order of variables, 130-133
Suci, G. J., 380, 382 Tomkins-Horn Picture Arrangement
Suicide Test, 283
and social cohesion, 318-319 Too Late the Phalarope, 54-55
statistics on, 323 Topic, research
Suicide, Le, 318 formulation of, 9,13-14,30-35,47
Summated scales, 359, 366-370 selection of, 26-30
Suppes, P., 383 Torgerson, W. S., 195n.
Survey, experience, 55-59 Training
Survey research (See Descriptive study) of coders, 75, 405
Survey Research Center, 348 of observers, 232-233
Swanson, G., 218n. Transition, individuals in, 61-62
Symbol analysis, 340 Tryon, R. C., 166n., 175n., 321
Systematic sampling, 523n. Type I error, 417-419
Type II error, 417-419
Tables, in research report, 453
Tabulation, 406-409
Tannenbaum, P. H., 380,382 Unemployment, Marienthal study of,
Tape recordings, for structured observa- 219,320
tion,230,232 fantasyand,436
Telephone surveys, 239n. Unidimensional scale, 186, 373-378
Tension systems, 498 United Nations publicity campaign, 117
Tertullian, 481, 484 United States Army, 27
Test-retest procedure, 170 United States Treasury, 27
Tests University Microfilms, 54
immediate memory, 304 Unstructured interview (See Inter-
inferences in indirect, 311-312 views)
information, 301-302 Unstructured material, categories for,
sentence-completion, 283, 289-290 399-401
word-association, 283, 288-289 Unstructured observation, 207-221
Thelen, H., 224, 229
Thematic Apperception Test, 282, 284,
294, 312 Validation
for tests of small groups, 285 of questionnaires and interviews, 277
Theory of projective techniques, 311-314
contribution of research to, 492-498 Validity
definition of, 480 concurrent, 157
frustration-aggression, 483-484 construct, 158-164
and hypotheses, 36-38 and errors, 154
observation and, 481-487 face, 164-166
research and, 498-499 of measurements, 154-163
Thermometers, feeling, 55* pragmatic, 156-158, 163-164, 166,
Thielens, W., 132n. 179,311
Thomas, W. 1., 327 predictive, 157
Thompson, W. R., 202 and rating scales, 355-356
Thomson, G. H., 379n. , and reliability, 178-179
Thorndike, R. L., 150n., 175n. in scaling, 383-384
Thurstone, L. L., 193n., 359, 361n., Values
365n. appealed to, by Hitler and Roosevelt,
Thurstone scale, 359-365, 368-369 334
Time and application of research, 475-476
estimate of, for researcl,J project, 502- of Army, and promotion of soldiers,
507 132-133
unit, in structured observation, 226 change in, 133-134
622 INDEX
Values-( cont'd) Weyl, H., 187
in research, 3 White, B. W., 341, 375n.
in selection of research topic, 28-29 White, R. K., 333
in social issues, 7n. Whitehead, A. N., 28,456
Variability, measures of, 412-413 Why, ways of asking, 253-255
Variables Whyte, W. F., 207n., 215, 217, 274-
association among, 413-414,423-424 275
causal, 128-136 Willerman, B., 289n.
criterion, 87 Wilner, D. M., 22, 137, 310
dependent, 20, 87,94,95 Withall, J., 224
effect, 87 Wolf, K. M., 202
independent, 87-88 Wolfenstein, M., 331
reliability in relationship between, Woolley, E. C., 453
179-180 Wordassociation test, 283, 288-289
time order of, 85-87, 429-430 Wark output, Hawthorne study of, 497
Variance, analysis of, 124-125, 431n. Workingdefinitions,A2-44, 147,495
Variation Wormser, M. H., 460n.
concomitant, 83-85, 87-88, 90, 94- Wright, Q., 340
9, Wright, R., 55
Visual aids in interviewing, 272-276 Wsigley, C., 40Sn.
Voting
and demographic pattern, 321-322
predictions of, 511-512 Yale Communication Research Pro-
gram,494
Young Girl's Diary, A, 32.5-32.6
Walker, H. M., 418n.
Walkley, R. P., 22,137,310 Zander, A., 221n., 229
Warner, W. L., 317 Zawadzki, B., 486
Watson, C. B., 302, 306 Zeisel, H., 219, 254,320,436, 470n.
Weber, M.,40 Zero-point,absolute, 194-195, 197, 198
Weighting, of sample, 517-518, 529 Zevin, B. D., 333
Weitz, J., 292 Znaniecki, F., 327
Weschler, 1. R., 302 :Wber, M., 298
I
-
PROPE}rr~
The Kansas S.ate
-Uniu{rsity ~
OF
of
Agriculture & ApplIed Science ':'
TC M India. 1

Claire Selltiz - Research Methods in Social Relations

Uploaded by

Copyright:

Available Formats

Claire Selltiz - Research Methods in Social Relations

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Claire Selltiz - Research Methods in Social Relations

Uploaded by

Copyright:

Available Formats

RESEARCH METHODS IN SOCIAL RELATIONS

CLAIRE SELLTIZ MARIE JAHODA

MORTON DEUTSCH STUART W. COOK

ISIDOR CHEIN HAROLD M. PROSHANSKY

PUBLISHED FOR THE SOCIETY FOR THE

PSYCHOLOGICAL STUDY OF SOCIAL ISSUES

Holt, Rinehart and Winston

Grateful acknowledgment is made to the following for the use of copyrighted

University of Minnesota Press for the paragraphs from Interracial Housing:

ALTHOUGH THERE HAVE BEEN many technical publications on specific

Research Center for Human Relations

IN REVISING Research Methods in Social Relations, we have had two

sion of these concepts has taken account of these new formulations. At

1. The Research Process 1

2. Selection and Formulation of a Research Problem 25

Causal Inference from Other Study Designs, 127

5. Some General Problems of Measurement 145

8. Data Collection: III

10. Placing Individuals on Scales 343

11. Analysis and Interpretation 385

12. The Research Report 441

13. The Applic,ation of Social Research 455

14. Research and Theory 479

A. Estimating the Time and Personnel Needed for a

C. Questionnaire Construction and Interview Procedure 546

The Importance of Knowing How Research Is Done'

Major Steps in Research

Organization of the Text

T HE PURPOSE OF RESEARCH is to discover answers to questions

of information alone but cannot be answered by research at the present

systematic body of knowledge about social interactions. Moreover,

The Importance of Knowing How Research Is Done

, Even though "research cannot provide any Ena1 answers to the

methods, there is the satisfaction of acquiring a new intellectual tool.

curing juvenile delinquency by fining the parents of delinquents; or if

project. As an illustration, we shall use a study of interracial housing

Not only may such research be socially useful ... but it

influences in the subject's social milieu . . . . there is reason to

With a census of problems and issues in the field of inter-

studying and from a knowledge of basic socio-psychological

We designed our study to overcome as many as possible of

provided matching segregated bi-racial projects. But interviewing

Because 9f the limited number of interviews that our re-

evidence on the practical question of the effects on attitudes of living

occupancy pattern1might produce the observed differences in attitudes

the Negro women as neighbors in the building, or on benches outside

Organization of the Text

;lnd points to the significance of theoretical development for the prac-

Selecting a Topic for Rese~rch

Relating the Findings to Other Knowledge

Selecting aTopic tor Research

T HE RANGE OF POTENTIAL TOPICS for social research is as broad as

let campaign in influencing votes, the effects of a movie about Jackie

factors. For example, let us say that he is interested in the extent to

for pursuing research on one topic rather than another. Different

Formulating the Research Problem

otheism begins with the hypothesis that Moses was actually an

But the sources of the hypotheses of a study have an important

tion on a theory of social stratification which holds that individuals

to the United States. National status was defined as "the relative