Unprogrammed Decision Making*

PEER SOELBERG
Massachusetts Institute of Technology

The research reported below has implications for management prac-

tice if one accepts the following three propositions:
i. Information processing and decision making are central func-
tions in modern organizations.
ii. To improve management decision making it is useful to know
how organizations at present make decisions.
This parallels the now familiar argument why engineers ought to
know the science underlying their engineering rules of thumb: The less
validated the engineering principles are, the more an engineer needs to
understand the science on which his practice rests.
iii. So long as people remain the chief instrument of corporate
policy a key feature of management decisions will be the choice
processes of individual human beings.
This paper is a report of how individuals make important, difficult,
and highly judgemental decisions. It has become customary to contrast
so-called "non-programmed" with more highly "programmed" types of
decisions.^ The latter are choices, or actions, that follow routinely from
the decision maker's (Dm's) application of explicit decision rules to
whatever stimulus or input data face him in his task environment.
The management of most company's daily operations abounds with
highly programmed decisions: Consider merely the highly routinized
rules that normally guide the everyday management of inventories, pro-
duction schedules, machine and manpower allocations, cost estimation,
mark-up pricing, etc. The more famous scientific description of a case
of highly programmed decision making is perhaps G.P.E. Clarkson's, in
which it was demonstrated that the portfolio selection decisions made by
a bank trust investment officer were so well programmed that his decisions
could be predicted by a computer, six months after his investment rules
had been elicited by an interviewer.^
This study, in contrast, focuses on highly unprogrammed decision
making. This is a subject that usually gets relegated to the mystical realm,
managerial "judgement". Critical decisions are produced every day for
which the decision maker (Dm) has available no identifiable rules or pre-
programmed decision procedure. This is not to say that Dm may not be
following a set of generalized guidelines when rendering his so-called
• This paper summarizes findings and conclusions from the author's research report
A Study of Decision Making, Carnegie and Massachusetts Institutes of Technology,
1966, xix + 465 pp. This paper copyrighted 1967 by Peer Soelherg. Reproduced with
permission.
judgement. But if you asked him directly. Dm would insist that the un-
programmed problem confronting him had to be solved in its own unique
context. Moreover, observing him solve it, you would indeed find (i)
that Dm applied few special-purpose decision rules when arriving at his
choice; (ii) that a number of the decision criteria he applied were
initially unoperational; (iii) that many of the choice alternatives he con-
sidered were unknown to start with; (iv) that information about the
alternatives' consequences and their relative worth was not immediately
available from the task environment; and (v) that Dm might not even
be able to specify the nature of an ideal solution to his problem a priori.
Yet it is precisely this type of non-programmed decision making that
forms the basis for allocating billions of dollars worth of resources in the
economy every year. And, ironically, until we better understand the na-
ture of such unprogrammed human decision processes, our sophisticated
computer technology will be of slight aid for making this type of decision
more effectively. In other words, the potential pay-off to management
of scientific understanding of the economic, psychological, sociological,
and political "laws" of non-programmed human judgement is truly enor-
mous.

AVAILABLE THEORIES OF UNPROGRAMMED

DECISION MAKING
Traditional economists have long tried to make do with the concepts
"utility" and "probability" for explaining unprogrammed choice among
uncertain alternatives.^ Utility functions, either cardinal or ordinal, are
linear preference orderings of all possible combinations of valued goods
and services, assumed to be adequate descriptions of Dm's value structure.
Likewise objective or personal probability measures are felt to capture
the essence of how Dm's think about whatever "factual" connections
they perceive between their available choice alternatives and the possible,
but uncertain, consequences of their choosing a given alternative.* Even
today mathematically oriented psychologists, management theorists, and
political scientists attempt to get along with little more than these two
deceptively elegant concepts, when trying to describe, or prescribe, non-
programmed human choice behavior.
The best known exception to this generalization is the work of Her-
bert A. Simon. The latter's notion of limited rationality, his "means-ends
satisficing" model of problem solving, and insistence on preserving one-
to-one process veridity in his theories of behavior have significantly re-
oriented and vitalized social science research on decision making.^
Simon characterizes unprogrammed decision making in terms of the
following three-phase process model :8
L Intelligence — finding occasions for making a decision;
2. Design — finding, inventing, developing, and analyzing alterna-
tive courses of action;
3. Choice — selecting a particular course of action from those avail-
able.
 In our research we used the following, slightly expanded phase struc-
ture as our framework for analyzing unprogrammed decision processes:
I. PARTICIPATION: the decision maker (Dm) is somehow in-
duced to work in a given task environment, in which he is then
motivated to attain one or more non-trivial objectives.
The Barnard-Simon inducements contribution theory is one
model that purports to explain under what conditions Dm
will decide to participate in a given task environment. The
research reported below may offer an alternate basis for ex-
planation.
II. RECOGNITION AND DEFINITION: Dm surveys his task
environment; i.e., attends to it selectively, and then discovers,
selects, or is somehow provided with — and then defines opera-
tionally — the particular problem or part of a problem he intends
to devote his resources to resolving.
Problem recognition may occur in the form of:
i. Dm's discovery of a barrier to his progress toward a goal;
ii. someone else in his organization handing Dm a problem,
with a request that he solve it;
iii. a serial performance indicator of some kind dropping
below Dm's target level (a basis^ for simple, automated
problem recognition); or some
iv. previously coded "problem" pattern appearing on Dm's
perceptual horizon.
Problems definition may take the form of:
i. Dm's description of differences, along one or more at-
tributes, between his goal and his present state;
ii. the description of a previously encountered problem that
"fits" the stimulus configuration presently confronting
Dm; or
iii. Dm's description of an ideal solution to the encountered
problem.
III. UNDERSTANDING: Dm investigates his task environment, try-
ing (1) to develop an appropriate set of event classifications, or
concepts, and (2) to formulate and test hypotheses about the ap-
parent cause-effect relationships in the environment — which in
turn might suggest design operators for, or help generate, viable
solution alternatives.
Environmental understanding is in part made up of:
i. Dm's deriving hypotheses about relationships among sets
of more primitive descriptors (i.e., concepts) in his image
model of the task environment;
ii. Dm's testing these hypotheses against his experiences, for-
mally recorded data, or experimental manipulation of
the task environment;
iii. Dm's deriving normative operators; i.e., ways to achieve
his subgoals, from his subjectively validated knowledge
of the task environment.
IV. DESIGN: Dm develops or searches for alternative courses of
action for solving his problem; he estimates or tries to ascertain
expected consequences of choosing each perceived alternative.
Search for alternatives and estimation of consequences are the
two components of decision design that traditionally get
focused upon by theorists of rational choice. As we demon-
strate below, however, Dm's search for, investigation within,
and estimation of alternatives include design components
that to date have been inadequately represented in available
decision models.
V. EVALUATION: Dm assigns some sort of value measure to the
estimated consequences of his perceived decision alternatives.
Utility theory has too long been embraced as the sole basis
for formally representing Dm's value structure. This study
has a number of alternate suggestions to make in that regard.
VI. REDUCTION: Dm reduces his set of viable decision alternatives
to a single one; i.e., he makes a choice.
According to traditional models this phase of decision making
is nearly trivial; Dm simply selects as his choice the best of his
previously ranked set of alternatives. This study demonstrates
that the choice reduction phase is not at all this simple, that
in fact it often constitutes a major hurdle in decision making.
VII. IMPLEMENTATION: Dm introduces and manages his decision
solution in the task environment.
This phase of decision making, usually the more critical one
for practical purposes, is usually left out of formal models of
decision making (a fact that operations research consultants,
if they have not formally recognized it in their decision mod-
els, have long known in practice).
VIII. FEEDBACK AND CONTROL: Dm receives and evaluates in-
formation from the task environment regarding the effects of his
implemented decision, and, if required, (i) changes his problem
definition, or (ii) modifies his goals, or (iii) takes appropriate
follow-up action.
This is the feature of decision making that introduced dy-
namics into the process. Most traditional models do not ade-
quately take into account the dynamic consequences of deci-
sions that are made at different points in time, by Dm's related
to each other by an organizational network of constraints —
even though their temporal dynamics obviously constitute a
major source of the cyclical pathologies that are often observed
in rational decision making systems.'^

RESEARCH STRATEGY
In order to explore empirically the detail structure of the above gen-
eralized decision process outline we would obviously have to investigate,
at great length, the information processes of a large number of Dms
solving many different types of decision problems, in different task en-
vironments. The criteria by which we chose a specific unprogrammed
decision situation to study initially were the following, namely that the
Dms we were to focus on would be:
i. well trained for making decisions, as well as able and motivated
to talk at length about their information processing while ac-
tually engaged in producing the decisions;
ii. highly involved with the problem confronting them, it being
personally very important for each Dm to reach the "right"
decision;
iii. quite unfamiliar with the type of decision problem they were
faced with, having encountered few such problems before, and
not expecting to do so again in the near future;
iv. engaged in making the decision over a longer period of time —
like several weeks — in order to minimize possible observer
measurements effects, yet allow a number of observations to be
made at different phases of the decision process;
V. easily and inexpensively accessible to the investigator in rea-
sonable number, in order to minimize idiosyncratic interpreta-
tion of the data, through cross-comparison of the thinking-aloud
protocols of fairly large samples of decision makers.
The above criteria for our choice of subjects were designed to help
us focus on as pure and "uncontaminated" a set of decision process obser-
vations as we thought could be found in industrial practice. However,
M.I.T. Sloan School of Management Master's and Doctor's candidates
making their post-graduate job decisions seemed to fit our bill reasonably
well. And in addition to satisfying our selection criteria, these subjects
would allow us readily and validly to test whatever rejectable hypotheses
were generated by our initial investigation, on succeeding years' samples
of graduating management recruits.
Our research objective originally was to design a longitudinal ques-
tionnaire that could efficiently and adequately chart the course of our
Dm's job decision processes. For that purpose we constructed an elaborate
questionnaire instrument, which in its complete form took three or four
hours for each Dm to complete, every week. This was clearly too long,
trying as it did to cover every possible theoretical contingency. For ex-
ample, one central part of the questionnaire derived from classical prob-
abilistic utility theory, according to which Dm was asked to identify,
weight, and then rate whatever goal dimensions he felt entered into his
decision. Perhaps not surprisingly, it turned out that the goal weights
which Dms provided during decision making could not be trusted: The
reported weights varied quite unreliably with (i) the specific alternatives
that Dm referred to when answering the goal weight questions, and (ii)
the temporal phasing of his decision process.
We therefore gave up the questionnaire as a poor job. It had become
increasingly obvious that unless our questionnaire was made up largely
of items that were closely compatible with the manner in which Dm
actually stored and manipulated his decision information "internally,"
during his own thinking about the problem, the answers he provided to
our questions would, for explanatory as well as predictive purposes, be
spurious at best; entirely misleading at worst.
We therefore resolved to rely, almost exclusively at first, on periodic,
open-ended, and highly detailed interviews with the decision makers in
process. These interviews provided our first insight into some rather
surprising aspects of unprogrammed decision making. Preliminary anal-
ysis of nearly 100 open ended interviews, each ranging from 1/2 hour to
2-1/^ hours in length, with 20 different decision makers over 3-5 month
choice periods, provided the basis for our first generalizable decision pro-
cessing model (GDP-I). The latter was first presented at Carnegie Insti-
tute of Technology in June, 1964.
Each interview protocol was thereafter reduced to comparable format
by the following 3-step method. First, each protocol was transcribed ver-
batum and its decision phase structure, according to GDP-I, was an-
notated in the margin. Thereafter the revelant protocol contents were
summarized in a synoptic coding language derived directly from the
variables and process hypotheses of GDP-I. Finally the current state of
each Dm's decision process and his active solution alternatives, at that
point in time, were entered on a multi-dimensional, Gantt type process
chart. The standardized data produced by the last two steps of the anal-
ysis thereafter served as the basis for quantifying each protocol, to enable
comparisons of decision processes to be made across Dms. The latter
curve-fitting analysis provided (post hoc) support for a number of GDP-I
hypotheses, the more interesting of which are summarized below:
a. Dm defines his career problem by deriving an ideal solution to
it, which in turn guides his planning of a set of operational cri-
teria for evaluating specific job alternatives.
b. Dm believes a priori that he will make his decision by weighting
all the relevant factors with respect to each alternative, and
then "add up numbers" in order to identify the best one. In
fact. Dm does not generally do this; and if he does, it is done
after he has made a selection among the alternatives.
c. Dm will search in parallel for alternatives, by activating one
or more "alternatives generators" — procedures which, once
activated, allow Dm to search passively, by deciding whedier or
not to follow up investigating particular ones of a stream of
alternatives presented by his generators.
d. Dm will usually be evaluating more than one alternative at a
time, each evaluation consisting of a series of investigation and
evaluation cycles.
e. Evaluation during the search phase takes the form of screening
each alternative along a number of noncompared goal dimen-
sions; no evidence of factor weighting is apparent at this stage.
f. Search will not necessarily halt as soon as Dm has identified
an acceptable alternative, one that is not rejected by his various
screening criteria; conversely, when Dm ends his search for new
(initial processing of newly generated) alternatives he will
8
 usually have more than a single acceptable alternative in his
"active roster."
g. When Dm terminates his search for new alternatives before his
search resources run out, he will already have identified a favor-
ite alternative in his roster of acceptable alternatives; this alter-
native (his choice candidate) can be identified by considering
Dm's primary goal attributes (usually one or two) alone.
h. At the point of search termination Dm will not generally have
compared his alternatives with one another, will not possess a
transitive rank ordering of alternatives, and will refuse to admit
that his implicit choice has been made.
i. Before Dm will recognize his choice explicitly he will engage in,
at times quite lengthy (two, three months) confirmation pro-
cessing of his roster of acceptable alternatives, during which
alternatives will get compared to each other, factor by factor.
j. During confirmation processing the roster of acceptable alterna-
tives, if greater than two alternatives, will quickly be reduced
to two alternatives — the choice candidate and a confirmation
candidate. If only one alternative, the choice candidate, is
viable at the time. Dm will try to obtain another acceptable
alternative (confirmation candidate) as soon as pxjssible "in
order to have something to compare it with."
k. The goal of confirmation processing is (a) to resolve the residual
uncertainties and problems connected with the choice candidate,
and (b) to arrive at a decision rule which shows unequivocally
that the choice candidate dominates the confirmation candi-
date — Pareto dominance being the ideal goal strived for.
1. During confirmation processing a great deal of perceptual and
interpretational distortion takes place, in favor of the choice
candidate, to the detriment of the confirmation candidate; goal
attribute "weights" are arrived at, or changed, to fit the per-
ceived data and the desired decision outcome.
m. The decision is made when a satisfactorily Pareto dominant
decision rule has been constructed, or when Dm runs up against
a time deadline during confirmation processing.

A FOLLOW-UP LONGITUDINAL QUESTIONNAIRE STUDY

Having thus gained a fair degree of insight into the information
processes of unprogrammed human decision making we were now obvious-
ly in a much better position, than we were previously, to design a predic-
tively valid questionnaire instrument for testing, on new and large sam-
ples of Dms, some of the key hypotheses suggested by the GDP-I model.
In contrast to the protocol curve fitting exercise reported above, our
follow-up investigation was a bona fide prediction study: All hypotheses,
with process-valid measures of the variables, were specified a priori.
Moreover, apart from our own personal belief in the GDP-I model, one
would have derived small prior likelihood estimates from any other
available decision model that the hypotheses which we were about to test
were in fact true. To most orthodox theorists our predictions should
indeed appear to be "long shots in the dark."
Among the hypotheses that we tested by process questionnaires were
the following:
I. Search for new alternatives ends a significant period of time
before Dm is willing to admit having made his decision.
II. In observation periods prior to ending his search for new alter-
natives Dm will, more often than not, already have available
one or more acceptable choice alternatives.
III. When Dm ends his search for new alternatives he will report
significant uncertainty regarding which alternative he is finally
to select as his choice.
IV. Should Dm not have obtained a firm job offer from more than
one acceptable alternative at time of search termination, by the
time he is ready to announce his decision he will have tried
hard, and will usually have obtained, at least one other accept-
able offer (according to GDP-I, in order to have something
with which to compare his choice candidate).
V. When Dm ends his search for new alternatives, his favorite
alternative can be identified by asking him a set of quite simple
questions. When Dm's subsequent confirmation processing of
alternatives ends, i.e., at time of choice announcement, his deci-
sion will be to select that alternative.
VI. Affective dissonance reduction, in the form of a "spreading
apart" of Dm's liking for his accepted versus rejected alterna-
tives, will not be generally observed after choice has been an-
nounced.
Those familiar with sequential search, aspiration level, choice models
may recall that according to the latter theory the first five hypotheses
would not appear reasonable. Similarly, according to the cognitive dis-
sonance theory, the sixth would be a disturbing proposition.

RESULTS OF THE LONGITUDINAL QUESTIONNAIRE STUDY

Below we can no more than summarize the findings pertaining to
the above six hypotheses — based on data from 256 questionnaire response
sets provided by 32 members of the 1965 graduating class of M.I.T. Sloan
School of Management Master's and Doctoral candidates. Each Dm in
the sample provided answers to eight bi-weekly questionnaires over the
period when he made his job decision. (For small numbers of Dms, a
different subset with respect to each hypothesis, the path of their decision
processes, as recorded by the questionnaires, provided inadequate data
with which to test a given hypothesis. Thus the totals reported below may
add to somewhat less than 32.)
Hypothesis I: 27 of 31 Dms (87%) terminated search for new alterna-
tives 10 days or more before the date on which they
reported having made their decision. 15 of 31 Dms
10
 terminated search 3 weeks or more before choice was
made.
Hypothesis II: Using a highly conservative measure of an alternative's
acceptability, 17 of 24 Dms (74%) reported having
available one or more acceptable alternative two weeks
or more before they terminated search for new alterna-
tives.
Hypothesis III: The average personal probability distribution of 28
Dm's reporting, at time of search termination, regard-
ing the likelihood that they would choose either of the
alternatives which we independently identified as being
their "choice candidate," their second most preferred
alternative, and "all other alternatives," was respec-
tively: [.29 .24 .47]. In other words, great uncertainity
was expressed by Dm at the time of search termination
regarding which alternative he was to choose.
Hypothesis IV: 13 of 16 Dms (81%) who did not have, or had not been
promised, an offer from more than one alternative at
time of search termination, did report having at least
one such other offer in hand before they rnade their
decision.
Hypothesis V: 25-1/2 of 29 Dms (87%) [14 since one Dm could rea-
sonably be classified either way] eventually selected as
their final decision that alternative which We, one to
twelve weeks, a median of three weeks, earlier — at the
time of Dm's search termination — had independently
identified as being his favorite alternative; i.e., his
choice candidate.
Hypothesis VI: No Dm reported a consistent dissonance reduction
"spreading apart" of his liking for accepted versus re-
jected alternatives over the periods of observation im-
mediately following decision commitment. However,
2 Dms exhibited what we might call latent dissonance
reduction; i.e., one which took effect two or more weeks
after Dm had committed himself to the decision. 9 of
26 Dms (35%) showed an initial "spreading apart"
effect of their relative liking for alternatives, a gap
which, however, was reduced again in subsequent peri-
ods of observation. 10 of 26 Dms (38%) exhibited no
changes whatever in their reported liking differentials
in the observation periods following choice. The re-
maining 5 Dms exhibited post choice dissonance ex-
pansion; i.e., a narrowing down of their liking dif-
ferential between alternatives after they had made their
decision.
In summary, the six decision process hypotheses described above
were thus supported rather convincingly by the data in our longitudinal
prediction study.
n
 CHIEF IMPLICAllONS FOR A BEHAVIORAL THEORY
OF DECISION MAKING
Below are some of the study's more central implications for decision
theory, which may not yet be entirely obvious from the above, severely
summarized report of our findings:
A. Scalar utility theory is a poor way of representing the structure
of human values. Decision attributes are usually multi-dimensional; i.e.,
are not compared or substituted for each other during choice. No stable
utility weighting function can be elicited from Dm prior to his selection
of a preferred alternative; nor do such weights appear to enter into Dm's
decision processing. His non-comparison of goal attributes during the
alternatives screening and selection phase also obviates Dm's need for,
and the reasonableness of our postulating the existence of, a multi-dimen-
sional utility indifference map.
B. Probability theory, either in objective frequency or personal-
Bayesian form, does not provide an adequate representation of Dm's per-
ceived uncertainty during unprogrammed decision making. The "prob-
ability" indices with which our highly trained Dms provided us, on direct
questions about them, were neither additive nor cardinally scaled. It
seems that Dm does not normally think of his choice alternatives in terms
of multiple consequences, each of which is then seen to depend condi-
tionally on a diflEerent reaction to his decision by the task environment.
Dm instead thinks of each choice alternative in terms of a set of non-
comparable goal attributes. Uncertainty in this context is more appropri-
ately represented in terms of equally likely ranges of a specific alternative's
rating along its uncertain goal attributes. In other words, decision un-
certainty rarely takes the form of a "pure" or probability-risk consequence
uncertainty; more commonly it constitutes non-distributive uncertainty
with respect to Dm's goal attribute evaluation of an alternative.
The mathematics of how Dm compares such multiple-attribute un-
certainty-ranged choice alternatives is quite simple, but, unfortunately,
would still take too much space to illustrate here. By the same token of
limited rationality, one might argue that it is the simplicity of Dm's in-
formation processing computations that effectively prevents him from
operating with the m. conditional probability distributions for each al-
ternative, that according to distributive probability theory. Dm should
be associating with each multi-consequence, multi-valued alternative.
C. Search for alternatives is a paralleled process; i.e., several poten-
tially acceptable alternatives are considered by Dm. at a time. This pro-
position contrasts with sequential one-at-a-time search models. In addi-
tion, Dm's evaluation of an alternative is a multi-stage affair, at each step
of which new information is collected and evaluated about some of the
attributes of the given alternative. In other words, search within alterna-
tives is an important a process for us to understand formally as the tradi-
tionally described search across alternatives.
During the search phase Dm does not view his evaluation of alterna-
tives as final. Alternatives that fall short on important goal attributes are
12
rejected immediately, but acceptable alternatives are merely put into
Dm's "active roster," with little or no systematic comparison performed
across the different acceptable alternatives, until Dm is ready to make his
final decision. In other words. Dm may well continue to search for new
alternatives, even though he has already discovered a perfectly satisfactory
one (one that was not rejected by any of his important goal attributes).
D. Making the final decision, what we have called decision con-
firmation, takes place after Dm has terminated search for new alternatives,
and is cognitively a highly involved, affectively a most painful, process
for Dm to engage in. This is the period during which Dm has to reject
alternatives that seem perfectly satisfactory to him, in some ways perhaps
better than the one he finally ends up choosing. It is at this point that
Dm is forced systematically to compare patently non-comparable alterna-
tives.
It is a major thesis of this study that Dms generally solve this problem
in the simplest manner conceivable, by not entering into this difficult
period of decision making until one of the alternatives can be identified
as an implicit "favorite." In other words, decision making during its
confirmation phase is an exercise in prejudice, of making sure that one's
implicit favorite will indeed be the "right" choice. This proposition
gives the key to a surprising degree of predictability in decision making,
demonstrated with the data of Hypothesis V above, in which we predict
87% of the career jobs taken, two to eight weeks before Dms would
admit they had reached a decision.
It would be too lengthy here to go into details on the nature of the
confirmation process (see: A study of decision making). The following
are two of its more outstanding characteristics: (a) The criteria that Dm
uses for identifying his favorite alternative are very few; not more than
one or two what we have called primary goal attributes account for most
of the observed variance, (b) Dm's comparison among alternatives quick-
ly reduces to a pro-con argument between two, and only two, alternatives
(see Hypothesis IV), the objective being for Dm to bring (i) his percep-
tion of the facts, (ii) and his evaluation of goal attributes, into line with
his predisposition that the preferred choice candidate dominates his
second-best alternative (which we call the confirmation candidate) on
all of Dm's important goal attributes, secondary and primary. Dm finally
makes his decision when he has constructed himself a satisfactory decision
rule — a goal weighting function, if you please — which enables him to
explain the Pareto dominance of his choice candidate. (Unless, of course.
Dm is forced by some deadline to make his decision before that time.
If so, however. Dm will still choose his choice candidate, but with much
more felt uncertainty about the "rightness" of his decision.)
E. Dissonance reduction, in the sense that such has been described
by Leon Festinger,^ must be viewed as a conditional phenomenea: In a
loose sense, confirmation processing might be viewed as being part of Dm's
"dissonance reduction" process. But according to Festinger, the onset of
dissonance reduction awaits Dm's commitment to his choice, which in
our data is synonomous with the point of Dm's choice announcement. And
13
in our study dissonance reduction after that point in time was observed in
only 35% of the cases, in all of which the effect dissipated during sub-
sequent periods of observations.
We propose as a testable explanation of our data: Post-choice dis-
sonance reduction will be observed only when Dm, at the time of choice
commitment, is not satisfied with his confirmation decision rule, i.e., with
the intellectual rationale for why he chose the way he did. Thus dis-
sonance reduction constitutes an effective compensation, on the part of
Dm, for his lack of a socially acceptable intellectual justification for his
behavior.
This hypothesis also explains the observed second order dissonance
reduction effect: With time we expect all men to be able to invent better
and better rationales for why they behaved as they did. Correspondingly,
we should observe that any initial affective (dissonance reduction) com-
pensation, with which Dm at first may be protecting his decision, will be
dissolved over time as his intellectual argument gets better.

SOME OBVIOUS CONSEQUENCES OF THE FINDINGS

FOR MANAGEMENT PRACTICE
Let us conclude by briefly considering some lessons of these findings
for management. The listener can surely think of other implications,
which I hope will be brought forward during our discussion. Neverthe-
less, here are a few obvious observations:
i. Our generalizable decision process model (GDP-I) allows the
manager (Mgr) to recognize when others have readied an implicit deci-
sion; i.e., when they are merely confirming their favorite alternative. Such
knowledge should enable Mgr not to waste time, resources, or face by re-
maining a party to a choice process that for most purposes has already
been dosed. This lesson should be particularly useful in situations where
Mgr; or his company, has been cast in the role of "confirmation candidate"
by the decision maker.
ii. The existence of a confirmation process that goes into effect
prior to public choice commitment emphasizes the desirability of getting
one's alternatives into the decision process early. On bids for government
research and development contracts, for example, Edward Roberts has
uncovered evidence that Mgr needs to get in there well before the official
invitation to bid on a contract has left the government agency — that at
this time one can predict with disturbing success which firm will get the
contract, simply by looking at the order of names on the list of those in-
vited to bid.8
iii. The confirmation process also suggests a way of manipulating
decision deadlines in Mgfs favor: If he has evidence that his alternative
is the favorite, Mgr can safely clinch the deal by imposing a stringent
deadline, trust to dissonance reduction to carry the day, and save himself
time, needless anxiety, and the risk of that rare alternative arriving on
Dm's horizon in time to upset the apple cart.
iv. The existence of the confirmation process also explains the often
14
observed as symmetry of administrative decision making: Once made, deci-
sions are usually very hard to unmake, or to get re-made. A most obvious
explanation is that Mgr balks at having to go through all the pain of
changing his tailor-made decision rule to fit a new pair of alternatives.
(That might smell too much of rationalizing, and go against the grain of
men who like to think of themselves as principled, orderly decision mak-
ers.) Besides, the decision rule offers ready arguments, in m dimensions,
why few alternatives can be expected to be as good as the chosen one.
And these arguments get themselves strengthened and elaborated as time
passes — partly through the self-fulfilling prophency which will bias all
future interactions between Mgr and his rejected versus accepted alterna-
tives.
V. Our description of the nature of the confirmation process also
offers a complementary explanation of the observed difficulty of changing
people's cognitive attitudes: Mgr can argue till he is blue in face with
Dm's decision rule; as soon as Mgr is successful in winning a battle on one
secondary point in Dm's rule, the latter will simply mend his breach,
either by pooh-poohing that particular goal attribute, or by countering
with a compensating argument along some other goal dimension.
Only if Mgr manages to zero in on Dm's truly primary goal attributes
(often carried around by Dm quite inaccessibly, as an uncommunicated
existentialist "feel" for the problem situation), can Mgr hope to change
Dm's dedsion behavior. But note, in the latter case Mgr will face the
difficult task of demonstrating convincingly to Dm that his old favorite
is dearly dominated by Mgr's own favorite alternative. (This proposition
might help explain why public debates should be so ineffective as a means
of changing anyone's voting behavior or political allegiance.)

IMPROVEMENT OF MANAGEMENT DECISION

Below are three lessons that appear to follow from the study, re-
garding how to improve management decision making:
vi. Mgr should work on integrating his formal models of rational
decision making with his intuitive judgemental, common sense manner
of solving choice problems — and seek to adapt the former to fit the
latter, rather than submit to a bastardization of his intuition in the name
of some modern mathematical technique. Mechanical aids to manage-
ment decision, like computer based management information systems, will
be (and should be) resisted; i.e., not used, to the extent that its structure
is incompatible with (a) the manner Mgr codes relevant decision infor-
mation for his own use, or (b) the manner in which Mgr intuitively feels
that information should be reduced for arriving at a decision.
To be somewhat more specific, a formal goal attribute weighting
scheme, an imposed set of operating decision rules, or an explicit frame-
work for estimating and operating with personal probabilities will be
circumvented by reasonable managers, we hypothesize, to the extent that
the area in which the technique is to be applied has not been carefully
chosen to match the structure or Dm's intuitive (culturally learned) pro-
15
 cess of working through multi-valued and uncertain decision alternatives.
This is not to say, however, that Mgr should not attempt to educate his
own decision making judgment, though the more effective way to ac-
complish this is still hotly debated in schools of business. We would re-
commend that a major part of Mgr's effort in that regard be directed to-
ward making the processes of his managerial judgment more explicit to
himself, in a variety of different job situations.
vii. One way Mgr might start training himself to make better deci-
sions would thus be to become more aware of his personal predispositions
and prejudices (i.e., of his primary goals) when operating in different
task contexts. Rapid feedback regarding his apparent decision behavior,
from people whom Mgr already trusts and respects, should be of much
help in that regard. Trying to avoid forming opinions early about com-
plex sets of alternatives seems (from anectdotal evidence) to create an un-
comfortable state of tension in most people. Perhaps Mgr might try
devising private "holding" heuristics to allow "sufficient" unbiased in-
formation about his available alternatives to be collected, and to prevent
hini from modifying his decision criteria until he has reached an explicit
decision to start doing so. A counterargument, however, is that decision
making and action taking under time pressure is so difficult to get accom-
plished under any circumstances, that Mgr needs to use all the short-cuts
and tension-reducing rules of thumb he can devise, even if in some cases
these lead to biased results.
viii. Our theory would lead us to expect that different Mgrs will
exhibit different degrees of the tendency to commit themselves to alterna-
tives early in the decision process. Perhaps such a predisposition could be
effectively counteracted by pointing out to Mgr that this is the way he
tends to operate; but perhaps the characteristic is sufficiently diffidilt or
expensive to change that we should consider developing a standardized
test situation for helping screen out of critical managerial positions those
persons who too early, on too meager evidence, tend to jump to conclu-
sions about the solution of complex problems.
NOTES AND REFERENCES
1. J. G. March and H. A. Simon, Organizations, New York: Wiley, 1958, pp. 169-182.
2. G. P. E. Clarkson, Portfolio Selection: A Simulation of Trust Investment, Engle-
wood ClifEs, N.J.: Prentice-Hall, 1962.
3. K. J. Arrow. "Utility, Attitudes, Choices: A Review Note," Econometrica, 1958,
pp. 1-23.
4. See J. F. Rothenherg, The Measurement of Social Welfare, Enslewood ClifJs N T.-
Prentice-Hall, 1961, pp. 200-278. ' a , j
5. H. A. Simon, Administrative Behavior, New York: MacMillan, 1947; Models of Man
New York: Wiley, 1957; "Theories of Dedsion Making in Economics and Behavioral
Saence," Am. Econ. Review, 1958, 49, pp. 255-283; with A. Newell and J. C. Shaw,
"Elements of a Theory of Human Problem Solving," Psychol. Review, 1958, 65, pp.
151-166; The Shape of Automation, New York: Harper, 1966.
6. H. A. Simon, New Science of Management Decision, New York: Harper, 1960.
7. J. A. Forrester, Industrial Dynamics, New York: Wiley, 1961.
8. L. Festinger, Conflict, Choice, and Dissonance Reduction, Stanford University Press,
9. E. B. Roberts, "Questioning the Cost/Effectiveness of the R & D Procurement Pro-
cess," in M. Yovits, et al, (eds.), Research Program Effectiveness, New York: Gordon
and Breach, 1966.
16