ASTM Data and Control

Manual on Presentation of Data and Control Chart Analysis
7th Edition
Prepared by
COMMITTEE E-11 ON QUALITY AND STATISTICS

Stock #; MNL7A Revision of Special Technical Publication (STP) 15D
^l4|| ASTM International 100 Barr Harbor Drive PO Box C700 INTERNATIONAL West Conshohocken, PA 19428-2959
Ittirary at Congress Cataloging-in-Publication Data
Manual en presentation of data and control chart analysis / prepared by the Committee E-11 on statistical control. (ASTM manual series ; MNL 7) Includes bibliographical references. ISBN 0-8031-1289-0 1. MaterialsTestingHandbooks, manuals, etc. 2. Quality controlStatistical methodsHandbooks, manuals, etc. I. ASTM Committeie E-11 on Statistical Methods. II. Series. TA410.M355 1989 620.1'1'0287dc20 89-18047 CIP
Copyright 2002 by ASTM International, West Conshohocken, PA. Prior editions copyrighted 1995 and eariier, by ASTM International. All rights reserved. This material may not be reproduced or copied, in whole or in part, in any printed, mechanical, electronic, film, or other distribution and storage media, without consent of the publisher.
ASTM International Photocopy Rights Authorization to photocopy items for internal or personal use, or the internal or personal use of specific clients, is granted by ASTM International for users registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the base fee of $2.50 per copy, plus $0.50 per page is paid directly to CCC, 222 Rosewood Dr., Danvers, MA 01923; Phone: (508) 750-8400; Fax: (508) 740-4744; online: http://www.copyright.coni/. For those organizations that have been granted a photocopy license by CCC, a separate system of payment has been arranged. The fee code for users of the Transactional Reporting Service is 0-8031-1289-0 95 $2.50 + .50.
Printed in Bridgeport, NJ February 2002
Foreword
THIS A S T M Manual on Presentation of Data and Control Chart Analysis is the sixth revision of the original ASTM Manual on Presentation of Data first published in 1933. This sixth revision was prepared by the ASTM El 1.10 Subcommittee on Sampling and Data Analysis, which serves the ASTM Committee E-11 on Quality and Statistics.
Contents
Preface P r e s e n t a t i o n of D a t a Summary Recommendations for Presentation of Data Glossary of Symbols Used in Part 1 Introduction 1. Purpose 2. Type of Data Considered 3. Homogeneous Data 4. Typical Examples of Physical Data U n g r o u p e d Whole N u m b e r Distribution 5. Ungrouped Distribution 6. Empirical Percentiles and Order Statistics Grouped F r e q u e n c y Distributions 7. Introduction 8. Definitions 9. Choice of Bin Boundaries 10. Number of Bins 11. Rules for Constructing Bins 12. Tabular Presentation 13. Graphical Presentation 14. Cumulative Frequency Distribution 15. "Stem and L e a f Diagram 16. "Ordered Stem and L e a f Diagram and Box Plot F u n c t i o n s of a F r e q u e n c y Distribution 17. Introduction 18. Relative Frequency 19. Average (Arithmetic Mean) 20. Other Measures of Central Tendency 21. Standard Deviation 22. Other Measures of Dispersion 23. Skewnessgi 23a. Kurtosisg2 24. Computational Tutorial A m o u n t of Information Contained in p, X,s, gi, a n d gz 25. Summarizing the Information 26. Several Values of Relative Frequency, p 27. Single Percentile of Relative Frequency, p 28. Average XOnly 1 5 5 5 5 7 7 7 8 9 10 10 10 13 13 14 14 14 14 17 18 19 20 21 22 22 23 23 23 24 24 25 25 26 26 26 27 27 28
29. Average X and Standard Deviation s 30. Average X, Standard Deviation s, Skewness gi, and Kurtosis g2 31. Use of Coefficient of Variation Instead of the Standard Deviation 32. General Comment on Observed Frequency Distributions of a Series of ASTM Observations 33. SummaryÂmount of Information Contained in Simple Functions of the Data Essential Information 34. Introduction 35. What Functions of the Data Contain the Essential Information 36. Presenting X Only Versus Presenting X and s 37. Observed Relationships 38. Summary: Essential Information P r e s e n t a t i o n of R e l e v a n t Information 39. Introduction 40. Relevant Information 41. Evidence of Control Recommendations 42. Recommendations for Presentation of Data References for Part 1 P r e s e n t i n g P l u s or Minus Limits of U n c e r t a i n t y of a n Observed Average Glossary of Symbols Used in Part 2 1. Purpose 2. The Problem 3. Theoretical Background 4. Computation of Limits 5. Experimental Illustration 6. Presentation of Data 7. One-Sided Limits 8. General Comments on the Use of Confidence Limits 9. Number of Places to be Retained in Computation and Presentation Supplements A. Presenting Plus or Minus Limits of Uncertainty for <7 Normal Distribution B. Presenting Plus or Minus Limits of Uncertainty fory References for Part 2
28 30 33 34 35 35 35 35 36 37 39 39 39 39 40 40 40 41
42 42 42 42 43 43 47 47 48 49 49 51 51 53 55
Control Chart Method of Analysis and Presentation of Data Glossary of Terms and Symbols Used in Part 3 General Principles 1. Purpose 2. Terminology and Technical Background 3. Two Uses 4. Breaking up Data into Rational Subgroups 5. General Technique in Using Control Chart Method 6. Control Limits and Criteria of Control ControlNo Standard Given 7. Introduction 8. Control Charts for Averages, X, and for Standard Deviations, sLarge Samples 9. Control Charts for Averages, X, and for Standard Deviations, sSmall Samples 10. Control Charts for Averages, X, and for Ranges, RSmall Samples 11. Summary, Control Charts for X, s, and RNo Standard Given 12. Control Charts for Attributes Data 13. Control Chart for Fraction Nonconforming, p 14. Control Chart for Number of Nonconforming Units, np 15. Control Chart for Nonconformities per Unit, u 16. Control Chart for Number of Nonconformities, c 17. Summary, Control Charts for p, np, u, and cNo Standard Given Control with Respect to a Given Standard 18. Introduction 19. Control Charts for Averages X, and for Standard Deviation, s 20. Control Chart for Ranges, R 21. Summary, Control Charts for X, s, and R Standard Given 22. Control Charts for Attributes Data 23. Control Chart for Fraction Nonconforming, p 24. Control Chart for Number of Nonconforming Units, np 25. Control Chart for Nonconformities per Unit, u 26. Control Chart for Number of Nonconformities, c 27. Summary, Control Charts for p, np, u, and c Standard Given Control Charts for Individuals 28. Introduction 29. Control Chart for Individuals, XÛsing Rational Subgroups
56 56 58 58 59 60 60 60 61 64 64 64 65 66 66 66 69 70 71 73 74 74 74 75 76 76 76 76 78 78 79 80 80 80 81
30. Control Chart for Individuals, XUsing Ranges
Moving 81
Examples 82 31. Illustrative ExamplesControl, No Standard Given 82 Example 1: Control Charts for X and s, Large Samples of Equal Size (Section 8A) 84 Example 2: Control Charts for X and s. Large Samples of Unequal Size (Section 8B) 84 Example 3: Control Charts for X and s. Small Samples of Equal Size (Section 9A) 85 Example 4: Control Charts for X and s. Small Samples of Unequal Size (Section 9B) 86 Example 5: Control Charts for X and R, Small Samples of Equal Size (Section lOA) 86 Example 6: Control Charts for X and R, Small Samples of Unequal Size (Section lOB) 87 Example 7: Control Charts for p, Samples of Equal Size (Section ISA), and np, Samples of Equal Size (Section 14) 88 Example 8: Control Chart for p, Samples of Unequal Size (Section 13B) 90 Example 9: Control Charts for u, Samples of Equal Size (Section 15A), and c. Samples of Equal Size (Section 16A) 90 Example 10: Control Chart for u, Samples of Unequal Size (Section 15B) 92 Example 11: Control Charts for c. Samples of Equal Size (Section 16A) 93 32. Illustrative ExamplesControl With Respect to a Given Standard 95 Example 12: Control Charts for X and s, Large Samples of Equal Size (Section 19) 95 Example 13: Control Charts for X and s, Large Samples of Unequal Size (Section 19) 96 Example 14: Control Chart for X and s. Small Samples of Equal Size (Section 19) 96 Example 15: Control Chart for X and s, Small Samples of Unequal Size (Section 19) 97 Example 16: Control Charts for X and R, Small Samples of Equal Size (Section 19 and 20) 98 Example 17: Control Charts forp. Samples of Equal Size (Section 23), and np, Samples of Equal Size (Section 24) 99 Example 18: Control Chart forp (Fraction Nonconforming), Samples of Unequal Size (Section 23) 100 Example 19: Control Chart forp (Fraction Rejected), Total and Components, Samples of Unequal Size (Section 23) 101 Example 20: Control Chart for u, Samples of
Unequal Size (Section 25) Example 21: Control Charts for c, Samples of Equal Size (Section 26) 33. Illustrative ExamplesControl Chart for Individuals Example 22: Control Chart for Individuals, X Using Rational Subgroups, Samples of Equal
103 105 106
Size, No Standard GivenBased on X and R (Section 29) 106 Example 23: Control Chart for Individuals, X Using Rational Subgroups, Standard Given, Based on \io and OQ (Section 29) 107 Example 24: Control Charts for Individuals, X, and Moving Range, MR, of Two Observations, No Standard GivenBased on X and MR , the Mean Moving Range (Section 30A) 109 Example 25: Control Charts for Individuals, X, and Moving Range, MR, of Two Observations, Standard GivenBased on j o and OQ (Section30B)110 O Supplements A. Mathematical Relations and Tables of Factors for Computing Control Chart Lines B. Explanatory Notes References for Part 3 S e l e c t e d P a p e r s o n Control Chart T e c h n i q u e s Appendix List of Some Related Publications on Quality Control Index 111 111 117 121 122
125 127
MNL7A-EB/Feb. 2002
1 INTRODUCTORY INFORMATION
PREFACE
THIS
Manual on the Presentation of Data and Control Chart Analysis (MNL 7), was prepared by ASTM's Committee E-11 on Quality and Statistics to make available to the ASTM INTERNATIONAL membership, and others, information regarding statistical and quality control methods, and to make recommendations for their application in the engineering work of the Society. The quality control methods considered herein are those methods t h a t have been developed on a statistical basis to control the quality of product through the proper relation of specification, production, and inspection as parts of a continuing process. The purposes for which the Society was foundedthe promotion of knowledge of the materials of engineering, and the standardization of specifications and the methods of testinginvolve at every t u r n the collection, analysis, interpretation, and presentation of quantitative data. Such data form an important part of the source material used in arriving at new knowledge and in selecting standards of quality and methods of testing that are adequate, satisfactory, and economic, from the standpoints of the producer and the consumer.
predicting what variations in quality may be expected in subsequently produced material; to discover the distributions of the errors of measurement for particular test methods, which serve as a basis for comparing the relative merits of two or more methods of testing, for specifying the precision and accuracy of standard tests, for setting up economical testing and sampling procedures; (b) to discover the relationship between two or more properties of a material, such as density and tensile strength; and (c) to discover physical causes of the behavior of materials under particular service conditions; to discover the causes of nonconformance with specified standards in order to make possible the elimination of assignable causes and the attainment of economic control of quality. Problems falling in these categories can be treated advantageously by the application of statistical methods and quality control methods. This Manual limits itself to several of the items mentioned under (a). PART 1 discusses frequency distributions, simple statistical measures, and the presentation, in concise form, of the essential information contained in a single set of n observations. PART 2 discusses the problem of expressing + limits of uncertainty for various statistical measures, together with some working rules for rounding-off observed results to an appropriate number of significant figures. PART 3 discusses the control chart method for the analysis of observational data obtained from a series of samples, and for detecting lack of statistical control of quality. The present Manual is the sixth revision of earlier work on the subject. The original ASTM Manual on Presentation of Data, STP 15, issued in 1933 was prepared by a special committee of former
Broadly, the three general objects of gathering engineering data are to discover: (1) physical constants and frequency distributions, (2) the relationshipsboth functional and statisticalbetween two or more variables, and (3) causes of observed phenomena. Under these general headings, the following more specific objectives in the work of ASTM International may be cited: (a) to discover the distributions of quality characteristics of materials which serve as a basis for setting economic standards of quality, for comparing the relative merits of two or more materials for a particular use, for controlling quality at desired levels, for
C o p y r i g h f 2002 by A S T M International
www.astm.org
DATA AND CONTROL CHART ANALYSIS

Subcommittee IX on Interpretation and Presentation of Data of ASTM Committee E-1 on Methods of Testing. In 1935, Supplement A on Presenting Limits of Uncertainty of an Observed Average and Supplement B on "Control Chart" Method of Analysis and Presentation of Data were issued. These were combined with the original manual and the whole, with minor modifications, was issued as a single volume in 1937. The personnel of the Manual Committee that undertook this early work were: H. F. Dodge, W. C. ChanceUor, J. T. McKenzie, R. F. Passano, H. G. Romig, R. T. Webster, and A. E. R. Westman. They were aided in their work by the ready cooperation of the Joint Committee on the Development of Apphcations of Statistics in Engineering and Manufacturing (sponsored by ASTM International and the American Society of Mechanical Engineers (ASME)) and especially of the chairman of the Joint Committee, W. A. Shewhart. The nomenclature and symbolism used in this early work were adopted in 1941 and 1942 in the American War Standards on Quahty Control (Zl.l, Z1.2, and Z1.3) of the American Standards Association, and its Supplement B was reproduced as an appenduc with one of these standards. In 1946, ASTM Technical Committee E11 on Quality Control of Materials was established under the chairmanship of H. F. Dodge, and the manual became its responsibility. A major revision was issued in 1951 as ASTM Manual on Quality Control of Materials, STP 15C. The Task Group that undertook the revision of PART 1 consisted of R. F. Passano, Chairman, H. F. Dodge, A. C. Holman, and J. T. McKenzie. The same task group also revised PART 2 (the old Supplement A) and the task group for revision of PART 3 (the old Supplement B) consisted of A. E. R. Westman, Chairman, H. F. Dodge, A. I. Peterson, H. G. Romig, and L. E. Simon. In this 1951 revision, the term "confidence limits" was introduced and constants for computing 0.95 confidence hmits were added to the constants for 0.90 and 0.99 confidence hmits presented in prior printings. Separate treatment was given to control charts for "number of defectives," "number of defects," and "number of defects per unit" and material on control charts for individuals was added. In subsequent editions, the term "defective" has been replaced by "nonconforming unit" and "defect" by "nonconformity" to agree with definitions adopted by the American Society for Quality Control in 1978 (See the American National Standard, ANSI/ASQC Al-1987, Definitions, Symbols, Formulas and Tables for Control Charts.) There were more printings of ASTM STP 15C, one in 1956 and a second in 1960. The first added the ASTM Recommended Practice for Choice of Sample Size to Estimate the Average Quality of a Lot or Process (E 122) as an Appendix. This recommended practice had been prepared by a task group of ASTM Committee E-11 consisting of A. G. Scroggie, Chairman, C. A. Bicking, W. E. Deming, H. F. Dodge, and S. B. Littauer. This Appendix was removed from that edition because it is revised more often than the main text of this Manual. The current version of E 122, as well as of other relevant ASTM International pubhcations, may be procured from ASTM International. (See the hst of references at the back of this Manual.) In the 1960 printing, a modifications were made committee consisting of Chairman, Simon Collier, Hader, and E. G. Olds. number of minor by an ad hoc Harold Dodge, R. H. Ede, R. J.
The principal change in ASTM STP 15C introduced in ASTM STP 15D was the redefinition of the sample standard deviation to be s = J^^ '~ /-]) This
change required numerous changes throughout the Manual in mathematical equations and formulas, tables, and numerical illustrations. It also led to a sharpening of distinctions between sample values, universe values, and standard
INTRODUCTORY INFORMATION
values that necessary. were not formerly deemed the sample root-mean-square deviation. Replacing the universe values p', u' and c' by Greek letters was thought worse than leaving them as they are. Section 33, PART 1, on distributional information conveyed by Chebyshev's inequality, has been revised.
Summary of changes in definitions and notations. MNL7 H, a, p , u , c STP 15D
New material added in ASTM STP 15D included the following items. The sample measure of kurtosis, g2, was introduced. This addition led to a revision of Table 8 and Section 34 of PART 1. In PART 2, a brief discussion of the determination of confidence limits for a universe standard deviation and a universe proportion was included. The Task Group responsible for this fourth revision of the Manual consisted of A. J. Duncan, Chairman R. A. Freund, F. E. Grubbs, and D. C. McCune. In the twenty-two years between the appearance oi ASTM STP 15D and Manual on Presentation of Data and Control Chart Analysis, &^ Edition there were two reprintings without significant changes. In that period a number of misprints and minor inconsistencies were found in ASTM STP 15D. Among these were a few erroneous calculated values of control chart factors appearing in tables of PART 3. While all of these errors were small, the mere fact that they existed suggested a need to recalculate all tabled control chart factors. This task was carried out by A. T. A. Holden, a student at the Center for Quality and Applied Statistics at the Rochester Institute of Technology, under the general guidance of Professor E. G. Schilling of Committee E 11. The tabled values of control chart factors have been corrected where found in error. In addition, some ambiguities and inconsistencies between the text and the examples on attribute control charts have received attention. A few changes were made to bring the Manual into better agreement with contemporary statistical notation and usage. The symbol |i (Greek "mu") has replaced X (and X') for the universe average of measurements (and of sample averages of those measurements.) At the same time, the symbol o has replaced a' as the universe value of standard deviation. This entailed replacing a by s^j-ms) to denote
X',
(f, p', u', c'
( = universe values) |io, Oo, po, Uo, C o
( = universe values)
Xg,Oo', po', Uo', Co'
( = standard values)
( = standard values)
In the twelve-year period since this Manual was revised again, three developments were made t h a t had an increasing impact on the presentation of data and control chart analysis. The first was the introduction of a variety of new tools of data analysis and presentation. The effect to date of these developments is not fully reflected in PART 1 of this edition of the Manual, but an example of the "stem and l e a f diagram is now presented in Section 15. Manual on Presentation of Data and Control Chart Analysis, &'^ Edition from the first has embraced the idea that the control chart is an all-important tool for data analysis and presentation. To integrate properly the discussion of this established tool with the newer ones presents a challenge beyond the scope of this revision. The second development of recent years strongly affecting the presentation of data and control chart analysis is the greatly increased capacity, speed, and availability of personal computers and sophisticated hand calculators. The computer revolution h a s not only enhanced capabilities for data analysis and presentation, but has enabled
DATA AND CONTROL CHART ANALYSTS

techniques of high speed real-time datataking, analysis, and process control, which years ago would have been unfeasible, if not unthinkable. This has made it desirable to include some discussion of practical approximations for control chart factors for rapid if not real-time application. Supplement A has been considerably revised as a result. (The issue of approximations was raised by Professor A. L. Sweet of Purdue University.) The approximations presented in this Manual presume the computational ability to take squares and square roots of rational numbers without using tables. Accordingly, the Table of Squares and Square Roots that appeared as an Appendix to ASTM STP 15D was removed from the previous revision. Further discussion of approximations appears in Notes 8 and 9 of Supplement B, PART 3. Some of the approximations presented in PART 3 appear to be new and assume mathematical forms suggested in part by unpublished work of Dr. D. L. Jagerman of AT&T Bell Laboratories on the ratio of gamma functions with near arguments. The third development has been the refinement of alternative forms of the control chart, especially the exponentially weighted moving average chart and the cumulative sum ("cusum") chart. Unfortunately, time was lacking to include discussion of these developments in the fifth revision, although references are given. The assistance of S. J Amster of AT&T Bell Laboratories in providing recent references to these developments is gratefully acknowledged. Manual on Presentation of Data and Control Chart Analysis, &^ Edition by Committee E-11 was initiated by M. G. Natrella with the help of comments from A. Bloomberg, J. T. Bygott, B. A. Drew, R. A. Freund, E. H. Jebe, B. H. Levine, D. C. McCune, R. C. Paule, R. F. Potthoff, E. G. Schilling and R. R. Stone. The revision was completed by R. B. Murphy and R. R. Stone with further comments from A. J. Duncan, R. A. Freund, J. H. Hooper, E. H. Jebe and T. D. Murphy. Manual on Presentation of Data and Control Chart Analysis, 7" Edition has been directed at bringing the discussions around the various methods covered in PART 1 up to date. Especially, in the areas of whole number frequency distributions, empirical percentiles, and order statistics. As an example, an extension of the stem-and-leaf diagram has been added which is termed an "ordered stem-and-leaf," which makes it easier to locate the quartiles of the distribution. These quartiles, along with the maximum and minimum values, are then used in the construction of a box plot. In PART 3, additional material has been included to discuss the idea of risk, namely, the alpha (a) and beta (P) risks involved in the decision-making process based on data; and tests for assessing evidence of nonrandom behavior in process control charts. Also, use of the s(rms) statistic has been minimized in this revision in favor of the sample standard deviation s to reduce confusion as to their use. Furthermore, the graphics and tables throughout the text have been repositioned so that they appear more closely to their discussion in the text. Manual on Presentation of Data and Control Chart Analysis, Z"* Edition by Committee E-11 was initiated and led by Dean V. Neubauer, Chairman of the E l l . 1 0 Subcommittee on Sampling and Data Analysis that oversees this document. Additional comments from Steve Luko, Charles Proctor, Paul Selden, Greg Gould, Frank Sinibaldi, Ray Mignogna, Neil UUman, Thomas D. Murphy, and R. B. Murphy were instrumental in the vast majority of the revisions made in this sixth revision. Thanks must also be given to Kathy Dernoga and Monica Siperko of the ASTM International New Publications department for their efforts in the publication of this edition.
MNL7A-EB/Feb. 2002
Presentation of Dote
PART 1 IS CONCERNED solely with presenting information about a given sample of data. It contains no discussion of inferences that might be made about the population from which the sample came. To see how the data may depart from a Normal distribution, prepare the grouped frequency distribution and its histogram. Also, calculate skewness, gi, and kurtosis,
g2.
SUMMARY Bearing in mind that no rules can be laid down to which no exceptions can be found the committee believes that if the recommendations presented are followed, the presentations will contain the essential information for a majority of the uses made of ASTM data.
4. If the data seem not to be normally distributed, then one should consider presenting the median and percentiles (discussed in Section 6), or consider a transformation to make the distribution more normally distributed. The advice of a statistician should be sought to help determine which, if any, transformation is appropriate to suit the user's needs. 5. Present as much evidence as possible t h a t the data were obtained under controlled conditions. Present relevant information on precisely (a) the field of application within which the measurements are believed valid and (b) the conditions under which they were made.
6. RECOMMENDATIONS FOR PRESENTATION OF DATA Given a sample of n observations of a single variable obtained under the same essential conditions: 1. Present as a minimum, the average, the standard deviation, and the number of observations. Always state the number of observations. 2. Also, present the values of the maximum and minimum observations. Any collection of observations may contain mistakes. If errors occur in the collection of the data, then correct the data values, but do not discard or change any other observations. 3. The average and standard deviation are sufficient to describe the data, particularly so when they follow a Normal distribution.
2002 by ASTM International www.astni.org
GLOSSARY OF SYMBOLS U S E D PARTI
IN
gi
g2
Observed frequency (number of observations) in a single bin of a frequency distribution Sample coefficient of skewness, a measure of skewness, or lopsidedness of a distribution Sample coefficient of kurtosis Number of observed values (observations) Sample relative frequency or proportion, the ratio of the number of occurrences of a given type to the total possible number of occurrences, the ratio of the number of observations in any stated interval to
Copyright
DATA AND CONTROL CHART ANALYSTS

the total number of observations; sample fraction nonconforming for measured values the ratio of the number of observations lying outside specified limits (or beyond a specified limit) to the total number of observations Sample range, the difference between the largest observed value and the smallest observed value. Sample standard deviation Sample variance Sample coefficient of variation, a measure of relative dispersion based on the standard deviation (see Sect. 31) Observed values of a measurable characteristic; specific observed values are designated Xi, X2, X3, etc. in order of measurement, and X(i), X(2), X(3), etc. in order of their size, where X(i) is the smallest or minimum observation and X(n) is the largest or maximum observation in a sample of observations; also used to designate a measurable characteristic Sample average or sample mean, the sum of the n observed values in a sample divided by n defined as the expected value (see Note) of X; t h u s E(X) = [i, spelled "mu" and pronounced "mew." Population relative frequency Population standard deviation, spelled and pronounced "sigma." Population variance defined as the expected value (see Note) of the square of a deviation from the universe mean; thusE[(X-n)2]=a2 Population coefficient of variation defined as the population standard deviation divided by the population mean, also called the relative standard deviation, or relative error, (see Sect. 31)
p' a o^
s s^ cv
CV
NOTE
If a set of data is homogeneous in the sense of Section 3 of P A R T 1, it is usually safe to apply statistical theory and its concepts, like that of an expected value, to the data to assist in its analysis and interpretation. Only then is it meaningful to speak of a population average or other characteristic relating to a population (relative) frequency distribution function of X. This function commonly assumes the form of f(x), which is the probability (relative frequency) of an observation having exactly the value X, or the form of f(x)dx, which is the probability an observation has a value between x and x + dx. Mathematically the expected value of a function of X, say h(X), is defined as the sum (for discrete data) or integral (for continuous data) of that function times the probability of X and written E[h(X)]. For example, if the probability of X lying between x and x + dx based on continuous data is f(x)dx, then the expected value is lh(x)fix)dx = E[hix)].
'x
NOTE
The sample proportion p is an example of a sample average in which each observation is either a 1, the occurrence of a given type, or a 0, the nonoccurrence of the same type. The sample average is then exactly the ratio, p, of the total number of occurrences to the total number possible in the sample, n. If reference is to be made to the population from which a given sample came, the following symbols should be used. Yi Population skewness defined as the expected value (see Note) of (X - |i)^ divided by a^. It is spelled and pronounced "gamma one." Population coefficient of kurtosis defined as the amount by which the expected value (see Note) of (X - )x)* divided by a^ exceeds or falls short of 3; it is spelled and pronounced "gamma two." Population average or universe mean
If the probability of X lying between x and X + dx based on continuous data is f(x)dx, then the expected value is I.h(x)f(x)dx = E[h(x)].
72
|X
Sample statistics, like X, s^, gi, and g2, also have expected values in most practical cases, but these expected values relate to
PRESENTATION OF DATA the population frequency distribution of entire samples of n observations each, rather t h a n of individual observations. The expected value of X is [i, the same as that of an individual observation regardless of the population frequency distribution of X, and E(s2) = a^ likewise, but E(s) is less t h a n a in all cases and its value depends on the population distribution of X.
Firsi Type n Chvobsmafm l/iMffs mtachfhing V
Second Type One n Observaiions thing
INTRODUCTION 1. Purpose
PART 1 of the Manual discusses the application of statistical methods to the problem of: (a) condensing the information contained in a sample of observations, and (b) presenting the essential information in a concise form more readily interpretable than the unorganized mass of original data. Attention will be directed particularly to quantitative information on measurable characteristics of materials and manufactured products. Such characteristics will be termed quality characteristics.
a a
I
FIG. 1Two general types of data.
Data of the first type are commonly gathered to furnish information regarding the distribution of the quality of the material itself, having in mind possibly some more specific purpose; such as the establishment of a quality standard or the determination of conformance with a specified quality standard, for example, 100 observations of transverse strength on 100 bricks of a given brand. Data of the second type are commonly gathered to furnish information regarding the errors of measurement for a particular test method, for example, 50-micrometer measurements of the thickness of a test block.
2. Type of Data Considered

Consideration will be given to the treatment of a sample of n observations of a single variable. Figure 1 illustrates two general types: (a) the first type is a series of n observations representing single measurements of the same quality characteristic of n similar things, and (b) the second type is a series of n observations representing n measurements of the same quality characteristic of one thing. The observations in Figure 1 are denoted as Xi, where i = 1, 2, 3, ... , n. Generally, the subscript will represent the time sequence in which the observations were taken from a process or measurement. In this sense, we may consider the order of the data in Table 1 as being represented in a time-ordered manner.
NOTE The quality of a material in respect to some particular characteristic, such as tensile strength, is better represented by a frequency distribution function, t h a n by a single-valued constant. The variability in a group of observed values of such a quality characteristic is made up of two parts: variability of the material itself, and the errors of measurement. In some practical problems, the error of measurement may be large compared with the variability of the material; in others, the converse may be true. In any case, if one is interested in discovering the objective frequency distribution of the quality of the material, consideration must be given to correcting
8 DATA AND CONTROL CHARTANALYSIS

the errors of measurement (This is discussed in Ref. 1, pp. 379-384, in the seminal book on control chart methodology by Walter A. Shewhart.). If a given sample of data consists of two or more subportions collected under different test conditions or representing material produced under different conditions, it should be considered as two or more separate subgroups of observations, each to be treated independently in the analysis. Merging of such subgroups, representing significantly different conditions, may lead to a condensed presentation that will be of little practical value. Briefly, any sample of observations to which these methods are applied should be homogeneous. In the illustrative examples of PART 1, each sample of observations will be assumed to be homogeneous, that is, observations from a common universe of causes. The analysis and presentation by control chart methods of data obtained from several samples or capable of subdivision into subgroups on the basis of relevant engineering information is discussed in PART 3 of this Manual. Such methods enable one to determine whether for practical
3. H o m o g e n e o u s D a t a
While the methods here given may be used to condense any set of observations, the results obtained by using them may be of little value from the standpoint of interpretation unless the data are good in the first place and satisfy certain requirements. To be useful for inductive generalization, any sample of observations that is treated as a single group for presentation purposes should represent a series of measurements, all made under essentially the same test conditions, on a material or product, all of which has been produced under essentially the same conditions.
TABLE 1. Three groups of original data.

(a) Transverse Strength of 270 Bricks of a Typical Brand, psi
860 920
1200
1320 1100
820
1250 1100
1040 1480
1000 1150
1010
740
1070 1020 1170
1190 1080
1180
860
1380
1080 1000
1100
810
1360
1130 1000
850 920
1090
830 920
1070
890
1310
270
1330 1150
830
1390
940
1630
830
1120 1090
960 820
1170 2010
730
1330 1090
980
1160
670
1170 1340
920
1020
830
1040 1510
700 880
1080 1060 1230
910 870
1040
800 840
1240 1110
960
1180
890
1100
930
1260 1140
840
1020 1100
980 940
1060
800
1240 1020 1030
740 860
1290
880
1010
790
1130 1260
970
1050
870
1030 1100
740
1150 1000 1140 1030
990
1060
860 720
1080
840
1170
820 990
1020
860
1080
850
1070
900 890 970

1150
700 920 860 950

1020 1300
960 860
1100
800 990 870 660

1080
570 800
1180
970 790
1040
690
1070
820
1180
890 580
1350
700 820
1180
880
1060
980
1110
820
1230
950
1380 1380 1030
990 880 750 970

1030
890 970
1070
980 940
1000
800
1060
920 650
1610 1180
780 760 910 990 870

1180 1190 1050
950
1220 1020
900
1100 1380
760
1090 1010
900
1270
830
1100
830
1010
850
1230 1150
630 780
1400 1010
710
1000
900
1150
950 890
1360
860
1400
880
1010
730
1240 1190
910
890
1080
850
1110
970
1100
910 870 810
970
1070
960 870 910

1100
980 830 460
730
1030
920 800
1050 1070
780
1140
860
800 710 880
780 940
1010
980
1120
890
1240
940
860
PRESENTATION OF DATA purposes a given sample of observations may be considered to be homogeneous. specimens selected in a random manner to provide information about the quality of a larger quantity of materialthe general output of one brand of brick, a production lot of galvanized iron sheets, and a shipment of hard drawn copper wire. Consideration will be given to ways of arranging and condensing these data into a form better adapted for practical use.
4. Typical E x a m p l e s of P h y s i c a l D a t a
Table 1 gives three typical sets of observations, each one of these datasets represents measurements on a sample of units or
TABLE 1. Continued.
(b) Weight of Coating of 100 Sheets of Galvanized Iron Sheets, oz/ft^'' 1.467 1.623 1.520 1.767 1.550 1.533 1.377 1.373 1.637 1.460 1.627 1.537 1.533 1.337 1.603 1.373 1.457 1.660 1.323 1.647 1.603 1.603 1.383 1.730 1.700 1.600 1.603 1.477 1.513 1.533 1.593 1.503 1.600 1.543 1.567 1.490 1.550 1.577 1.483 1.600 1.577 1.577 1.323 1.620 1.473 1.420 1.450 1.337 1.440 1.557 1.480 1.477 1.550 1.637 1.570 1.617 1.477 1.750 1.497 1.717 1.563 1.393 1.647 1.620 1.530 1.470 1.337 1.580 1.493 1.563 1.543 1.567 1.670 1.473 1.633 1.763 1.573 1.537 1.420 1.513 1.437 1.350 1.530 1.383 1.457 1.443 1.473 1.433 1.637 1.500 1.607 1.423 1.573 1.753 1.467 1.563 1.503 1.550 1.647 1.690
(c) Breaking Strength of Ten Specimens of 0.104-in. HardDrawn Copper Wire, Ib'^ 578 572 570 568 572 570 570 572 576 584
" Measured to the nearest 10 psi. Test method used was ASTM Method of Testing Brick and Structural Clay (C 67). Data from ASTM Manual for Interpretation of Refractory Test Data, 1935, p. 83. ' Measured to the nearest 0.01 oz/ft^ of sheet, averaged for three spots. Test method used was ASTM Triple Spot Test of Standard Specifications for Zinc-Coated (Galvanized) Iron or Steel Sheets (A 93). This has been discontinued and was replaced by ASTM Specification for General Requirements for Steel Sheet, Zinc-Coated (Galvanized) by the Hot-Dip Process (A 525). Data from laboratory tests. 'Measured to the nearest 2 lb. Test method used was ASTM Specification for Hard-Drawn Copper Wire (B 1). Data from inspection report.
10 DATA AND CONTROL CHART ANALYSIS
500
1000 Transverse Strength, psi.
1500
2000
Fig. 2Showing graphicaiiy the ungrouped frequency distribution of a set of observations. Each dot represents one bricl<, data of Table 2(a}.
UNGROUPED WHOLE NUMBER DISTRIBUTION 5. Ungrouped Distribution

An arrangement of the observed values in ascending order of magnitude will be referred to in the Manual as the ungrouped frequency distribution of the data, to distinguish it from the grouped frequency distribution defined in Section 8. A further adjustment in the scale of the ungrouped distribution produces the whole number distribution. For example, the data of Table 1(a) were multiplied by lO^, and those of Table 1(b) by 103, ^ h i l e those of Table 1(c) were already whole numbers. If the data carry digits past the decimal point, just round until a tie (one observation equals some other) appears and then scale to whole numbers. Table 2 presents ungrouped frequency distributions for the three sets of observations given in Table 1. Figure 2 shows graphically the ungrouped frequency distribution of Table 2(a). In the graph, there is a minor grouping in terms of the unit of measurement. For the data of Fig. 2, it is the "rounding-off unit of 10 psi. It is rarely desirable to present data in the manner of Table 1 or Table 2. The mind cannot grasp in its entirety the meaning of so many numbers; furthermore, greater compactness is required for most of the practical uses t h a t are made of data.
sorted from smallest to largest. These features should make it easier to convert from an ungrouped to a grouped frequency distribution. More importantly, they allow calculation of the order statistics t h a t will aid in finding ranges of the distribution wherein lie specified proportions of the observations. A collection of observations is often seen as only a sample from a potentially huge population of observations and one aim in studying the sample may be to say what proportions of values in the population lie in certain ranges. This is done by calculating the percentiles of the distribution. We will see there are a number of ways to do this but we begin by discussing order statistics and empirical estimates of percentiles. A glance at Table 2 gives some information not readily observed in the original data set of Table 1. The data in Table 2 are arranged in increasing order of magnitude. When we arrange any data set like this the resulting ordered sequence of values are referred to as order statistics. Such ordered arrangements are often of value in the initial stages of an analysis. In this context, we use subscript notation and write X to denote the P'^ order statistic. For a sample of n values the order statistics are X(i) < X(2) < X(3) < ... < X(n). The index i is sometimes called the rank of the data point to which it is attached. For a sample size of n values, the first order statistic is the smallest or minimum value and has rank 1. We write this as X(i). The n" order statistic is the largest or maximum value and has rank n. We write this as X(n). The i*'' order statistic is written as X(i), for 1 < i < ;x. For the breaking strength data in Table 2c, the order statistics are: X(i)=568, X(2)=570, ... , X(io)=584. When ranking the data values, we may find some that are the same. In this situation, we say that a matched set of values constitutes a tie. The proper rank assigned to values that make up the tie is calculated by averaging the
6. Empirical Percentiles and Order Statistics

As should be apparent, the ungrouped whole number distribution may differ from the original data by a scale factor (some power of ten), by some rounding and by having been
11 PRESENTATION OF DATA
TABLE 2. Ungrouped frequency distributions in tabular form.
(a) Transverse Strength, psi (data of Table 1 (a))
270 460 570 580 630 650 660 670 690 700 700 700 710 710 720 730 730 730 740 740 740 750 760 760 780
780 780 780 790 790 800 800 800 800 800 800 800 810 810 820 820 820 820 820 820 820 830 830 830 830
830 830 830 840 840 840 850 850 850 850 860 860 860 860 860 860 860 860 860 860 870 870 870 870 870
870 880 880 880 880 880 880 890 890 890 890 890 890 890 890 900 900 900 900 910 910 910 910 910 920
920 920 920 920 920 930 940 940 940 940 940 950 950 950 950 960 960 960 960 970 970 970 970 970 970
970 980 980 980 980 980 980 990 990 990 990 990 1000 1000 1000
1020 1020 1020 1020 1020 1020 1020 1030 1030 1030 1030 1030 1030 1040 1040 1040 1040 1050 1050 1050 1060 1060 1060 1060 1060
1070 1070 1070 1070 1070 1070 1070 1080 1080 1080 1080 1080 1080 1080 1090 1090 1090 1090 1100 1100 1100 1100 1100 1100 1100
1100 1100 1100 1100 1110 1110 1110 1120 1120 1130 1130 1140 1140 1140 1150 1150 1150 1150 1150 1150 1160 1170 1170 1170 1170
1180 1180 1180 1180 1180 1180 1180 1190 1190 1190 1200 1220 1230 1230 1230 1240 1240 1240 1240 1250 1260 1260 1270 1290 1300
1310 1320 1330 1330 1340 1350 1360 1360 1380 1380 1380 1380 1390 1400 1400 1480 1510 1610 1630 2010
1000 1000 1000 1010 1010 1010 1010 1010 1010 1010
ranks that would have been determined by the procedure above in the case where each value was different from the others. For example, there are many ties present in Table 2. The rank associated with the three values of 700 would be the average of the ranks as if they were 700, 701, and 702, respectively. In other words, we see that the values of 700 occur in the 10*, llth^ and 12* positions, or represented as X(io), X(ii), and X(i2), respectively, if they were unequal. Thus, the value of 700 should carry a rank equal to (10+ll+12)/3 = 11, and each value specified as
X(ii).
less than that value. For example, the 5 0 * percentile, typically referred to as the median, is a value such t h a t half of the observations exceed it and half are below it. The 7 5 * percentile is a value such t h a t 25% of the observations exceed it and 75% are below it. The 9 0 * percentile is a value such that 10% of the observations exceed it and 90%) are below it. To aid in understanding the formulas that follow, consider finding the percentile t h a t best corresponds to a given order statistic. Although there are several answers to this question, one of the simplest is to realize that a sample of size n will partition the distribution from which it came into n+1 compartments as illustrated in the following figure.
The order statistics can be used for a variety of purposes, but it is for estimating the percentiles that they are used here. A percentile is a value that divides a distribution to leave a given fraction of the observations
12 DATA AND CONTROL CHART ANALYSTS

statistic. For X(i), the percentile is 100(l)/(24+l) - 4th; and for X(24), the percentile is 100(24/(24+1) = 96th. For the illustration in Figure 3, the point a corresponds to the 20*^ percentile, point b to the 40'^ percentile, point c to the GO'h percentile and point d to the 8 0 * percentile. It is not difficult to extend this application. From the figure it appears that the interval defined by a < x < d should enclose, on average, 60% of the distribution of X.
Fig. 3Any distribution is partitioned into n+1 compartments witti a sampie of n.
In Figure 3, the sample size is rt=4; the sample values are denoted as a, b, c and d. The sample presumably comes from some distribution as the figure suggests. Although we do not know the exact locations that the sample values correspond to along the true distribution, we observe that the four values divide the distribution into 5 roughly equal compartments. Each compartment will contain some percentage of the area under the curve so that the sum of each of the percentages is 100%. Assuming that each compartment contains the same area, the probability a value will fall into any compartment is 100[l/(n+l)]%. Similarly, we can compute the percentile that each value represents by 100[i/(n+l)]%, where i = 1, 2, ..., n. If we ask what percentile is the first order statistic among the four values, we estimate the answer as the 100[l/(4+l)]% = 20%, or 20th percentile. This is because, on average, each of the compartments in Figure 3 will include approximately 20% of the distribution. Since there are ?i+l=4+l=5 compartments in the figure, each compartment is worth 20%. The generalization is obvious. For a sample of n values, the percentile corresponding to the i'h order statistic is 100[i/(7i+l)]%, where i = 1, 2, ..., n. For example, if n=24 and we want to know which percentiles are best represented by the l^t and 24th order statistics, we can calculate the percentile for each order
We now extend these ideas to estimate the distribution percentiles. For the coating weights in Table 2(b), the sample size is n.=100. The estimate of the 50*^ percentile, or sample median, is the number lying halfway between the 50th and Sl'^t order statistics (X(50) = 1.537 and X(5i) = 1.543, respectively). Thus, the sample median is (1.537 +1.543)/2 = 1.540. Note that the middlemost values may be the same (tie). When the sample size is an even number, the sample median will always be taken as halfway between the middle two order statistics. Thus, if the sample size is 250, the median is taken as (X(i25)+X(i26))/2. If the sample size is an odd number, the median is taken as the middlemost order statistic. For example, if the sample size is 13, the sample median is taken as X(7). Note that for an odd numbered sample size, n, the index corresponding to the median will be i in+l)/2. We can generalize the estimation of any percentile by using the following convention. Let p be a proportion, so that for the 50th percentile p equals 0.50, for the 25th percentile p = 0.25, for the lO'h percentile p = 0.10, and so forth. To specify a percentile we need only specify p. An estimated percentile will correspond to an order statistic or weighted average of two adjacent order statistics. First, compute an approximate rank using the formula i = (n+l)p. If i is an integer then the lOOp"* percentile is estimated as X and we are done. If i is not an integer, then drop the decimal portion and keep the integer portion of i. Let k be the retained integer portion and r be the dropped decimal portion (note: 0<r<l).
TABLE 2 Continued.
(b) Weight of Coating, oz/ft^ (data of Table 1(6)) 1.323 1.323 1.337 1337 1.337 1.350 1.373 1.373 1.377 1.383 1.383 1.393 1.420 1.420 1.423 1.433 1.437 1.440 1.443 1.450 1.457 1.457 1.460 1.467 1.467 1.470 1.473 1.473 1.473 1.477 1.477 1.477 1.480 1.483 1.490 1.493 1.497 1.500 1.503 1.503 1.513 1.513 1.520 1.530 1.530 1.533 1.533 1.533 1.537 1.537 1.543 1.543 1.550 1.550 1.550 1.550 1.557 1.563 1.563 1.563 1.567 1.567 1.570 1.573 1.573 1.577 1.577 1.577 1.580 1.593 1.600 1.600 1.600 1.603 1.603 1.603 1.603 1.607 1.617 1.620 1.620 1.623 1.627 1.633 1.637 1.637 1.637 1.647 1.647 1.647 1.660 1.670 1.690 1.700 1.717 1.730 1.750 1.753 1.763 1.767
(c) Breaking Strength, lb. (data of Table 1(c)) 568 570 570 570 572 572 572 576 578 584
The estimated lOOp"* percentile is computed from the formula X(k) + r(X(k+i) - X(k)). Consider the transverse strengths with ?i=270 and let us find the 2.5'^ and 97.5*^ percentiles. For the 2.5*^^ percentile, p = 0.025. The approximate rank is computed as i = (270+1) 0.025 = 6.775. Since this is not an integer, we see that k-6 and r=0.775. Thus, the 2.5*'^ percentile is estimated by X(6) + r(X(7)X(6)), which is 650 + 0.775(660-650) = 657.75. For the 97.5'^ percentile, the approximate r a n k is i = (270+1) 0.975 = 264.225. Here = again, i is not an integer and so we use ^=264 and r=0.225; however; notice that both X(264) and X(265) are equal to 1400. In this case, the value 1400 becomes the estimate.
GROUPED FREQUENCY DISTRIBUTIONS
7. I n t r o d u c t i o n
Merely grouping the data values may condense the information contained in a set of observations. Such grouping involves some loss of information but is often useful in presenting engineering data. In the following sections, both tabular and graphical presentation of grouped data will be discussed.
8. Definitions
A grouped frequency distribution of a set of observations is an arrangement that shows the frequency of occurrence of the values of the variable in ordered classes. The interval, along the scale of measurement, of each ordered class is termed a bin. The frequency for any bin is the number of observations in t h a t bin. The frequency for a bin divided by the total number of observations is the relative frequency for that bin. Table 3 illustrates how the three sets of observations given in Table 1 may be organized into grouped frequency distributions. The recommended form of presenting tabular distributions is somewhat more compact, however, as shown in Table 4. Graphical presentation is used in Fig. 4 and discussed in detail in Section 14.
10. Number of Bins

The number of bins in a frequency distribution should preferably be between 13 and 20. (For a discussion of this point. See Ref. 1, p. 69, and Ref. 18, pp. 9-12.) Sturge's rule is to make the number of bins equal to l-t-3.31ogio(n). If the number of observations is, say, less t h a n 250, as few as 10 bins may be of use. When the number of observations is less than 25, a frequency distribution of the data is generally of little value from a presentation standpoint, as for example the 10 observations in Table 3(c). In general, the outline of a frequency distribution when presented graphically is more irregular when the number of bins is larger. This tendency is illustrated in Fig. 4.
11. Rules for Constructing Bins

After getting the ungrouped whole number distribution, one can use a number of popular computer programs to automatically construct a histogram. For example, a spreadsheet program, e.g., Excel^, can be used by selecting the Histogram item from the Analysis Toolpack menu. Alternatively, you can do it manually by applying the following rules: The number of bins (or "cells" or "levels") is set equal to NL = CEIL(2.1 log(n)), where n is the sample size and CEIL is an Excel spreadsheet function t h a t extracts the largest integer part of a decimal number, e.g., 5 is CEIL(4.1)). Compute the bin interval as LI = CEIL(RG/NL), where RG = LW-SW, and LW is the largest whole number and SW is the smallest among the n observations. Find the stretch adjustment as SA = CEIL((NL*LI-RG)/2). Set the start boundary at START = SW-SA-0.5 and then add LI successively NL times to get the bin boundaries. Average successive pairs of boundaries to get the bin midpoints.
9. Choice of Bin Boundaries

It is usually advantageous to make the bin intervals equal. It is recommended that, in general, the bin boundaries be chosen halfway between two possible observations. By choosing bin boundaries in this way, certain difficulties of classification and computation are avoided (See Ref. 2, pp. 73-76). With this choice, the bin boundary values will usually have one more significant figure (usually a 5) t h a n the values in the original data. For example, in Table 3(a), observations were recorded to the nearest 10 psi, hence the bin boundaries were placed at 225, 375, etc., rather than at 220, 370, etc., or 230, 380, etc. Likewise, in Table 3(6), observations were recorded to the nearest 0.01 oz/ft^, hence bin boundaries were placed at 1.275, 1.325, etc., rather than at 1.28, 1.33, etc.
' Excel is a trademark of Microsoft Corporation.
TABLE 3. Three examples of grouped frequency distribution, showing bin midpoints and bin boundaries. Bin Midpoint (a) Transverse strength, psi (data of Table 1 {a)) Bin Boundaries 225 300 375 450 525 600 675 750 825 900 975 1050 1125 1200 1275 1350 1425 1500 1575 1650 1725 1800 1875 1950 2025 Total (b) Weight of coating, oz/fl^ (data of Table 1 (b)) 1.275 1.300 1.325 1.350 1.375 1.400 1.425 1.450 1.475 1.500 1.525 1.550 1.575 1.600 1.625 1.650 1.675 1.700 1.725 1.750 1.775 Total (c) Breaking strength, lb (data Table 1 (c)) 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 Total 10 1 0 0 1 1 0 3 3 1 100 5 3 10 17 22 14 14 7 6 2 270 1 0 2 2 17 39 83 80 Observed Frequency
38
16 DATA AND CONTROL CHART ANALYSIS TABLE 4. Four methods of presenting a tabular frequency distribution (data of TABLE 1(a)).
(6) Relative Frequency (expressed in percentages) Percentage of Bricks Having Strength Within Given Limits 0.4 0.4 2.2 14.1 29.6 30.7 14.5 6.3 0.7 0.7 0.0 0.4 100.0
(a) Frequency Number of Bricks Having Strength Within Given Limits 1 1 6 38 80 83 39 17 2 2 0 1 270
Transverse Strength, psi 225 to 375 375 to 525 525 to 675 675 to 825 825 to 975 975 to 1125 1125 to 1275 1275 to 1425 1425 to 1575 1575 to 1725 1725 to 1875 1875 to 2025 Total
Transverse Strength, psi 225 to 375 375 to 525 525 to 675 675 to 825 825 to 975 975 to 1125 1125 to 1275 1275 to 1425 1425 to 1575 1575 to 1725 1725 to 1875 1875 to 2025
Total Number of observations = 270
(c) Cumulative Frequency Number of Bricks Having Strength less than Given Values 1 2 8 46 126 209 248 265 267 269 269 270
(d) Cumulative Relative Frequency (expressed in percentages) Percentage of Bricks Having Strength less t h a n Given Values 0.4 0.8 3.0 17.1 46.7 77.4 91.9 98.2 98.9 99.6 99.6 100.0 = 270
Transverse Strength, psi 375 525 675 825 975 1125 1275 1425 1575 1725 1875 2025
Transverse Strength, psi 375 525 675 825 975 1125 1275 1425 1575 1725 1875 2025
Number of observations
NOTE"Number of observations" should be recorded with tables of relative frequencies.
100 > 80
Using 12 cells, (Table III [a])
Using 19 cells
S 60h
I 40
500 1000 1500 2000
500 1000 1500 2000
Fig. 4Illustrating increased irregularity with larger number of cells, or bins.
17 PRESENTATION OF DATA Having defined the bins, the last step is to count the whole numbers in each bin and thus record the grouped frequency distribution as the bin midpoints with the frequencies in each. The user may improve upon the rules but they will produce a useful starting point and do obey the general principles of construction of a frequency distribution.
12. Tabular Presentation

Methods of presenting tabular frequency distributions are shown in Table 4. To make a frequency tabulation more understandable, relative frequencies may be listed as well as actual frequencies. If only relative frequencies are given, the table cannot be regarded as complete unless the total number of observations is recorded. Confusion often arises from failure to record bin boundaries correctly. Of the four methods, A to D, illustrated for strength measurements made to the nearest 10 lb., only Methods A and B are recommended (Table 5). Method C gives no clue as to how observed values of 2100, 2200, etc., which fell exactly at bin boundaries were classified. If such values were consistently placed in the next higher bin, the real bin boundaries are those of Method A. Method D is liable to misinterpretation since strengths were measured to the nearest 10 lb. only.
Figure 5 illustrates a convenient method of classifying observations into bins when the number of observations is not large. For each observation, a mark is entered in the proper bin. These marks are grouped in five's as the tallying proceeds, and the completed tabulation itself, if neatly done, provides a good picture of the frequency distribution. If the number of observations is, say, over 250, and accuracy is essential, the use of a computer may be preferred.
Transverse Strength, psi. 225 to 375 375 to 525 525 675 825 975 to to to to 675 825 975 1125 1275 1425 1575 1775
Frequency
1 1
1 1 6 38 80 83 39 17 2 2 0 1 Total
Fig. 5Method of classifying observations. Data of Table 1(a).
mi
mmmitfritfritiMttnii
1tfrltfr1tH~ltH*1tfrltfrltfrlHKllH-ttfr1lfr1Hf1tfr1^
1125 to 1275 to 1425 to 1575 to
itfrmitfritfrittt-iifrttfrmitfrmmmmmmm iifritifiiifmittntifmiiii
II II 1
mmitifii
1725 to 1875 1875 to 2025
270

TABLE 5. Methods A through D illustrated for strength measurements to the nearest 10 lb. RECOMMENDED METHOD A NUMBER OF OBSERVATIONS METHOD B NUMBER OF OBSERVATIONS NOT RECOMMENDED METHOD C NUMBER OF OBSERVATIONS METHOD D NUMBER OF OBSERVATIONS
STRENGTH, lb. 1995 to 2095 2095 to 2195 2195 to 2295 2295 to 2395 2395 to 2495 etc.
1 3 17 36 82
etc.
1 3 17 36 82
etc.
1 3 17 36 82
etc.
1 3 17 36 82
etc.
100 80 60 40 20 0 80 60 40 20 0
CD
Frequency 30 Bar Chart (Bars centered on cell midpoints) 20 10 0 Alternate Form 30 of Frequency Bar Chart 20 (Line erected at cell midpoints) 10
13. Graphical Presentation

Using a convenient horizontal scale for values of the variable and a vertical scale for bin frequencies, frequency distributions may be reproduced graphically in several ways as shown in Fig. 6. The frequency bar chart is obtained by erecting a series of bars, centered on the bin midpoints, with each bar having a height equal to the bin frequency. An alternate form of frequency bar chart may be constructed by using lines rather t h a n bars. The distribution may also be shown by a series of points or circles representing bin frequencies plotted at bin midpoints. The frequency polygon is obtained by joining these points by straight lines. Each endpoint is joined to the base at the next bin midpoint to close the polygon. Another form of graphical representation of a frequency distribution is obtained by placing along the graduated horizontal scale a series of vertical columns, each having a width equal to the bin width and a height equal to the bin frequency. Such a graph, shown at the bottom of Fig. 6, is called the frequency histogram of the distribution. In the histogram, if bin widths are arbitrarily given the value 1, the area enclosed by the steps represents frequency exactly, and the sides of the columns designate bin boundaries.
80 60 40 20 0 80 60 40 20 0 500 30 20 10 Frequency - Histogram 0 (Columns erected on cells) 30 20

Q.
-R
1000
1500
2000
10 0
Transverse Strength, psi. Frequency I I I 11 6l38l80l83l39ll7l2 I 2 lo I 1 I C e l l B o u n d r i e s S ^ g g m ^ | |

Cell Midpoint I30ol45ol60ol75ol90oll05oll20oll35oll50oll65olia0oll95ol
FIG. 6Graphical presentations of a frequency distribution. Data of Table 1(a) as grouped in Table 3(a).
19 PRESENTATION OF DATA The same charts can be used to show relative frequencies by substituting a relative frequency scale, such as that shown in Fig. 6. It is often advantageous to show both a frequency scale and a relative frequency scale. If only a relative frequency scale is given on a chart, the number of observations should be recorded as well. The lower chart shows a preferable method by plotting the relative cumulative frequencies on a normal probability scale. A Normal distribution (see Fig. 14) will plot cumulatively as a straight line on this scale. Such graphs can be drawn to show the number of observations either "less than" or "greater than" the scale values. (Graph paper with one dimension graduated in terms of the summation of Normal law distribution has been described in Refs. 3,18). It should be noted t h a t the cumulative percents need to be adjusted to avoid cumulative percents from equaling or exceeding 100%. The probability scale only reaches to 99.9% on most available probability plotting papers. Two methods which will work for estimating cumulative percentiles are [cumulative frequency/(n+1)], and [(cumulative frequency 0.5)/n].
14. Cumulative Frequency Distribution

Two methods of constructing cumulative frequency polygons are shown in Fig. 7. Points are plotted at bin boundaries. The upper chart gives cumulative frequency and relative cumulative frequency plotted on an arithmetic scale. This type of graph is often called an ogive or "s" graph. Its use is discouraged mainly because it is usually difficult to interpret the tail regions.
300
100
50 S
1500 Transverse Strength, psi. (a) Using arithmetic scale for frequency. (b) Using probability scale for relative frequency.
Fig. 7Graphical presentations of a cumulative frequency distribution. Data of Table 4: (a) using arithmetic scale for frequency, and (b) using probability scale for relative frequency.

For some purposes, the number of observations having a value "less than" or "greater than" particular scale values is of more importance t h a n the frequencies for particular bins. A table of such frequencies is termed a cumulative frequency distribution. The "less than" cumulative frequency distribution is formed by recording the frequency of the first bin, then the sum of the first and second bin frequencies, then the sum of the first, second, and third bin frequencies, and so on. Because of the tendency for the grouped distribution to become irregular when the number of bins increases, it is sometimes preferable to calculate percentiles from the cumulative frequency distribution rather than from the order statistics. This is recommended as n passes the hundreds and reaches the thousands of observations. The method of calculation can easily be illustrated geometrically by using Table 4(d), Cumulative Relative Frequency and the problem of getting the 2.5*11 and 97.5'^ percentiles. We first define the cumulative relative frequency function, F(x), from the bin boundaries and the cumulative relative frequencies. It is just a sequence of straight lines connecting the points (X=235, F(235)=0.000), (X=385, F(385)=0.0037), (X=535, F(535)=0.0074), and so on up to (X=2035, F(2035)=1.000). Notice in Fig. 7, with a n arithmetic scale for percent, and you can see the function. A horizontal line at height 0.025 will cut the curve between X=535 and X=685, where the curve rises from 0.0074 to 0.0296. The full vertical distance is 0.02960.0074 = 0.0222, and the portion lacking is 0.0250-0.0074 = 0.0176, so this cut will occur at (0.0176/0.0222) 150+535 = 653.9 psi. The horizontal at 97.5% cuts the curve at 1419.5 psi. The first step is to reduce the data to two or three-digit numbers by: (1) dropping constant initial or final digits, like the final zero's in Table 1(a) or the initial one's in Table 1(b); (2) removing the decimal points; and finally, (3) rounding the results after (1) and (2), to two or three-digit numbers we can call coded observations. For instance, if the initial one's and the decimal points in the data of Table 1(b) are dropped, the coded observations r u n from 323 to 767, spanning 445 successive integers. If forty successive integers per class interval are chosen for the coded observations in this example, there would be 12 intervals; if thirty successive integers, then 15 intervals; and if twenty successive integers then 23 intervals. The choice of 12 or 23 intervals is outside of the recommended interval from 13 to 20. While either of these might nevertheless be chosen for convenience, the flexibility of the stem and leaf procedure is best shown by choosing thirty successive integers per interval, perhaps the least convenient choice of the three possibilities. Each of the resulting 15 class intervals for the coded observations is distinguished by a first digit and a second. The third digits of the coded observations do not indicate to which intervals they belong and are therefore not needed to construct a stem and leaf diagram in this case. But the first digit may change (by one) within a single class interval. For instance, the first class interval with coded observations beginning with 32, 33 or 34 may be identified by 3(234) and the second class interval by 3(567), but the third class interval includes coded observations with leading digits 38, 39 and 40. This interval may be identified by 3(89)4(0) The intervals, identified in this manner, are listed in the left column of Fig. 8. Each coded observation is set down in t u r n to the right of its class interval identifier in the diagram using as a symbol its second digit, in the order (from left to right) in which the original observations occur in Table 1(b). In spite of the complication of changing some first digits within some class intervals, this stem and leaf diagram is quite simple to construct. In this particular case, the diagram reveals "wings" at both ends of the diagram.
15. "Stem and Leaf Diagram

It is sometimes quick and convenient to construct a "stem and l e a f diagram, which has the appearance of a histogram turned on its side. This kind of diagram does not require choosing explicit bin widths or boundaries.
First (and second) Digit: 3(234) 3(567) 3(89)4(0) 4(123) 4(456) 4(789) 5(012) 5(345) 5(678) 5(9)6(01) 6(234) 6(567) 6(89)7(0) 7(123) 7(456) Second Digits Only 32233 7775 898 22332 66554546 798787797977 2 10100 53333455534335 677776866776 000090010 23242342334 67 09 31 6565
The 'TJOX" is formed by the 2 5 * and 75'*^
percentiles, the center of the data is dictated by the 50'^ percentile (median) and "whiskers" are formed by extending a line from either side of the box to the minimum, X(i) point, and to the maximum, X(n) point. Fig. 8b shows the box plot for the data from Table 1(b). For further information on boxplots, see Ref. 18.
First (and second) Digit: 3(234) 3(567) 3(89)4(0) 4(123) 4(456) 4(789) 5(012) 5(345) 5(678) 5(9)6(01) 6(234) 6(567) 6(89)7(0) 7(123) 7(456) Second Digits Only 22333 5777 889 22233 44555666 777777788999 000 112 33333334455555 666667777778 90 0 0 0 0 0 0 1 22223333444 67 90 13 5566
FIG. 8Stem and leaf diagram of data from Table 1(b) with groups based on triplets of first and second decimal digits.
As this example shows, the procedure does not require choosing a precise class interval width or boundary values. At least as important is the protection against plotting and counting errors afforded by using clear, simple numbers in the construction of the diagrama histogram on its side. For further information on stem and leaf diagrams see Refs. 4 and 18.
FIG. 8aOrdered stem and leaf diagram of data from Table 1(b) with groups based on triplets of first and second decimal digits. The 25'", 50*^ and 75"^ quartiles are shown in bold type and are underlined.
1.323 1.4678 1.540 1.6030
1.767
16. "Ordered Stem and Leaf Diagram and Box Plot

The stem and leaf diagram can be extended to one that is ordered. The ordering pertains to the ascending sequence of values within each "leaf. The purpose of ordering the leaves is to make the determination of the quartiles an easier task. The quartiles represent the 2b''^, 50"i (median), and 75'^ percentiles of the frequency distribution. They are found by the method discussed in Section 6. In Fig. 8a, the quartiles for the data are bold and underlined. The quartiles are used to construct another graphic called a box plot.
FIG. 8bBox plot of data from Table 1(b)
The information contained in the data may also be summarized by presenting a tabular grouped frequency distribution, if the number of observations is large. A graphical presentation of a distribution makes it possible to visualize the nature and extent of the observed variation. While some condensation is effected by presenting grouped frequency distributions, further reduction is necessary for most of the uses t h a t are made of ASTM data. This need can be fulfilled by means of a few simple functions of the observed distribution, notably, the average and the standard deviation.
22 DATA AND CONTROL CHART ANALYSIS FUNCTIONS OF A FREQUENCY DISTRIBUTION 17. Introduction
In the problem of condensing and summarizing the information contained in the frequency distribution of a sample of observations, certain functions of the distribution are useful. For some purposes, a statement of the relative frequency within stated limits is all t h a t is needed. For most purposes, however, two salient characteristics of the distribution which are illustrated in Fig. 9a are: (a) the position on the scale of measurementthe value about which the observations have a tendency to center, and (b) the spread or dispersion of the observations about the central value. A third characteristic of some interest, but of less importance, is the skewness or lack of symmetrythe extent to which the observations group themselves more on one side of the central value than on the other (see Fig. 9b). A fourth characteristic is "kurtosis" which relates to the tendency for a distribution to have a sharp peak in the middle and excessive frequencies on the tails as compared with the Normal distribution or conversely to be relatively flat in the middle with little or no tails (see Fig. 10).
Swwd
Different pMrifions, Mmc
..itl 1 Mil....
.lllllh.
yilllllllllllllii.......
Same Powtion, different spreads
illlllll
Scale of mtesurtmenf
..lllllll..
>
Different Positions, different spreads
FIG. 9aIllustrating two salient characteristics of distributionsposition and spread.
Negative Skewness 91 =-1.00
Positive Skewness
g, = 0
g, = +1.00
. . . i i I'l
1111
iu.
I llm.
Scale of Measurement
FIG. 9bIllustrating a third characteristic of frequency distributionsskewness, and particular values of skewness, g,.
Leptokurtic
92=100
Mesokurtic
92=0
Platykurtic
92=-0.9
ILO.
JLx.
FIG. 10Illustrating the kurtosis of a frequency distribution and particular values of g^
Several representative sample measures are available for describing these characteristics, but by far the most useful are the arithmetic mean X, the standard deviation s, the skewness factor g^, and the kurtosis factor g2all algebraic functions of the observed values. Once the numerical values of these particular measures have been determined, the original data may usually be dispensed with and two or more of these values presented instead. The four characteristics of the distribution of a sample of observations just discussed are most useful when the observations form a single heap with a single peak frequency not located at either extreme of the sample values. If there is more than one peak, a tabular or graphical representation of the frequency distribution conveys information the above four characteristics do not.
19. A v e r a g e (Arithmetic Mean)

The average (arithmetic mean) is the most widely used measure of central tendency. The term average and the symbol X will be used in this Manual to represent the arithmetic mean of a sample of numbers. The average, X, of a sample of n numbers, Xi, Xg,..., Xn, is the sum of the numbers divided by n, that is
(1) where the expression .Z^ Xi means "the sum of all values of X, from Xi to X, inclusive." Considering the n values of X as specifying the positions on a straight line of n particles of equal weight, the average corresponds to the center of gravity of the system. The average of a series of observations is expressed in the same units of measurement as the observations, t h a t is, if the observations are in pounds, the average is in pounds.
18. Relative F r e q u e n c y
The relative frequency p within stated limits on the scale of measurement is the ratio of the number of observations lying within those limits to the total number of observations. In practical work, this function has its greatest usefulness as a measure of fraction nonconforming, in which case it is the fraction, p, representing the ratio of the number of observations lying outside specified limits (or beyond a specified limit) to the total number of observations.
20. Other M e a s u r e s of Central Tendency

The geometric mean, of a sample of n numbers, Zi, Xj,..., Xn, is the n."" root of their product, that is

geometric mean = ^X\X2 X or log (geometric mean) logX,+logX2+ +logX
v\2
(2)
s =.
(Xi - Xf
+{X2-Xf n-\
++ {X - Xf (4)
(3)
where X is defined by Eq 1. The quantity s^ is called the sample variance. The standard deviation of any series of observations is expressed in the same units of measurement as the observations, t h a t is, if the observations are in pounds, the standard deviation is in pounds. (Variances would be measured in pounds squared.) A frequently more convenient formula for the computation of s is
Equation 3, obtained by taking logarithms of both sides of Eq 2, provides a convenient method for computing the geometric mean using the logarithms of the numbers. NOTE The distribution of some quality characteristics is such that a transformation, using logarithms of the observed values, gives a substantially Normal distribution. When this is true, the transformation is distinctly advantageous for (in accordance with Section 29) much of the total information can be presented by two functions, the average, X, and the standard deviation, s, of the logarithms of the observed values. The problem of transformation is, however, a complex one that is beyond the scope of this Manual. See Ref. 18. The median of the frequency distribution of n numbers is the middlemost value. The mode of the frequency distribution of n numbers is the value that occurs most frequently. With grouped data, the mode may vary due to the choice of the interval size and the starting points of the bins.
Is^-(=1
n-\
(5)
but care must be taken to avoid excessive rounding error when n is larger t h a n s.
NOTE A useful quantity related to the standard deviation is the root-mean-square deviation
mx-x)
i=l
T7\2 = S,
\n-\
^(rms)
21. S t a n d a r d D e v i a t i o n The standard deviation is the most widely used measure of dispersion for the problems considered in PART 1 of the Manual. For a sample of n numbers, Xi, X2..., X^, the sample standard deviation is commonly defined by the formula
22. O t h e r M e a s u r e s of D i s p e r s i o n The coefficient of variation, cv, of a sample of n numbers, is the ratio (sometimes the coefficient is expressed as a percentage) of their standard deviation, s, to their average X. It is given by
(6)
25 PRESENTATION OF DATA The coefficient of variation is an adaptation of the standard deviation, which was developed by Prof. Karl Pearson to express the variability of a set of numbers on a relative scale rather t h a n on an absolute scale. It is thus a dimensionless number. Sometimes it is called the relative standard deviation, or relative error. The average deviation of a sample of n numbers, X^, X^, ..., X, is the average of the absolute values of the deviations of the numbers from their average X that is
X(^,-^)'
!=1
^1
(8)
ns
This measure of skewness is a pure number and may be either positive or negative. For a symmetrical distribution, gi is zero. In general, for a nonsymmetrical distribution, g^ is negative if the long tail of the distribution extends to the left, towards smaller values on the scale of measurement, and is positive if the long tail extends to the right, towards larger values on the scale of measurement. Figure 9 shows three unimodal distributions with different values of gj.
t,\x-x\
average deviation (7) 23a. Kurtosisg2 where the symbol | | denotes the absolute value of the quantity enclosed. The range i? of a sample of n numbers is the difference between the largest number and the smallest number of the sample. One computes R from the order statistics as R = X(n)-X(i). This is the simplest measure of dispersion of a sample of observations. The peakedness and tail excess of a sample frequency distribution is generally measured by the coefficient of kurtosis ^2The coefficient of kurtosis ga for a sample of n numbers, Xi, X^, ..., X, is defined by the expression g2 = k4/s^ and
23. Skewnessg^
A useful measure of the lopsidedness of a sample frequency distribution is the coefficient of skewness gi. The coefficient of skewness g^, of a sample of n numbers, X^, X^, ..., X^, is defined by the expression gi = ks/s^. Where ks is the third kstatistic as defined by R. A. Fisher. The kstatistics were devised to serve as the moments of small sample data. The first moment is the mean, the second is the variance, and the third is the average of the cubed deviations and so on. Thus, ki= X ,k2s^,
K = {n-\){n-2){n-3)
Notice that when n is large
(-2)(-3)
g2=-^ ns
(9)
K'l
(n-l)(-2)
Notice that when n is large
Again this is a dimensionless number and may be either positive or negative. Generally, when a distribution has a sharp peak, thin shoulders, and small tails relative to the bell-shaped distribution characterized by the Normal distribution, g2 is positive. When a distribution is flat-topped with fat tails, relative to the Normal distribution, gz is negative. Inverse relationships do not necessarily follow. We cannot definitely infer anything about the shape of a distribution from knowledge of g2 unless we are willing to assume some theoretical curve, say a Pearson curve, as being

appropriate as a graduation formula (see Fig. 14 and Section 30). A distribution with a positive g2 is said to be leptokurtic. One with a negative ^2 is said to be platykurtic. A distribution with ^2 = 0 is said to be mesokurtic. Figure 10 gives three unimodal distributions with different values of ^2-
SE{x)=sl4n,
SB (g-2) = yjlAIn , respectively.
24. Computational Tutorial

The method of computation can best be illustrated with an artificial example for n=4 with Xi = 0, X2 = 4, Xs = 0, and X4 = 0. Please first verify that X= 1. The deviations from this mean are found as - 1 , 3, - 1 , and - 1 . The sum of the squared deviations is thus 12 and s^ = 4. The sum of cubed deviations is 1+27-1-1 = 24, and thus ks = 16. Now we find gi = 16/8 2. Please verify that g2 = 4. Since both gi and g2 are positive, we can say that the distribution is both skewed to the right and leptokurtic relative to the Normal distribution. Of the many measures that are available for describing the salient characteristics of a sample frequency distribution, the average X, the standard deviation s, the skewness gi, and the kurtosis g2, are particularly useful for summarizing the information contained therein. So long as one uses them only as rough indications of uncertainty we list approximate sampling standard deviations of the quantities X, s^, gi and g2, as When using a computer software calculation, the ungrouped whole number distribution values will lead to less round off in the printed output and are simple to scale back to original units. The results for the data from Table 2 are given in Table 6.
AMOUNT OF INFORMATION CONTAINED IN p , J^, s, g^, AND g^ 25. Summarizing the Information
Given a sample of n observations, Xi, X2, X3, ..., Xn, of some quality characteristic, how can we present concisely information by means of which the observed distribution can be closely approximated, that is, so that the percentage of the total number, n, of observations lying within any stated interval from, say, X-atoX = b, can be approximated?
Table 6. Summary Statistics for Three Sets of Data
Datasets Transverse Strength, psi Weight of Coating,

Oz/ft2
X
999.8
s 201.8
gi
g2
0.611
2.567
1.535
0.1038
0.013
-0.291
Breaking Strength, lb
573.2
4.826
1.419
1.797
27 PRESENTATION OF DATA The total information can be presented only by giving all of the observed values. It will be shown, however, that much of the total information is contained in a few simple functionsnotably the average X, the standard deviation s, the skewness î, and the kurtosis ^2-
NOTE
For the purposes of PART 1 of this Manual, the curves of Figs. 11 and 12 may be taken to represent frequency histograms with small bin widths and based on large samples. In a frequency histogram, such as that shown at the bottom of Fig. 5, let the percentage relative frequency between any two bin boundaries be represented by the area of the histogram between those boundaries, the total area being 100 percent. Since the bins are of uniform width, the relative frequency in any bin is then proportional to the height of t h a t bin and may be read on the vertical scale to the right. Average
X, Xj X3
26. Several Values of Relative Frequency, p

By presenting, say, 10 to 20 values of relative frequency p, corresponding to stated bin intervals and also the number n of observations, it is possible to give practically all of the total information in the form of a tabular grouped frequency distribution. If the ungrouped distribution has any peculiarities, however, the choice of bins may have an important bearing on the amount of information lost by grouping.
1 1
\ 1
1
>' t
211
1 ' . ^ ^
,' ( 3\.'.
27, Single Percentile of Relative Frequency, p

If we present but a percentile value, Qp, of relative frequency p, such as the fraction of the total number of observed values falling outside of a specified limit and also the number n of observations, the portion of the total information presented is very small. This follows from the fact that quite dissimilar distributions may have identically the same percentile value as illustrated in Fig. 11.
Specified Limit (min.
^.'-'
FIG. 12Quite different distributions may have the same average.
Q FIG. 11Quite different distributions may have the same percentile value of p, fraction of total observations below specified limit.
If the sample size is increased and the bin width reduced, a histogram in which the relative frequency is measured by area approaches as a limit the frequency distribution of the population, which in many cases can be represented by a smooth curve. The relative frequency between any two values is then represented by the area under the curve and between ordinates erected at those values. Because of the method of generation, the ordinate of the curve may be regarded as a curve of relative frequency density. This is analogous to the representation of the variation of density along a rod of uniform cross section by a smooth curve. The weight between any two points along the rod is proportional to the area under the curve between the two ordinates and we may speak of the density (that is, weight density) at any point but not of the weight at any point.
28 DATA AND CONTROL CHART ANALYSIS 28. Average X Only

If we present merely the average, X, and number, n, of observations, the portion of the total information presented is very small. Quite dissimilar distributions may have identically the same value of X as illustrated in Fig. 12. In fact, no single one of the five functions, Qp, X, s, gi, or g2, presented alone, is generally capable of giving much of the total information in the original distribution. Only by presenting two or three of these functions can a fairly complete description of the distribution generally be made. An exception to the above statement occurs when theory and observation suggest that the underlying law of variation is a distribution for which the basic characteristics are all functions of the mean. For example, "life" data "under controlled conditions" sometimes follows a negative exponential distribution. For this, the cumulative relative frequency is given by the equation F{X) = l-e
-x/Q
of the exponential function, yield a fitting formula from which estimates can be made of the percentage of cases lying between any two specified values of X. Presentation of X and n is sufficient in such cases provided they are accompanied by a statement that there are reasons to believe that X has a negative exponential distribution.
29. Average X and S t a n d a r d Deviation s

These two functions contain some information even if nothing is known about the form of the observed distribution, and contain much information when certain conditions are satisfied. For example, more than 1 - 1/k'' of the total number n of observations lie within the closed interval X ks (where k is not less t h a n 1). This is Chebyshev's inequality and is shown graphically in Fig. 13. The inequality holds true of any set of finite numbers regardless of how they were obtained. Thus if X a n d s are presented, we may say at once that more t h a n 75 percent of the numbers lie within the interval X 2s; stated in another way, less than 25 percent of the numbers differ from X by more t h a n 2s. Likewise, more t h a n 88.9 percent lie within the interval X 3s, etc. Table 7 indicates the conformance with Chebyshev's inequality of the three sets of observations given in Table 1.
0<X<oo
(14)
This is a single parameter distribution for which the mean and standard deviation both equal 0. That the negative exponential distribution is the underlying law of variation can be checked by noting whether values of 1 F(X) for the sample data tend to plot as a straight line on ordinary semi-logarithmic paper. In such a situation, knowledge of X will, by taking 0 = X in Eq. 14 and using tables
Percentage .75.00. 0 40 60 70 I 80
I,
.88.89.
.93.75. 90 92 94
1 1,1 I, 1 I ^ l | i . i i i . . ^ l . . I i i I I I ^ I I I, I ,
FIG. 13Percentage of the total observations lying within the interval x Acs always exceeds the percentage given on this chart.
29 PRESENTATION OF DATA TABLE 7. Comparison of observed percentages and Chebyshev's minimum percentages of the total observations lying within given intervals.
CHEBYSHEVS MINIMUM OBSERVATIONS LYING WITHIN THE GIVEN_ INTERVAL X ks 75.0
84.0
OBSERVED PERCENTAGES" DATA OF TABLE 1(a) {n = 270) DATA OF TABLE 1(6) {n = 100) DATA OF TABLE 1(c) (n = 10) 90 90 100
INTERVAL,
Xks
X 2.0s X 2.5s X 3.0s
96.7 97.8 98.5
94 100 100
88.9
"Data of Table 1(a): X= 1000, s = 202; data of Table 1(&): X= 1.535, s = 0.105; data of Table 1(c): X = 573.2, s = 4.58.
To determine approximately just what percentages of the total number of observations lie within given limits, as contrasted with minimum percentages within those limits, requires additional information of a restrictive nature. If we present X, s, and n, and are able to add the information "data obtained under controlled conditions," then it is possible to make such estimates satisfactorily for limits spaced equally above and below X. What is meant technically by "controlled conditions" is discussed by Shewhart (see Ref. 1) and is beyond the scope of this Manual. Among other things, the concept of control includes the idea of homogeneous dataa set of observations resulting from measurements made under the same essential conditions and representing material produced under the same essential conditions. It is sufficient for present purposes to point out t h a t if data are obtained under "controlled conditions," it may be assumed that the observed frequency distribution can, for most practical purposes, be graduated by some theoretical curve say, by the Normal law or by one of the nonnormal curves belonging to the system of frequency curves developed by Karl Pearson. (For an extended discussion of Pearson curves, see Ref. 5). Two of these are illustrated in Fig. 14.
The applicability of the Normal law rests on two converging arguments. One is mathematical and proves that the distribution of a sample mean obeys the Normal law no matter what the shape of the distributions are for each of the separate observations. The other is that experience with many, many sets of data show that more of them approximate the Normal law t h a n any other distribution. In the field of statistics, this effect is known as the central limit theorem. Supposing a smooth curve plus a gradual approach to the horizontal axis at one or both sides derived the Pearson system of curves. The Normal distribution's fit to the set of data may be checked roughly by plotting the cumulative data on Normal probability paper (see Section 13). Sometimes if the original data do not appear to follow the Normal law, some transformation of the data, such as log X, will be approximately normal. Thus, the phrase "data obtained under controlled conditions" is taken to be the equivalent of the more mathematical assertion t h a t "the functional form of the distribution may be represented by some specific curve." However, conformance of the shape of a frequency distribution with some curve should by no means be taken as a sufficient criterion for control.
Ben snaped
Examples of two Pearson non-normal frequency curves
FIG. 14Â frequency distribution of observations obtained under controlled conditions will usually have an outline that conforms to the Normal law or a non-normal Pearson frequency curve.
Percentage .6827 0 10 20 30 40 50 60 ,70 80 90 I I . I. .1 I .1 l l l i u i ' m i l i i u l iUJ I 0.6745 1.282 k

FIG. 15Normal law integral diagram giving percentage of total area under Normal law curve falling within the range \i ka. This diagram is also useful in probability and sampling problems, expressing the upper (percentage) scale values in decimals to represent "probability."
95.45 95;
-W
99.73. 99 T 2.326 2.576 99.5
1.645
1.960
Generally for controlled conditions, the percentage of the total observations in the original sample lying within the interval Xks may be determined approximately from the chart of Fig. 15, which is based on the Normal law integral. The approximation may be expected to be better the larger the number of observations. Table 8 compares the observed percentages of the total number of observations lying within several symmetrical intervals about X with those estimated from a knowledge of X and s, for the three sets of observations given in Table 1.
30. A v e r a g e X, S t a n d a r d D e v i a t i o n s, S k e w n e s s gi, a n d Kurtosis g2

If the data are obtained under "controlled conditions" and if a Pearson curve is assumed appropriate as a graduation formula, the presentation of ^1 and g2 in addition to X and s will contribute further information. They will give no immediate help in determining the percentage of the total observations lying within a symmetric interval about the average X, t h a t is, in the interval of X ks.
TABLE 8. Comparison of observed percentages and theoretical estimated percentages of the total observations lying within given intervals.
THEORETICAL ESTIMATED PERCENTAGES" OF TOTAL OBSERVATIONS LYING WITHIN THE INTERVAL, GIVEN INTERVAL X 50.0 68.3 86.6 95.5 98.7 99.7 ks
OBSERVED PERCENTAGES DATA OF TABLE 1(a) {n = 270) 52.2 76.3 89.3 96.7 97.8 98.5 DATA OF TABLE 1(6) {n = 100) 54 72 84 94 100 100 DATA OF TABLE 1(c) {n = 10) 70 80 90 90 90 100
X ks
X 0.6745s X 1.0s X 1.5s X 2.0s X 2.5s X3.0s
"Use Fig. 15 with X and s as estimates of |x and a.
What they do is to help in estimating observed percentages (in a sample already taken) in an interval whose limits are not equally spaced above and below X. If a Pearson curve is used as a graduation formula, some of the information given by g^ and g^ may be obtained from Table 9 which is taken from Table 42 of the Biometrika Tables for Statisticians. For j3, = gf and jS^ = g^ + 3, this table gives values of ^^ for use in estimating the lower 2.5 percent of the data and values of k^j for use in estimating the upper 2.5 percent point. More specifically, it may be estimated that 2.5 percent of the cases are less than X-k^s and 2.5 percent are greater than X + k^s Put another way, it may be estimated that 95 percent of the cases are between X-k^s and X + k^sTable 42 of the Biometrika Tables for Statisticians also gives values of ki and ku for 0.5, 1.0, and 5.0 percent points.
(6) we may estimate that approximately 95 percent of the 270 cases lie between X k^s and X + kyS,or between 1000 - 1.801 (201.8) = 636.6 and 1000 + 2.17 (201.8) = 1437.7. The actual percentage of the 270 cases in this range is 96.3 percent (see Table 2(a)). Notice that using just Xl.96s gives the interval 604.3 to 1395.3 which actually includes 95.9% of the cases versus a theoretical percentage of 95%. The reason we prefer the Pearson curve interval arises from knowing that the gi = 0.63 value has a standard error of 0-15 (= V6/270) and is thus about four standard errors above zero. That is, if future data come from the same conditions it is highly probable that they will also be skewed. The 604.3 to 1395.3 interval is symmetric about the mean, while the 636.6 to 1437.7 interval is offset in line with the anticipated skewness. Recall that the interval based on the order statistics was 657.8 to 1400 and that from the cumulative frequency distribution was 653.9 to 1419.5. When computing the median, all methods will give essentially the same result but we need to choose among the methods when estimating a percentile near the extremes of the distribution.
Example For a sample of 270 observations of the transverse strength of bricks, the sample distribution is shown in Fig. 5. From the sample values of g^ = 0.61 and ga = 2.57, we take pi = gi2 = (0.61)2 = 0.37 and P2 = g2 + 3 = 2.57 + 3 = 5.57. Thus, from Tables 9(a) and
TABLE 9. Lower and upper 2.5 percent points ^^ and k^ of the standardized deviate (X- |x)/o, given by Pearson frequency curves for designated values of Pi(estimated as equal to g^) and PJ (estimated as equal to ^2 + 3). Pilp2 (a) Lower" /JL 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 5.0 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 5.0 0.00 1.65 1.76 1.83 1.88 1.92 1.94 1.96 1.97 1.98 1.99 1.99 1.99 2.00 2.00 2.00 2.00 2.00 1.65 1.76 1.83 1.88 1.92 1.94 1.96 1.97 1.98 1.99 1.99 1.99 2.00 2.00 2.00 2.00 2.00 0.01 1.68 1.76 1.82 1.86 1.89 1.91 1.93 1.94 1.95 1.95 1.96 1.96 1.96 1.96 1.97 1.97 1.82 1.89 1.94 1.97 1.99 2.01 2.02 2.02 2.02 2.03 2.03 2.03 2.03 2.03 2.03 2.03 0.03 1.62 1.71 1.77 1.82 1.85 1.87 1.89 1.90 1.91 1.92 1.93 1.93 1.94 1.94 1.94 1.94 1.86 1.93 1.98 2.01 2.03 2.04 2.05 2.05 2.05 2.05 2.05 2.05 2.05 2.05 2.05 2.05 0.05 1.56 1.66 1.73 1.78 1.82 1.84 1.86 1.88 1.89 1.90 1.91 1.91 1.92 1.92 1.93 1.93 1.89 1.96 2.01 2.03 2.05 2.06 2.07 2.07 2.07 2.07 2.07 2.07 2.07 2.07 2.07 2.07 0.10 0.15 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
1.57 1.65 1.71 1.76 1.79 1.81 1.83 1.85 1.86 1.87 1.88 1.88 1.89 1.89 1.90
1.49 1.58 1.64 1.70 1.74 1.77 1.79 1.81 1.82 1.84 1.84 1.85 1.86 1.87 1.87
1.41 1.51 1.58 1.65 1.69 1.72 1.75 1.77 1.79 1.81 1.82 1.83 1.83 1.84 1.85
1.39 1.47 1.55 1.60 1.65 1.68 1,71 1.73 1.75 1.76 1.78 1.79 1.80 1.81
1.37 1.45 1.52 1.57 1.61 1.65 1.67 1.70 1.72 1.73 1.75 1.76 1.77
1.35 1.42 1.49 1.54 1.58 1.62 1.64 1.67 1.69 1.70 1.72 1.73
1.33 1.40 1.46 1.51 1.56 1.59 1.62 1.64 1.66 1.68 1.69
1.32 1.39 1.44 1.49 1.53 1.56 1.59 1.62 1.64 1.65
1.24 1.31 1.38 1.43 1.47 1.51 1.54 1.57 1.59 1.61
1.23 1.30 1.36 1.41 1.45 1.49 1.52 1.55 1.57
1.23 1.29 1.35 1.40 1.44 1.47 1.50 1.53
(b) Upper ^L
2.00 2.05 2.08 2.09 2.10 2.11 2.11 2.11 2.11 2.11 2.10 2.10 2.10 2.10 2.09
2.04 2.08 2.11 2.13 2.13 2.14 2.14 2.14 2.13 2.13 2.13 2.13 2.12 2.12 92.1
2.06 2.11 2.14 2.15 2.16 2.16 2.16 2.16 2.16 2.15 2.15 2.15 2.14 2.14 22.1
2.15 2.18 2.20 2.21 2.21 2.21 2.20 2.20 2.19 2.19 2.18 2.18 2.17 2.17
2.22 2.24 2.25 2.25 2.25 2.24 2.24 2.23 2.22 2.22 2.21 2.21 2.20
2.27 2.28 2.29 2.28 2.28 2.27 2.26 2.25 2.25 2.24 2.23 2.23
2.32 2.32 2.32 2.31 2.30 2.29 2.28 2.28 2.27 2.26 2.25
2.35 2.35 2.34 2.33 2.32 2.31 2.31 2.30 2.29 42.2
2.38 2.38 2.37 2.36 2.35 2.34 2.33 2.32 2.31 2.30
2.41 2.41 2.40 2.38 2.37 2.36 2.35 2.34 82.3
2.44 2.43 2.41 2.40 2.39 2.38 2.36 2.35
NOTESThis table was reproduced from Biometrika Tables for Statisticians, Vol. 1, p. 207, with the kind permission of the Biometrika Trust. The Biometrika Tables also give the lower and upper 0.5, 1.0, and 5 percent points. Use for a large sample only, say n > 250. Take /x = X and a = s.
''When gj > 0, the skewness is taken to be positive, and the deviates for the lower percentage points are negative.
As a first step, one should scan the data to assess its approach to the Normal law. We suggest dividing gi and g2 by their standard errors and if either ratio exceeds 3 then look to see if there is an outlier. An outlier is an observation so small or so large that there are no other observations near it and so extreme that persons familiar with the measurements can assert that such extreme value will not arise in the future under ordinary conditions. A glance at Fig. 2 suggests the presence of
outliers but we must suppose that the second criterion was not satisfied. If any observations seem to be outliers then discard them. If n is very large, say M>10000, then use the percentile estimator based on the order statistics. If the ratios are both below 3 then use the Normal law for smaller sample sizes. If n is between 1000 and 10000 but the ratios suggest skewness and/or kurtosis, then use the cumulative frequency function. For smaller sample sizes and
33 PRESENTATION OF DATA evidence of skewness and/or kurtosis, use the Pearson system curves. Obviously, these are rough guidehnes and the user must adapt them to the actual situation by trying alternative calculations and then judging the most reasonable. are respectively the largest and smallest values in the sample. If the population distribution is known to be Normal it might also be said, with a 0.90 probability of being right, that 99 percent of the values of the population will lie in the interval X 2.703J. Further information on statistical tolerances of this kind is presented in Refs. 6, 7, and 18.
NOTE ON TOLERANCE LIMITS In Sections 33 through 34, the percentages of X values estimated to be within a specified range pertain only to the given sample of data which is being represented succinctly by selected statistics, X, s, etc. The Pearson curves used to derive these percentages are used simply as graduation formulas for the histogram of the sample data. The aim of Sections 33 to 34 is to indicate how much information about the sample is given by X, s, gi, and g2. It should be carefully noted that in an analysis of this kind the selected ranges of X and associated percentages are not to be confused with what in the statistical literature are called "tolerance limits." In statistical analysis, tolerance limits are values on the X scale that denote a range which may be stated to contain a specified minimum percentage of the values in the population there being attached to this statement a coefficient indicating the degree of confidence in its truth. For example, with reference to a random sample of 400 items, it may be said, with a 0.91 probability of being right, that 99 percent of the values in the population from which the sample came will be in the interval X(4oo) - X(i) where X(4oo) and X(i)
31. Use of Coefficient of Variation I n s t e a d of t h e S t a n d a r d D e v i a t i o n

So far as quantity of information is concerned, the presentation of the sample coefficient of variation, cv, together with the average, X, is equivalent to presenting the sample standard deviation, s, and the average, X, since s may be computed directly from the values of cv = sjX and X. In fact, the sample coefficient of variation (multiplied by 100) is merely the sample standard deviation, s, expressed as a percentage of the average, X. The coefficient of variation is sometimes useful in presentations whose purpose is to compare variabilities, relative to the averages, of two or more distributions. It is also called the relative standard deviation or relative error.
Example 1 Table 10 presents strength test results for two different materials. It can be seen that whereas the standard deviation for Material B is less than the standard deviation for Material A, the latter shows the greater relative variability as measured by the coefficient of variability.
TABLE 10. Strength test results. NUMBER OF AVERAGE STANDARD OBSERVATIONS, STRENGTH, LB, DEVIATION, LB, n s X 160 150 1100 800 225 200 COEFFICIENT OF VARIATION, % cv 20.4 25.0
MATERIAL1 A B
34 DATA AND CONTROL CHART ANALYSTS TABLE 11. Data for two test conditions. TEST CONDITION A B NUMBER OF SPECIMENS n 50 50 AVERAGE LIFE o 14 h 80 min STANDARD DEVIATION s 4.2 h 23.2 min COEFFICIENT OF VARMTION, m, % 30.0 29.0
The coefficient of variation is particularly applicable in reporting the results of certain measurements where the variability, a, is known or suspected to depend upon the level of the measurements. Such a situation may be encountered when it is desired to compare the variability (a) of physical properties of related materials usually at different levels, (b) of the performance of a material under two different test conditions, or (c) of analyses for a specific element or compound present in different concentrations.
Example 2 The performance of a material may be tested under widely different test conditions as for instance in a standard life test and in an accelerated life test. Further, the units of measurement of the accelerated life tester may be in minutes and of the standard tester, in hours. The data shown in Table 11 indicate essentially the same relative variability of performance for the two test conditions.
Taking this as a starting point and considering the fact that in ASTM work the intention is, in general, to avoid throwing together into a single series data obtained under widely different conditionsdifferent in an important sense in respect to the characteristic under inquirywe believe that it is possible, in general, to use the methods indicated in Sections 33 and 34 for making rough estimates of the observed percentages of a frequency distribution, at least for making estimates (per Section 33) for symmetric ranges around the average, that is, X ks. This belief depends, to be sure, upon our own experience with frequency distributions and upon the observation that such distributions tend, in general, to be unimodalto have a single peakas in Fig. 14. Discriminate use of these methods is, of course, presumed. The methods suggested for controlled conditions could not be expected to give satisfactory results if the parent distribution were one like that shown in Fig. 16a bimodal distribution representing two different sets of conditions. Here, however, the methods could be applied separately to each of the two rational subgroups of data.
32. General Comment on Observed Frequency Distributions of a Series of ASTM Observations

Experience with frequency distributions for physical characteristics of materials and manufactured products prompts the committee to insert a comment at this point. We have yet to find an observed frequency distribution of over 100 observations of a quality characteristic and purporting to represent essentially uniform conditions, that has less t h a n 96 percent of its values within the range X 3s. For a Normal distribution, 99.7 percent of the cases should theoretically lie between ^ 3a as indicated in Fig. 15.
Fig. 16A bimodal distribution arising from two different systems of causes.
33. SummaryAmount of Information Contained in Simple Functions of t h e Data

The material given in Sections 24 to 32, inclusive, may be summarized as follows:
1. If a sample of observations of a single variable is obtained under controlled conditions, much of the total information contained therein may be made available by presenting four functionsthe average X, the standard deviation, s, the skewness, î the kurtosis, g^ and the number n, of observations. Of the four functions, X and s contribute most; g^ and g2 contribute in accord with how small or how large are their standard errors, namely -J^ln and ^lAIn2. The average, X, and the standard deviation, s, give some information even for data that are not obtained under controlled conditions. No single function, such as the average, of a sample of observations is capable of giving much of the total information contained therein unless the sample is from a universe that is itself characterized by a single parameter. To be confident, the population that has this characteristic will usually require much previous experience with the kind of material or phenomenon under study.
supporting evidence for his or her conclusions. The objective is to present t h a t portion of the total information given by the original data that is believed to be essential for the intended use. Essential information will be described as follows: "We take data to answer specific questions. We shall say that a set of statistics (functions) for a given set of data contains the e s s e n t i a l information given by the data when, through the use of these statistics, we can answer the questions in such a way that further analysis of the data will not modify our answers to a practical extent," (taken from PART 2, Ref. 1). The Preface to this Manual lists some of the objectives of gathering ASTM data of the type under discussiona sample of observations of a single variable. Each such sample constitutes an observed frequency distribution, and the information contained therein should be used efficiently in answering the questions that have been raised.
3.
35. W h a t F u n c t i o n s of t h e D a t a Contain the Essential Information The nature of the questions asked determine what part of the total information in the data constitutes the essential information for use in interpretation. If we are interested in the percentages of the total number of observations that have values above (or below) several values on the scale of measurement, the essential information may be contained in a tabular grouped frequency distribution plus a statement of the number of observations n. But even here, if n is large and if the data represent controlled conditions, the essential information may be contained in the four sample functionsthe average X, the standard deviation s, the skewness g^, the kurtosis g^, and the number of observations n. If we are interested in the average and variability of the quality of a material, or in the average quality of a material and some measure of the variability of averages for successive samples, or in a comparison of the
J u s t what functions of the data should be presented, in any instance, depends on what uses are to be made of the data. This leads to a consideration of what constitutes the "essential information."
ESSENTIAL INFORMATION 34. Introduction

Presentation of data presumes some intended use either by others or by the author as
36 DATA AND CONTROL CHART ANAT,YSTS

average and variability of the quality of one material with t h a t of other materials, or in the error of measurement of a test, or the like, then the essential information may be contained in the X , s, and n of each sample of observations. Here, if n is small, say 10 or less, much of the essential information may be contained in the X, R (range), and n of each sample of observations. The reason for use of R when n < 10 is as follows: It is important to note (see Ref. 8) that the expected value of the range R (largest observed value minus smallest observed value) for samples of n observations each, drawn from a Normal universe having a standard deviation a varies with sample size in the following manner. The expected value of the range is 2.1(7 for n = 4, 3.1(7 for n = 10, 3.9(7 for n = 25, and 6.1(7 for n = 500. From this it is seen that in sampling from a Normal population, the spread between the maximum and the minimum observation may be expected to be about twice as great for a sample of 25, and about three times as great for a sample of 500, as for a sample of 4. For this reason, n should always be given in presentations which give R. In general, it is better not to use R i n exceeds 12. If we are also interested in the percentage of the total quantity of product that does not conform to specified limits, then part of the essential information may be contained in the observed value of fraction defective p. The conditions under which the data are obtained should always be indicated, i.e., (a) controlled, (b) uncontrolled, or (c) unknown. If the conditions under which the data were obtained were not controlled, then the maximum and minimum observations may contain information of value. It is to be carefully noted that if our interest goes beyond the sample data themselves to the processes that generated the samples or might generate similar samples in the future, we need to consider errors that may arise from sampling. The problems of sampling errors that arise in estimating process means, variances, and percentages are discussed in PART 2. For discussions of sampling errors in comparisons of means and variabilities of different samples, the reader is referred to texts on statistical theory (for example, Ref. 9). The intention here is simply to note those statistics, those functions of the sample data, which would be useful in making such comparisons and consequently should be reported in the presentation of sample data.
36. P r e s e n t i n g X Only V e r s u s Presenting X and s

Presentation of the essential information contained in a sample of observations commonly consists in presenting X, s, and n. Sometimes the average alone is givenno record is made of the dispersion of the observed values nor of the number of observations taken. For example. Table 12 gives the observed average tensile strength for several materials under several conditions.
TABLE 12. Information of value may be lost if only the average is presented. Tensile Strength psi Condition a Average, J Condition b Average, x Condition c Average, x
Material
A B C
51430 59 060 57 710
47 200 57 380 74 920
49 010 60 700 80 460
The objective quality in each instance is a frequency distribution, from which the set of observed values might be considered as a sample. Presenting merely the average, and failing to present some measure of dispersion
TABLE 13. Presentation of essential information (data of TABLE 8). Tensile Strength, psi Condition a Average, Material A B C Tests 20 18 27 51430 59 060 75 710 Standard Deviation, s 920 1320 1840 Condition b Standard Average, Deviation, s 47 200 57 380 74 920 830 1360 1650 Condition c Average, 49 010 60 700 80 460 Standard Deviation, 1070 1480 1910
and the number of observations generally loses much information of value. Table 13 corresponds to Table 12 and provides what will usually be considered as the essential information for several sets of observations, such as data collected in investigations conducted for the purpose of comparing the quality of different materials.
37. Observed R e l a t i o n s h i p s
ASTM work often requires the presentation of data showing the observed relationship between two variables. Although this subject does not fall strictly within the scope of PART 1 of the Manual, the following material is included for general information. Attention will be given here to one type of relationship, where one of the two variables is of the nature of temperature or timeone that is controlled at will by the investigator and considered for all practical purposes as capable of "exact" measurement, free from experimental errors. (The problem of presenting information on the observed relationship between two statistical
variables, such as hardness and tensile strength of an alloy sheet material, is more complex and will not be treated here. For further information, see Refs. 1, 2, and 9.) Such relationships are commonly presented in the form of a chart consisting of a series of plotted points and straight lines connecting the points or a smooth curve which has been "fitted" to the points by some method or other. This section will consider merely the information associated with the plotted points, i.e., scatter diagrams. Figure 17 gives an example of such an observed relationship. (Data are from records of shelf life tests on die-cast metals and alloys, former Subcommittee 15 of ASTM Committee B-2 on Non-Ferrous Metals and Alloys.) At each successive stage of an investigation to determine the effect of aging on several alloys, five specimens of each alloy were tested for tensile strength by each of several laboratories. The curve shows the results obtained by one laboratory for one of these alloys. Each of the plotted points is the average of five observed values of tensile strength and thus attempts to summarize an observed frequency distribution.
Years
Fig. 17Exampte of graph showing an observed relationship.
34.000 32,000 0 2 Years 3
Observed value *' Average of observed value Ôbjective distribution
Fig. 18Showing pictorially what lies bacl( of the plotted points in Fig. 17. Each plotted point in Fig. 17 is the average of a sample from a universe of possible observations.
Figure 18 has been drawn to show pictorially what is behind the scenes. The five observations made at each stage of the life history of the alloy constitute a sample from a universe of possible values of tensile strengthan objective frequency distribution whose spread is dependent on the inherent variability of the tensile strength of the alloy and on the error of testing. The dots represent the observed
values of tensile strength and the bellshaped curves the objective distributions. In such instances, the essential information contained in the data may be made available by supplementing the graph by a tabulation of the averages, the standard deviations, and the number of observations for the plotted points in the manner shown in Table 14.
TABLE 14. Summary of essential information for Fig. 18.
Tensile strength, psi Time of Test Initial 6 months 1 year 2 years 5 years Number of Specimens 5 5 5 5 5 Average, x Standard Deviation, ^ 950 668 869 655 319
35 400 35 980 36 220 37 460 36 800
39 PRESENTATION OF DATA 38. S u m m a r y : E s s e n t i a l Information The material given in Sections 34 to 37, inclusive, may be summarized as follows. 1. What constitutes the essential information in any particular instance depends on the nature of the questions to be answered, and on the nature of the hypotheses that we are willing to make based on available information. Even when measurements of a quality characteristic are made under the same essential conditions, the objective quality is a frequency distribution t h a t cannot be adequately described by any single numerical value. 2. Given a series of observations of a single variable arising from the same essential conditions, it is the opinion of the committee that the average, X, the standard deviation, s, and the number, n, of observations contain the essential information for a majority of the uses made of such data in ASTM work. Empirical knowledge is not contained in the observed data alone, rather it arises from interpretationan act of thought. (For an important discussion on the significance of prior information and hypothesis in the interpretation of data, see Ref. 10; a treatise on the philosophy of probable inference which is of basic importance in the interpretation of any and all data is presented in Ref. 11.) Interpretation consists in testing hypotheses based on prior knowledge. Data constitute but a part of the information used in interpretationthe judgments that are made depend as well on pertinent collateral information, much of which may be of a qualitative rather t h a n of a quantitative nature. If the data are to furnish a basis for most valid prediction, they must be obtained under controlled conditions and must be free from constant errors of measurement. Mere presentation does not alter the goodness or badness of data. However, the usefulness of good data may be enhanced by the manner in which they are presented.
40. R e l e v a n t I n f o r m a t i o n NOTE If the observations are not obtained under the same essential conditions, analysis and presentation by the control chart method, in which order (see P A R T 3 of this Manual) is taken into account by rational subgrouping of observations, commonly provides important additional information. Presented data should be accompanied by any or all available relevant information, particularly information on precisely the field within which the measurements are supposed to hold and the condition under which they were made, and evidence t h a t the data are good. Among the specific things that may be presented with ASTM data to assist others in interpreting them or to build up confidence in the interpretation made by an author are:
PRESENTATION OF RELEVANT INFORMATION
1. The kind, grade, and character of material or product tested. 2. The mode and conditions of production, if this has a bearing on the feature under inquiry.

3. The method of selecting the sample; steps taken to ensure its randomness or representativeness. (The manner in which the sample is taken has an important bearing on the interpretability of data and is discussed by Dodge (see Ref. 12).) The specific method of test (if an ASTM or other standard test, so state; together with any modifications of procedure). The specific conditions of test, particularly the regulation of factors that are known to have an influence on the feature under inquiry. The precautions or steps taken to eliminate systematic or constant errors of observation. The difficulties encountered and eliminated during the investigation. Information regarding parallel but independent paths of approach to the end results. Evidence that the data were obtained under controlled conditions; the results of statistical tests made to support belief in the constancy of conditions, in respect to the physical tests made or the material tested, or both. (Here, we mean constancy in the statistical sense, which encompasses the thought of stability of conditions from one time to another and from one place to another. This state of affairs is commonly referred to as "statistical control." Statistical criteria have been developed by means of which we may judge when controlled conditions exist. Their character and mode of application are given in PART 3 of this Manual; see also Ref. 13.) be vague, yet without it the interpretation of the data and the conclusions reached may be misleading or of little value to others.
41. Evidence of Control

One of the fundamental requirements of good data is that they should be obtained under controlled conditions. The interpretation of the observed results of an investigation depends on whether or not there is justification for believing that the conditions were controlled. If the data are numerous and statistical tests for control are made, evidence of control may be presented by giving the results of these tests. (For examples, see Refs. 14-18.) Such quantitative evidence greatly strengthens inductive arguments. In any case, it is important to indicate clearly just what precautions were taken to control the essential conditions. Without tangible evidence of this character, the reader's degree of rational belief in the results presented will depend on his faith in the ability of the investigator to eliminate all causes of lack of constancy.
4.
5.
6.
7.
8.
9.
RECOMMENDATIONS 42. Recommendations for Presentation of Data

The following recommendations for presentation of data apply for the case where one has at hand a sample of n observations of a single variable obtained under the same essential conditions. 1. Present as a minimum, the average, the standard deviation, and the number of observations. Always state the number of observations taken. If the number of observations is moderately large (7i>30), present also
Much of this information may be qualitative in character, some may even
2.
41 PRESENTATION OF DATA the value of the skewness, î, and the value of the kurtosis, ^2- An additional procedure when n is large (n>100) is to present a graphical representation, such as a grouped frequency distribution. If the data were not obtained under controlled conditions and it is desired to give information regarding the extreme observed effects of assignable causes, present the values of the maximum and minimum observations in addition to the average, the standard deviation, and the number of observations. 4. Present as much evidence as possible that the data were obtained under controlled conditions. Present relevant information on precisely (a) the field within which the measurements are believed valid and (6) the conditions under which they were made.
[7] Bowker, A. H. and Lieberman, G. J., Engineering Statistics, 2nd ed., Section 8.12, Prentice-Hall, 1972. [8] Tippett, L. H. C , "On the Extreme Individuals and the Range of Samples Taken From a Normal Population," Biometrika, Vol. 17, Dec. 1925, pp. 364-387. [9] Hoel, P. G., Introduction to Mathematical Statistics, 5th ed., Wiley, New York, 1984. [10] Lewis, C. I., Mind and Scribner, New York, 1929. the World Order,
[11] Keynes, J. M., A Treatise MacMillan, New York, 1921.
on
Probability,
[12] Dodge, H. F., "Statistical Control in Sampling Inspection," presented at a round table discussion on "Acquisition of Good Data," held at the 1932 Annual Meeting of the American Society for Testing and Materials; published in American Machinist, 26 Oct. and 9 Nov. 1932. [13] Pearson, E. S., "A Survey of the Uses of Statistical Method in the Control and Standardization of the Quality of Manufactured Products," Journal of the Royal Statistical Society, Vol. XCVI, P a r t 11, 1933, pp. 21-60. [14] Passano, R. F., "Controlled Data from an Immersion Test," Proceedings, American Society for Testing and Materials, Vol. 32, Part 2, 1932, p. 468. [15] Skinker, M. F., "Application of Control Analysis to the Quality of Varnished Cambric Tape," Proceedings, American Society for Testing and Materials, Vol. 32, Part 3, 1932, p. 670. [16] Passano, R. F. and Nagley, F. R., "Consistent Data Showing the Influences of Water Velocity and Time on the Corrosion of Iron," Proceedings, American Society for Testing and Materials, Vol. 33, Part 2, p. 387. [17] Chancellor, W. C , "Application of Statistical Methods to the Solution of Metallurgical Problems in Steel Plant," Proceedings, American Society for Testing and Materials, Vol. 34, P a r t 2, 1934, p. 891. [18] Ott, E. R., SchiUing, E. G., and Neubauer, D. V., Process Quality Control, 3rd ed., McGraw-Hill, New York, N.Y., 2000.
REFERENCES
[1] Shewhart, W. A., Economic Control of Quality of Manufactured Product, Van Nostrand, New York, 1931; republished by ASQC Quahty Press, Milwaukee, WI, 1980. [2] Yule, G. U. and Kendall, M. G., An Introduction to the Theory of Statistics, 14th ed., Charles Griffin and Company, Ltd., London, 1950. [3] Box, G. E. P., Hunter, W. G., and Hunter, J. S., Statistics for Experimenters, Wiley, New York, 1978, pp. 329-330. [4] Tukey, J. W., Exploratory Data Analysis, Addison-Wesley, Reading, PA, 1977, pp. 1-26. [5] Elderton, W. P. and Johnson, N. L., Systems of Frequency Curves, Cambridge University Press, Bentley House, London, 1969. [6] Duncan, A. J., Quality Control and Industrial Statistics, 5th ed., Chapter 6, Sections 4 and 5, Richard D. Irwin, Inc., Home Wood, IL, 1986.
MNL7A-EB/Feb. 2002
Presenting Plus or Minus Limits of Uncertainty of on Observed Average

GLOSSARY OF SYMBOLS USED IN PART 2
1^ a Population mean Factor, given in Table 2 of PART 2, for computing confidence limits for [i associated with a desired value of probability, P, and a given number of observations, n Deviation of a Normal variable Number of observed values (observations) Sample fraction nonconforming Population fraction nonconforming Population standard deviation Probability; used in PART 2 to designate the probability associated with confidence limits; relative frequency with which the averages |J, of sampled populations may be expected to be included within the confidence limits (for |x) computed from samples Sample standard deviation Estimate of a based on several samples Observed value of a measurable characteristic; specific observed values are designated X^, X2, X3, etc.; also used to designate a measurable characteristic Sample average (arithmetic mean), the sum of the n observed values in a set divided by n
2. T h e P r o b l e m
A n o b s e r v e d a v e r a g e X, is subject to t h e u n c e r t a i n t i e s t h a t a r i s e from s a m p l i n g fluctuations a n d t e n d s to differ from t h e population mean. The smaller the n u m b e r of o b s e r v a t i o n s , n, t h e larger the fluctuations a r e likely to b e . W i t h a s e t of n o b s e r v e d v a l u e s of a v a r i a b l e X, w h o s e a v e r a g e ( a r i t h m e t i c m e a n ) is X, a s i n T a b l e 1, i t is often d e s i r e d to i n t e r p r e t t h e r e s u l t s in s o m e w a y . O n e w a y is to c o n s t r u c t a n i n t e r v a l s u c h t h a t t h e m e a n , ft = 573.2 3.5 lb., lies w i t h i n l i m i t s b e i n g e s t a b l i s h e d from the quantitative d a t a along with t h e i m p l i c a t i o n s t h a t t h e m e a n \i of t h e p o p u l a t i o n s a m p l e d is i n c l u d e d w i t h i n t h e s e l i m i t s w i t h a specified p r o b a b i l i t y . How should such limits be computed, a n d w h a t m e a n i n g m a y b e a t t a c h e d to t h e m ?
k n P P'
(J
s c X
TABLE 1. Breaking strength often specimens of 0.104-in. hard-drawn copper wire.

Specimen 1 2 3 4 5 6 7 8 9 10 re = 10 Average, ^ Standard deviation, s Breaking Strength, X, lb 578 572 570 568 572 570 570 572 576 584 5732 573.2 4.83
1. Purpose
P A R T 2 of t h e M a n u a l d i s c u s s e s t h e p r o b l e m of p r e s e n t i n g p l u s or m i n u s l i m i t s to i n d i c a t e t h e u n c e r t a i n t y of t h e a v e r a g e of a n u m b e r of o b s e r v a t i o n s o b t a i n e d u n d e r t h e s a m e essential conditions, and s u g g e s t s a form of p r e s e n t a t i o n for u s e in ASTM reports and publications where needed.
Copyright 2002 by A S T M International
www.astm .org
43 LIMITS OF UNCERTAINTY
NOTE
The mean |i, is the value of X that would be approached as a statistical limit as more and more observations were obtained under the same essential conditions and their cumulative averages computed.
NOTE While the method of Section 4 represents the best that can be done at present in interpreting a sample X and s, when no other information regarding the variability of the population is available, a much more satisfactory interpretation can be made in general if other information regarding the variability of the population is at hand, such as a series of samples from the universe or similar populations for each of which a value of s or Jf? is computed. If s or i? displays statistical control, as outlined in PART 3 of this Manual, and a sufficient number of samples (preferably 20 or more) are available to obtain a reasonably precise estimate of c, designated as a, the limits of uncertainty for a sample containing any number of observations, n, and arising from a population whose true standard deviation can be presumed equal to a, can be computed from the following formula:
3. Theoretical Background
Mention should be made of the practice, now mostly out of date in scientific work, of recording such limits as
X 0.6745 V where X = observed average, s = observed standard deviation, and n = number of observations. and referring to the value 0.6145 s/^Jn as the "probable error" of the observed average, X. (Here the value of 0.6745 corresponds to the Normal law probability of 0.50, see Table 8 of PART 1.) The term "probable error" and the probability value of 0.50 properly apply to the errors of sampling when sampling from a universe whose average, p., and whose standard deviation, o, are known (these terms apply to limits \i 0.6745CT/Vn), but they do not apply in the inverse problem when merely sample values of X and s are given. Investigation of this problem [1-3] has given a more satisfactory alternative (section 4), a procedure which provides limits that have a definite operational meaning.
Xk
where k = 1.645, 1.960, and 2.576 for probabilities of P = 0.90, 0.95, and 0.99, respectively.
4. Computation of Limits
The following procedure applies to any long run series of problems for each of which the following conditions are met: Given: A sample of n observations, Xi, X2, X3, ..., Xn, having an average = X a n d a standard deviation = s.
TABLE 2. Factors for calculating 90, 95, and 99 percent confidence limits for averages."
Confidence Limits, * X :kas
N u m b e r of O b s e r v a t i o n s in Sample, n 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 n greater t h a n 25
90% Confidence Limits ( P = 0.90) Value of a 1.177 0.953 0.823 0.734 0.670 0.620 0.580 0.546 0.518 0.494 0.473 0.455 0.438 0.423 0.410 0.398 0.387 0.376 0.367 0.358 0.350 0.342
1.645
95% Confidence Limits ( P = 0.95) Value of a 1.591 1.241 1.050 0.925 0.836 0.769 0.715 0.672 0.635 0.604 0.577 0.554 0.533 0.514 0,497 0.482 0.468 0.455 0.443 0.432 0.422 0.413
1560
99% Confidence Limits (P = 0.99) Value of a 2.921 2.059 1.646 L401 1.237 1.118 1.028 0.955 0.897 0.847 0.805 0.769 0.737 0.708 0.683 0.660 0.640 0.621 0.604 0.588 0.573 0.559
2.576
approximately
approximately
approximately
*Limits which may be expected to include |X (9 times in 10, 95 times in 100, or 99 times in 100) in a series of problems, each involving a single sample of n observations. Values of a computed from Fisher, R. A., "Table of t" Statistical Methods for Research Workers, Table IV based on Student's distribution of t. " Recomputed in 1975. The a of this Table 2 equals Fisher's for n - 1 degrees of freedom divided by Vn. See also Fig. 1.
45 LIMITS OF UNCERTAINTY 5.0
4.0
3.0
4--
2.0
''^
/
/ s
> 6
z_..v -ii: /' 7

^ y
1.0 0.9
.8
SO.8 o J 0.7
0.6
?î;:i:î!i=Hii?i^
1 M
T'
^ ; i : E = :12EE
^^,i__j,.2o--
= z j ! " = i;:25 = = 0.5 0.4
0.3
0.2
H
/./././'^'Z^^
4^-i
U
11^
.zzzzzztzy.zt^t/tt^ Z Z ZZ/lZZ^.Zt/t^Z'
II III lilil 1
r
I.A
'
J- ' i '
'?7 .' >
^ ^'
J /
^ ^
z
>^
0.1
o o
Value of P o
CD O O O O
00
en
oo
05
03
FIG. 1Curves giving factors for calculating 50 to 99 percent confidence limits for averages (see also Table 2). Redrawn in 1975 for new values of a. Error in reading a not likely to be > 0.01. The numbers printed by curves are the sample sizes n.

Conditions: (a) The population sampled is homogeneous (statistically controlled) in respect to X, the variable measured, (b) The distribution of X for the population sampled is approximately Normal, (c) The sample is a random sample.' Procedure: Compute limits These limits are commonly referred to as "confidence limits" [4,5,18] for the three columns of Table 2 they may be referred to as the "90 percent confidence limits," "95 percent confidence limits" and "99 percent confidence limits," respectively. The magnitude P = 0.95 applies to the series of samples, and is approached as a statistical limit as the number of instances in the series is increased indefinitely; hence it signifies "statistical probability." If the values of a given in Table 2 for P = 0.99 are used in a series of samples, we may, in like manner, expect 99 percent of the sample intervals so computed to include the population mean |i. Other values of P could, of course, be used if desiredthe use of chances of 95 in 100, or 99 in 100 are, however, often found convenient in engineering presentations. Approximate values of a for other values of P may be read from the curves in Fig. 1, for samples otn -25 or less. For larger samples (n greater than 25), the constants, 1.645, 1.960, and 2.576, in the expressions 1645 1960 2.576
Xas
where the value of a is given in Table 2 for three values of P and for various values of
Meaning: If the values of o given in Table 2 for P = 0.95 are used in a series of such problems, then, in the long run, we may expect 95 percent of the intervals bounded by the limits so computed, to include the population averages p. of the populations sampled. If in each instance, we were to assert t h a t jx is included within the limits computed, we should expect to be correct 95 times in 100 and in error 5 times in 100; t h a t is, the statement "ji is included within the interval so computed" has a probability of 0.95 of being correct. But there would be no operational meaning in the following statement made in any one instance: "The probability is 0.95 that |j. falls within the limits computed in this case" since |a, either does or does not fall within the limits. It should also be emphasized that even in repeated sampling from the same population, the interval defined by the limits Xas will vary in width and position from sample to sample, particularly with small samples (see Fig. 2). It is this series of ranges fluctuating in size and position which will include, ideally, the population mean |J,, 95 times out of 100 for P - 0.95.
a =
, and a =
at the foot of Table 2 are obtained directly from Normal law integral tables for probability values of 0.90, 0.95, and 0.99. To find the value of this constant for any other value of P, consult any standard text on statistical methods or read the value approximately on the "^" scale of Fig. 15 of PART 1 of this Manual. For example, use of a = l/Vn yields P = 68.27% and the limits plus or minus one standard error, which some scientific journals print without noting a percent.
If the population sampled is finite, that is, made up of a finite number of separate units that may be measured in respect to the variable, X, and if interest centers on the [i of this poptJation, then this procedure assumes that the number of units, n, in the sample is relatively small compared with the number of units, A^, in the population, say n is less than about 5 percent of A^. However, correction for relative size of sample can be made by multiplying > s by the factor Jl-(n/ N). On the other hand, if interest centers on the
\i of the underlying process or soiffce of the finite population, then this correction factor is not used.
47 LIMITS OF UNCERTAINTY 2.0

00 CO o +1
O W .^ _ ^
Case (4) P=0.50
i|||||i1j.|||l|''|l''i|Ul''i|ili'lj'i'||i|'''i'i|i''l ii'i'i'
Case (B) P=0.90
I X .E
-2.0 4.0
2.0
oo "o
^ CO
)<
-2.0
-4.0 10 20 30 40 50 60 70 80 90 100 Sample Number

FIG. 2Illustration showing computed intervals based on sampling experiments; 100 samples of n = 4 observations each, from a Normal universe having \i=0 and o = case A is taken from Fig. 8 of Ref. 2, and Case B gives corresponding intervals for limits Jf 1.18 s, based on P = 0.90.
5. Experimental Illustration
Figure 2 gives two diagrams illustrating the results of sampling experiments for samples of ?i = 4 observations each drawn from a Normal population, for values of Case A, P = 0.50, and Case B, P = 0.90. For Case A, the intervals for 51 out of 100 samples included p, and for Case B, 90 out of 100 included yi. If, in each instance (that is, for each sample) we had concluded that the population mean |J. is included within the limits shown for Case A, we would have been correct 51 times and in error 49 times, which is a reasonable variation from the expectancy of being correct 50 percent of the time. In this experiment all samples were taken from the same population. However, the same reasoning applies to a series of samples each drawn from a population from the same universe as evidenced by conformance to the three conditions set forth in Section 4.
6. Presentation of Data
In presentation of data, if it is desired to give limits of this kind, it is quite important t h a t the probability associated with the limits be clearly indicated. The three values P = 0.90, P = 0.95, and P = 0.99 given in Table 2 (chances of 9 in 10, 95 in 100, and 99 in 100) are arbitrary choices that may be found convenient in practice.
Example Given a sample of 10 observations of breaking strength of hard-drawn copper wire as in Table 1, for which X = 573.2 lb s = 4.83 lb Using this sample to define limits of uncertainty based on P = 0.95 (Table 2), = we have
48 DATA AND CONTROL CHART ANALYSIS X 0.715^ = 573.2 3.5 = 569.7 and 576.7 Two pieces of information are needed to supplement this numerical result: (a) the fact t h a t 95 in 100 limits were used, and (6) that this result is based solely on the evidence contained in 10 observations. Hence, in the presentation of such limits, it is desirable to give the results in some such way as the following 573.2 3.5 lb ( P = 0.95, = 10) The essential information contained in the data is, of course, covered by presenting X, s, and n (see PART 1 of this Manual) and the limits under discussion could be derived directly therefrom. If it is desired to present such limits in addition to X, s, and n, the tabular arrangement given in Table 3 is suggested. A satisfactory alternative is to give the value in the column designated "Average, X " and to add a note giving the significance of this entry, as shown in Table 4. If one omits the note, it will be assumed that that P = 68%. a = was used and For a confidence coefficient of 0.95 use the a hsted in Table 2 under P = 0.90. For a confidence coefficient of 0.975 use the a listed in Table 2 under P = 0.95. For a confidence coefficient of 0.995 use the a listed in Table 2 under P = 0.99. In general, for a confidence coefficient of Pi use the a derived from Fig. 1 for P = 1 - 2(1 - Pi). For example, with TI = 10, X = 573.2, and s = 4.83 the one-sided upper Pi = 0.95 confidence limit would be to use a = 0.58 for P = 0.90 in Table 2, which yields 573.2-H 0.58(4.83) = 573.2 + 2.8 = 576.0.
7. One-Sided Limits
Sometimes we are interested in limits of uncertainty in one direction only. In this case we would present X + a s or X-as (not both), a one-sided confidence limit below or above which the population mean may be expected to lie in a stated proportion of an indefinitely large number of samples. The a to use in this one-sided case and the associated confidence coefficient would be obtained from Table 2 or Fig. 1 as follows.
TABLE 3. Suggested tabular arrangement. NUMBER OF TESTS, n 10 LIMITS FOR n (95 PERCENT CONFIDENCE LIMITS) 573.2 3.5 STANDARD DEVIATION, s 4.83
AVERAGE, X 573.2
TABLE 4. Alternative to Table 3. NUMBER OF TESTS, n 10 AVERAGE, X 573.2 ( 3.5) STANDARD DEVIATION, s 4.83
"The + entry indicates 95 percent confidence limits for |x.
level of measurement value constant
p-r true level
- I I r T
560
580
600
620
FIG. 3Showing how plus or minus limits (Li and L2) are unrelated to a systematic or constant en-or.
8. General C o m m e n t s o n t h e Use of Confidence Limits

In making use of limits of uncertainty of the type covered in this part, the engineer should keep in mind: (1) the restrictions as to (a) controlled conditions, (b) approximate Normality of population, (c) randomness of sample; and (2) the fact that the variability under consideration relates to fluctuations around the level of measurement values, whatever that may be, quite regardless of whether or not the population mean \i, of the measurement values is widely displaced from the true value, [IT , of the thing measured, as a result of the systematic or constant errors present throughout the measurements. For example, breaking strength values might center around a value of 575.0 lb (the population mean |i, of the measurement values) with a scatter of individual observations represented by the dotted distribution curve of Fig. 3, whereas the true average [ij. for the batch of wire under test is actually 610.0 lb, the difference between 575.0 and 610.0 representing a constant or systematic error present in all the observations as a result, say, of an incorrect adjustment of the testing machine. The limits thus have meaning for series of like measurements, made under
like conditions, including the constant errors if any be present.
same
In the practical use of these limits, the engineer may not have assurance that conditions (a), (b), and (c) given in section 4 are met, hence it is not advisable to lay great emphasis on the exact magnitudes of the probabilities given in Table 2, but rather to consider them as orders of magnitude to be used as general guides.
9. N u m b e r of P l a c e s to be R e t a i n e d in Computation a n d Presentation
The following working rule is recommended in carrying out computations incident to determining averages, standard deviations, and "limits for averages" of the kind here considered, for a sample of n observed values of a variable quantity: "In all operations on the sample of n observed values, such as adding, subtracting, multiplying, dividing, squaring, extracting square root, etc., retain the equivalent of two more places of figures than in the single observed values. For example, if observed values are read or determined to the nearest 1 lb., carry numbers to the nearest 0.01 lb. in the computations; if observed values are read or determined to the nearest 10

lb., carry numbers to the nearest 0.1 lb. in the computations, etc." Deleting places of figures should be done after computations are completed, in order to keep the final results substantially free from computation errors. In deleting places of figures the actual rounding-off procedure should be carried out as follows.^ 39 5(50) becomes 39 600. The number of places of figures to be retained in presentation depends on what use is to be made of the results and on the sampling variation present. No general rule, therefore, can safely be laid down. The following working rule has, however, been found generally satisfactory by the committee in presenting the results of testing in technical investigations and development work:
1. When the figure next beyond the last figure or place to be retained is less t h a n 5, the figure in the last place retained should be kept unchanged. 2. When the figure next beyond the last figure or place to be retained is more than 5, the figure in the last place retained should be increased by 1. 3. When the figure next beyond the last figure or place to be retained is 5, and (a) there are no figures, or only zeros, beyond this 5, if the figure in the last place to be retained is odd, it should be increased by 1; if even, it should be kept unchanged; but (b) if the 5 next beyond the figure in the last place to be retained is followed by any figures other than zero, the figure in the last place retained should be increased by 1, whether odd or even. For example, if in the following numbers, the places of figures in parenthesis are to be rejected 39 4(49) becomes 39 400, 39 4(50) becomes 39 400, 39 4(51) becomes 39 500, and
(a) See TABLE 5 for averages. (b) For standard deviations, three places of figures. retain
(c) If "limits for averages" of the kind here considered are presented, retain the same places of figures as are retained for the average. For example, ii n = 10, and if observed values were obtained to the nearest 1 lb, present averages and "limits for averages" to the nearest 0.1 lb, and present the standard deviation to three places of figures. This is illustrated in the tabular presentation in section 6. Rule (a) will result generally in one and conceivably in two doubtful places of figures in the average, that is, places which may have been affected by the rounding-off (or observation) of the n individual values to the nearest number of units stated in the first column of the table. Referring to Tables 3 and 4 the third place figures in the average, X = 573.2, corresponding to the first place of figures in the 3.5 value is doubtful in this sense. One might conclude that it would be suitable to present the average to the nearest pound, thus,
^ This rounding-off procedure agrees with t h a t adopted in ASTM Recommended Practice for Indicating which Places of Figures Are to be Considered Significant in Specified Limiting Values (E 29).
573 3 lb ( F = 0.95,
n^lO)
TABLE 5. Averages. WHEN THE SINGLE VALUES ARE OBTAINED TO THE NEAREST 0.1, 1, 10, etc., units 0.2, 2, 20, etc., units 0.5, 5, 50, etc., units RETAIN THE FOLLOWING NUMBER OF PLACES OF FIGURES IN THE AVERAGE AND WHEN THE NUMBER OF OBSERVED VALUES IS 2 to 20 4 to 40 10 to 100 1 more place t h a n in single values 21 to 200 41 to 400 101 to 1000 2 more places than in single values
less t h a n 4 less t h a n 10
[same number of places as in ^ [single values
TABLE 6. Effect of rounding. NOT ROUNDED LIMITS 573.5 1 . 4 573.5 1 . 5 572.1 572.0 574.9 575.0 DIFFERENCE 2.8 3.0 574 1 574 2 LIMITS 573 572 575 576 ROUNDED DIFFERENCE 2 4
SUPPLEMENT A This might be satisfactory for some purposes. However, the effect of such rounding off to the first place of figures of the plus or minus value may be quite pronounced if the first digit of the plus or minus value is small, as indicated in Table 6. If further use were to be made of these datacollecting additional observations to be combined with these, gathering other data to be compared with these, etc. then the effect of such rounding off of X in a presentation might seriously interfere with proper subsequent use of the information. The number of places of figures to be retained or to be used as a basis for action in specific cases cannot readily be made subject to any general rule. It is, therefore, recommended that in such cases the number of places be settled by definite agreements between the individuals or parties involved. In reports covering the acceptance and rejection of material ASTM E 29 gives specific rules that are applicable when reference is made to this recommended practice. P r e s e n t i n g P l u s or Minus Limits of U n c e r t a i n t y forCTNormal Distribution^ When observations Xi, X2, ..., X^ are made under controlled conditions, and there is reason to believe the distribution of X is Normal, two-sided confidence limits for the standard deviation of the population with confidence coefficient, P, will be given by lower confidence limit for
= sp-l)/xl
upper confidence limit for
P)I2
(1)
'^4(n-\)lxl
! + /)/2
where the quantity xli-p)i2 (o^^ Xl\+p)i2) is the x^ value of a chi-square variable with n
The analysis is strictly valid only for an unlimited population such as presented by a manufacturing or measurement process. When the population sampled is relatively small compared with the sample size n, the reader is advised to consult a statistician.

- 1 degrees of freedom which is exceeded with probabihty (1 - P)/2 or (1 + P)/2 as found in most statistics text books. To facihtate gives values of computation, Table 7 and for the upper confidence limit û=bu^ (3)
^ = y ( - l)/-^(i_P)/2
and
(2)
Example Table 1 of PART 2 gives the standard deviation of a sample of 10 observations of breaking strength of copper wire as s = 4.83 lb. If we assume that the breaking strength has a Normal distribution, which may actually be somewhat questionable, we have as 0.95 confidence limits for the universe standard deviation a that yield a
for P = 0.90, 0.95, and 0.99. Thus we have, for a Normal distribution, the estimate of the lower confidence limit for o as
TABLE 7. 6-factors for calculating confidence limits for a, Normal distribution."

Number of Observations in Sample, n 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 For larger n and
1 90% Confidence Limits bL bu
95% Confidence Limits bv bL 0.446 0.521 0.566 0.600 0.625 0.645 0.661 0.676 0.688 0.698 0.709 0.718 0.725 0.732 0.739 0.745 0.750 0.756 0.760 0.765 0.769 0.773 0.777 0.780 0.785 0.788 0.791 0.793 0.797 0.799 31.9 6.29 3.73 2.87 2.45 2.20 2.04 1.92 1.83 1.75 1.70 1.65 1.61 1.58 1.55 1.52 1.50 1.48 1.46 1.44 1.43 1.41 1.40 1.39 1.38 1.37 1.36 1.35 1.35 1.34
1 99% Confidence Limits bv (-L
0.510 0.578 0.619 0.649 0.671 0.690 0.705 0.718 0.730 0.739 0.747 0.756 0.762 0.769 0.775 0.780 0.785 0.789 0.794 0.798 0.801 0.806 0.808 0.812 0.814 0.818 0.821 0.823 0.826 0.828
16.0 4.41 2.92 2.37 2.09 1.91 1.80 1.71 1.64 1.59 1.55 1.51 1.49 1.46 1.44 1.42 1.40 1.38 1.37 1.35 1.35 1.34 1.33 1.32 1.31 1.30 1.29 1.29 1.28 1.27
0.356 0.434 0.484 0.518 0.547 0.569 0.587 0.603 0.618 0.630 0.641 0.651 0.660 0.669 0.676 0.683 0.690 0.696 0.702 0.707 0.712 0.717 0.721 0,725 0.730 0.734 0.738 0,741 0,745 0,747
159.5 14.1 6.47 4.40 3,48 2,98 2,66 2,44 2,28 2,15 2,06 1,98 1,91 1.85 1.81 1,76 1,73 1,70 1,67 1,64 1,62 1,60 1,58 1,56 1.54 1.52 1.51 1,50 1,49 1.47
1645 /Jli^ 1/(1 + 1

1/(1-1IMS /Jzi^)
[Smijln) 1/(1 + 1 1/(1-1[960/.j2n)
l/(l + 2 5 7 6 / ^ ) 1/(1-2.576/.,^
Confidence limits for a = bi^s and b^s.
53 LIMITS OF UNCERTAINTY lower 0.95 confidence limit of d i = 0.688^4.83; = 3.32 lb. and upper 0.95 confidence limit of Ojj = 1.83^4.83; = 8.83 lb. When there is a fraction p of a given category, for example, the fraction nonconforming, in n observations obtained under controlled conditions, 95 percent confidence limits for the population fraction p' may be found from the chart in Fig. 41 of Biometrika Tables for Statisticians, Vol. 1. A reproduction of this fraction is entered on the abscissa and the upper and lower 0.95 confidence limits are chart for selected sample sizes is shown in Fig. 4. To use the chart, the sample read on the vertical scale for various values of n. Approximate limits for values of n not shown on the Biometrika chart may be obtained by graphical interpolation. The Biometrika Tables for Statisticians also give a chart for 0.99 confidence limits. In general, for an np and n(l - p) at least 6 and preferably 0.10 < p < 0.90, the following formulas can be applied approximate 0.90 confidence limits pL645.^p{l-p)/n approximate 0.95 confidence limits pL960yjp{l-p)/n
Û Û(0.90)'^
SUPPLEMENT B
P r e s e n t i n g P l u s or Minus Limits of U n c e r t a i n t y for />'*
If we wish a one-sided confidence limit on the low side with confidence coefficient P, we estimate the lower one-sided confidence limit as
G^=s^{n-\)lxl.
For a one-sided confidence limit on the high side with confidence coefficient P, we estimate the upper one-sided confidence limit as
^4in-^)lxl
Thus for P = 0.95, 0.975, and 0.995 we use the 6L or b^ factor from Table 7 in the columns headed 0.90, 0.95, and 0.99, respectively. For example, a 0.95 upper one-sided confidence limit for o based on a sample of 10 items for which s = 4.83 will be
(4)
= 1.64(4.83) = 7.92 A lower 0.95 one-sided confidence limit would be a, =bL(0.90)'' = 0.730(4.83) = 3.53
approximate 0.99 confidence limits p 2.576^ p{\-p)/n.
Example Refer to the data of Table 2(a) of PART 1 and Fig. 4 of PART 1 and suppose that the lower specification limit on transverse
The analysis is strictly valid only for an unlimited population such as presented by a manufacturing or measurement process. When the population sampled is relatively small compared with the sample size n, the reader is advised to consult a statistician.
0-70
0-65
0-60
0-55
0-50
0-45
0-40
0-35
0-30
0-25
0-20
0-15
0-10
0-05
0-00 0-02 0-04 0-06 0-08 0-10 0-12 0-14 0-16 0-18 0-20 0-22 0-24 0-26 0-28 0-30 P Reproduced by permission of the Biometrika Trust.
FIG. 4Chart providing confidence iimits for p' in binomiai sampling, given a sample fraction. Confidence coefficient = 0.95. The numbers printed along the curves indicate the sample size n. If for a given value of the abscissa, p^ and Pg are the ordinates read from (or interpolated between) the appropriate lower and upper curves, then Pr{pA < p' < PB} > 0.95.
55 LIMITS OF UNCERTAINTY strength is 675 psi and there is no upper specification Hmit. Then the sample percentage of bricks nonconforming (the sample "fraction nonconforming" p) is seen to be 8/270 = 0.030. Rough 0.95 confidence limits for the universe fraction nonconforming p' are read from Fig. 4 as 0.02 to 0.07. Using Eq 4, we have approximate 95 percent confidence limits as 0.030 1.960^0.030(1 - 0.030/270) 0.030 1.960(0.010) 0.05 0.01
REFERENCES
[1] Shewhart, W. A., "Probability as a Basis for Action," presented at the joint meeting of the American Mathematical Society and Section K of the A.A.A.S., 27 Dec. 1932. [2] Shewhart, W. A., Statistical Method from the Viewpoint of Quality Control, W. E, Deming, Ed., The Graduate School, Department of Agriculture, Washington, D.C., 1939. [3] Pearson, E. S., The Application of Statistical Methods to Industrial Standardization and Quality Control, B. S. 600-1935, British Standards Institution, London, Nov. 1935. [4] Snedecor, G. W. and Cochran, W. G., Statistical Methods, 7th ed., Iowa State University Press, Ames, LA, 1980, pp. 54-56. [5] Ott, E. R., Schilling, E. G.,. and Neubauer, D. v.. Process Quality Control, 3rd ed., McGraw-Hill, New York, N.Y., 2000.
Even though p > 0.10, the two results agree reasonably well. One-sided confidence limits for a population fraction p' can be obtained directly from the Biometrika chart or Fig. 4, but the confidence coefficient will be 0.975 instead of 0.95 as in the two-sided case. For example, with n = 200 and the sample p = 0.10 the 0.975 upper one-sided confidence limit is read from Fig. 4 to be 0.15. When the Normal approximation can be used, we will have the following approximate one-sided confidence limits forp' P = n qo- lower limit =p l.2S2ylp(l-p)/n
upper limit =p + l.282.yjp{l-p)/n lower limit -p upper limit =p + 1.645^p{l-p)/n 1.645^p(l-p)/n
P = 0.95:
. lower limit =/? 2326-^p{[-p)/n P = 0.99: upper limit =p + 2.326-yJp{l-p)/n
MNL7A-EB/Feb. 2002
Control Chart Method of Analysis and Presentation of Data

GLOSSARY OF TERMS AND SYMBOLS USED IN PART 3
In general, the terms and symbols used in PART 3 have the same meanings as in preceding parts of the Manual. In a few cases, which are indicated in the following glossary, a more specific meaning is attached to them for the convenience of a portion or all of PART 3. Mathematical definitions and derivations are given in S u p p l e m e n t A. material, taken from a larger collection of units or quantity of material, which serves to provide information that can be used as a basis for action on the larger quantity or on the production process. May be referred to as a subgroup in the construction of a control chart. s u b g r o u p , n.one of a series of groups of observations obtained by subdividing a larger group of observations; alternatively, the data obtained from one of a series of samples taken from a series of lots or from sublots taken from a process. One of the essential features of the control chart method is to break up the inspection data into rational subgroups, that is, to classify the observed values into subgroups, within which variations may, for engineering reasons, be considered to be due to nonassignable chance causes only, but between which there may be differences due to one or more assignable causes whose presence is considered possible. May be referred to as a sample from the process in the construction of a control chart. unit, n.one of a number of similar articles, parts, specimens, lengths, areas, etc. of a material or product. sublot, n.an identifiable part of a lot.
Glossary of Terms
assignable cause, n.an identifiable factor that contributes to variation in quality and which it is feasible to detect and identify. Sometimes referred to as a special cause. c h a n c e cause, n.an identifiable factor that exhibits variation which is random and free from any recognizable pattern over time. Sometimes referred to as a common cause. lot, n.a definite quantity of some commodity produced under conditions that are considered uniform for sampling purposes. sample, n.a group of units, or portion of
Glossary of Symbols
Symbol General In P A R T 3, Control Charts number of nonconformities; more specifically t h e n u m b e r of nonconformities in a sample (subgroup)
Copyright
2002 by A S T M Inteniational
www.astin.org
57 CONTROL CHART METHOD
Glossary of SymbolsContinued
Symbol General In PART 3, Control Charts a factor that is a function of n and expresses the ratio between the expected value of J for a large number of samples of n observed values each and the a of the universe sampled. (Values of f. = E(I)la ^"^^ given in Tables 6 and 16, and in Table 49 in Supplement A, based on a Normal distribution.) a factor that is a function of n and expresses the ratio between expected value of R for a large number of samples of n observed values each and the a of the universe sampled. (Values of di = E(R)I o are given in Tables 6 and 16 and in Table 49 in Supplement A, based on a Normal distribution.) number of subgroups or samples under consideration the absolute value of the difference of two successive values plotted on a control chart.
d2
k MR
MR
typically the absolute value of the difference of two successive values plotted on a control chart. It may also be the range of a group of more than two successive values. average of n-1 moving ranges from a series of n values
average moving range of n-1 moving ranges from a series of n values
MR
number of observed values (observations) relative frequency or proportion, the ratio of the number of occurrences to the total possible number of occurrences
\X^-X^\ + --- + \X,-X
np R
number of occurrences range of a set of numbers, that is, the difference between the largest number and the smallest number sample standard deviation
subgroup or sample size, that is, the number of units or observed values in a sample or subgroup fraction nonconforming, the ratio of the number of nonconforming units (articles, parts, specimens, etc.) to the total number of units under consideration; more specifically, the fraction nonconforming of a sample (subgroup) number of nonconforming units; more specifically, the number of nonconforming units in a sample of n units range of the n observed values in a subgroup (sample) (the symbol R is also used to designate the moving range in Figures 29 and 30) standard deviation of the n observed values in a subgroup (sample)
\x^-'xf+ +(x-lcf
or expressed in a form more convenient but sometimes less accurate for computation purposes
7 ( x f + + xl) - {x,+ +xf u X

observed value of a measurable characteristic; specific observed values are designated X^, X2, X^, etc.; also used to designate a measurable characteristic average (arithmetic mean); the sum of the n observed values divided hyn n{n - 1) nonconformities per units, the number of nonconformities in a sample of n units divided by n
average of the n observed values in a subgroup (sample) Z +X + +X X=-
58 DATA AND CONTROL CHART ANALYSIS Qualified Symbols oaaa^

X,s,R,p,etc. |X, o, p', u', c'
^lo, Oo, Po, " o . Co,
;etc.
standard deviation of values oiX, s, R, p, etc. average of a set of values oi X,s, R, p, etc. (the overbar notation signifies an average value) mean, standard deviation, fraction nonconforming, etc., of the population
standard deviation of the sampling distribution of X, s, R, p, etc. average of the set of k subgroup (sample) values of X s, R, p, etc., under consideration; for samples of unequal size, an overall or weighted average
standard value of |x, 0, p', etc., adopted for computing control limits of a control chart for the case. Control with Respect to a Given Standard (see Sections 18 to 27) alpha risk of claiming that a hypothesis is true when it is actually true beta risk of claiming that a hypothesis is false when it is actually false risk of claiming that a process is out of statistical control when it is actually in statistical control, a.k.a.. Type I error. 100(l-a)% is the percent confidence. risk of claiming that a process is in statistical control when it is actually out of statistical control, a.k.a.. Type 11 error. 100(1-|3)% is the power of a test that declares the hypothesis is false when it is actually false.
GENERAL PRINCIPLES
for scientific evaluation of data and the determination of quality. The control chart method lays emphasis on the order or grouping of the observations in a set of individual observations, sample averages, number of nonconformities, etc., with respect to time, place, source, or any other consideration that provides a basis for a classification which may be of significance in terms of known conditions under which the observations were obtained. This concept of order is illustrated by the data in Table 1 in which the width in inches to the nearest 0.0001-in. is given for 60 specimens of Grade BB zinc which were used in ASTM atmospheric corrosion tests. At the left of the table, the data are tabulated without regard to relevant information. At the right, they are shown arranged in ten subgroups, where each subgroup relates to the specimens from a separate milling. The information regarding origin is relevant engineering
1. P u r p o s e PART 3 of the Manual gives formulas, tables, and examples that are useful in applying the control chart method [1] of analysis and presentation of data. This method requires that the data be obtained from several samples or that the data be capable of subdivision into subgroups based on relevant engineering information. Although the principles of PART 3 are applicable generally to many kinds of data, they will be discussed herein largely in terms of the quality of materials and manufactured products. The control chart method provides a criterion for detecting lack of statistical control. Lack of statistical control in data indicates that observed variations in quality are greater than should be attributed to chance. Freedom from indications of lack of control is necessary

TABLE 1. Comparison of data before and after subgrouping (width in inches of specimens of Grade BB zinc).
Before Subgrouping After Subgrouping Specimen 0.5005 0.5000 0.5008 0.5000 0.5005 0.5000 0.4998 0.4997 0.4998 0.4994 0.4999 0.4998 0.4995 0.4995 0.4995 0.4995 0.4995 0.4996 0.4998 0.5005 0.5005 0.5002 0.5003 0.5004 0.5000 0.5005 0.5008 0.5007 0.5008 0.5010 0.5008 0.5009 0.5010 0.5005 0.5006 0.5009 0.5000 0.5001 0.5002 0.4995 0.4996 0.4997 0.4993 0.4994 0.4999 0.4996 0.4996 0.4997 0.4995 0.4995 0.4997 0.4992 0.4995 0.4992 0.4994 0.4998 0.5000 0.4990 0.5000 0.5000 Subgroup
(Mining) 1 2 3 4 5 6 7 8 9 10
1
0.5005 0.4998 0.4995 0.4998 0.5000 0.5008 0.5000 0.4993 0.4995 0.4994
2
0.5000 0.4997 0.4995 0.5005 0.5005 0.5009 0.5001 0.4994 0.4995 0.4998
3
0.5008 0.4998 0.4995 0.5005 0.5008 0.5010 0.5002 0.4999 0.4997 0.5000
4
0.5000 0.4994 0.4995 0.5002 0.5007 0.5005 0.4995 0.4996 0.4995 0.4990
5
0.5005 0.4999 0.4995 0.5003 0.5008 0.5006 0.4996 0.4996 0.4995 0.5000
6
0.5000 0.4998 0.4996 0.5004 0.5010 0.5009 0.4997 0.4997 0.4992 0.5000
information, which makes it possible to apply the control chart method to these data (see Example 3).
2. T e r m i n o l o g y a n d Technical Background
Variation in quality from one unit of product to another is usually due to a very large number of causes. Those causes, for which it is possible to identify, are termed special causes, or assignable causes. Lack of control indicates one or more assignable causes are operative. The vast majority of causes of variation may be found to be inconsequential and cannot be identified. These are termed chance causes, or common causes. However, causes of large variations in quality generally admit of ready identification.
In more detail we may say t h a t for a constant system of chance causes, the average X, the standard deviations s, the value of fraction nonconforming p, or any other functions of the observations of a series of samples will exhibit statistical stability of the kind that may be expected in random samples from homogeneous material. The criterion of the quality control chart is derived from laws of chance variations for such samples, and failure to satisfy this criterion is taken as evidence of the presence of an operative assignable cause of variation. As applied by the manufacturer to inspection data, the control chart provides a basis for action. Continued use of the control chart and the elimination of assignable causes as their presence is disclosed, by failures to meet its criteria, tend to reduce variability and to stabilize quality at aimed-at levels [28,12]. While the control chart method has been devised primarily for this purpose, it

provides simple techniques and criteria that have been found useful in analyzing and interpreting other types of data as well.
4. B r e a k i n g u p D a t a into Rational Subgroups

One of the essential features of the control chart method is what is referred to as breaking up the data into rationally chosen subgroups called rational subgroups. This means classifying the observations under consideration into subgroups or samples, within which the variations may be considered on engineering grounds to be due to nonassignable chance causes only, but between which the differences may be due to assignable causes whose presence are suspected or considered possible. This part of the problem depends on technical knowledge and familiarity with the conditions under which the material sampled was produced and the conditions under which the data were taken. By identifying each sample with a time or a source, specific causes of trouble may be more readily traced and corrected, if advantageous and economical. Inspection and test records, giving observations in the order in which they were taken, provide directly a basis for subgrouping with respect to time. This is commonly advantageous in manufacture where it is important, from the standpoint of quality, to maintain the production cause system constant with time. It should always be remembered that analysis will be greatly facilitated if, when planning for the collection of data in the first place, care is taken to so select the samples that the data from each sample can properly be treated as a separate rational subgroup, and that the samples are identified in such a way as to make this possible.
3. Two Uses
The control chart method of analysis is used for the following two distinct purposes.
A. ControlNo Standard
Given
To discover whether observed values of X, s, p, etc., for several samples of n observations each, vary among themselves by an amount greater than should be attributed to chance. Control charts based entirely on the data from samples are used for detecting lack of constancy of the cause system.
B. Control Standard
with
Respect
to a
Given
To discover whether observed values of X, s, p, etc., for samples of n observations each, differ from standard values (XQ, OO, Po, etc., by an amount greater t h a n should be attributed to chance. The standard value may be an experience value based on representative prior data, or an economic value established on consideration of needs of service and cost of production, or a desired or aimed-at value designated by specification. It should be noted particularly t h a t the standard value of a, which is used not only for setting up control charts for s or i? but also for computing control limits on control charts for X, should almost invariably be an experience value based on representative prior data. Control charts based on such standards are used particularly in inspection to control processes and to maintain quality uniformly at the level desired.
5. General Technique in Using Control Chart Method

The general technique (see Ref. 1, Criterion I, Chapter XX) of the control chart method variations in quality generally admit of ready identification is as follows. Given a set of

observations, to determine whether an assignable cause of variation is present: a. Classify the total number of observations into k rational subgroups (samples) having n^, îj, ..., n,^ observations, respectively. Make subgroups of equal size, if practicable. It is usually preferable to make subgroups not smaller than n = A for variables X, s, or R, nor smaller t h a n n = 25 for (binary) attributes. (See Sections 13, 15, 23, and 25 for further discussion of preferred sample sizes and subgroup expectancies for general attributes.) For each statistic (X, s, R, p, etc.) to be used, construct a control chart with control limits in the manner indicated in the subsequent sections. If one or more of the observed values of X, s, R, p, etc., for the k subgroups (samples) fall outside the control limits, take this fact as an indication of the presence of an assignable cause. n. Note that this standard deviation is not the standard deviation computed from the subgroup values (of X, s, R, p, etc.) plotted on the chart, but is computed from the variations within the subgroups (see S u p p l e m e n t B, Note 1). Throughout this part of the Manual such standard deviations of the sampling distributions will be designated as a ^, Gg, GR, Cp, etc., and these symbols, which consist of a and a subscript, will be used only in this restricted sense. For measurement data, if [i and a were known, we would have Control limits for: average standard deviations ranges (expected z ) + 3 cr_ (expected s) 3cr, (expected R) 3CT^
b.
c.
where the various expected values are derived from estimates of (i or o. For attribute data, if p' were known, we would have control limits for values of p (expected p) 30p, where expected p = p'. The use of 3-sigma control limits can be attributed to Walter Shewhart who based this practice upon the evaluation of numerous datasets (see Ref. 1). Shewhart determined t h a t based on a single point relative to 3sigma control limits the control chart would signal assignable causes affecting the process. Use of 4-sigma control limits would not be sensitive enough, and use of 2-sigma control limits would produce too many false signals (too sensitive) based on the evaluation of a single point. Figure 1 indicates the features of a control chart for averages. The choice of the factor 3 (a multiple of the expected standard deviation of X, s, R, p, etc.) in these limits, as Shewhart suggested (Ref. 1), is an economic choice based on experience that covers a wide range of industrial applications of the control chart,
6, Control Limits and Criteria of Control

In both uses indicated in Section 3, the control chart consists essentially of symmetrical limits (control limits) placed above and below a central line. The central line in each case indicates the expected or average value of X, s, R, p, etc. for subgroups (samples) of n observations each. The control limits used here, referred to as 3-sigma control limits, are placed at a distance of three standard deviations from the central line. The standard deviation is defined as the standard deviation of the sampling distribution of the statistical measure in question {X, s, R, p, etc.) for subgroups (samples) of size

rather than on any exact value of probabiUty (see S u p p l e m e n t B, Note 2). This choice has proved satisfactory for use as a criterion for action, that is, for looking for assignable causes of variation.
Observed Values of X Upper Control Limlt-
2 4 6 8 Subgroup (Sample) Number
10
worthwhile to investigate points that fall outside what some have called a set of warning limits (often placed two standards deviation about the centerline). The chances that any single point would fall two standard deviations from the average is roughly 1 in 20, or 5% of the time, when the process is indeed centered and in statistical control. Thus, stopping to investigate a false alarm once for every 20 plotting points on a control chart would be too excessive. Alternatively, an effective rule of nonrandomness would be to take action if two consecutive points were beyond the warning limits on the same side of the centerline. The risk of such an action would only be roughly 1 in 800! Such an occurrence would be considered an unlikely event and indicate that the process is not in control, so justifiable action would be taken to identify an assignable cause. A control chart may be said to display a lack of control under a variety of circumstances, any of which provide some evidence of nonrandom behavior. Several of the best known nonrandom patterns can be detected by the manner in which one or more tests for nonrandomness are violated. The following list of such tests are given below: 1. Any single point beyond 3o limits 2. Two consecutive points beyond 2a limits on the same side of the centerline Eight points in a row on one side of the centerline Six points in a row that are moving away or toward the centerline with no change in direction (a.k.a., trend rule) Fourteen consecutive points alternating up and down (sawtooth pattern) Two out of three points beyond 2o limits on the same side of the centerline
FIG. 1Essential features of a control chart presentation; chart for averages.
This action is presumed to occur in the normal work setting, where the cost of too frequent false alarms would be uneconomic. Furthermore, the situation of too frequent false alarms could lead to a rejection of the control chart as a tool if such deviations on the chart are of no practical or engineering significance. In such a case, the control limits should be re-evaluated to determine if they correctly reflect the system of chance, or common, cause variation of the process. For example, a control chart on a raw material assay may have understated control limits if the data on which they were based encompassed only a single lot of the raw material. Some lot-to-lot raw material variation would be expected since nature is in control of the assay of the material as it is being mined. Of course, in some cases, some compensation by the supplier may be possible to correct problems with particle size and the chemical composition of the material in order to comply with the customer's specification. In exploratory research, or in the early phases of a deliberate investigation into potential improvements, it may be
3.
4.
5.
6.

7. Four out of five points beyond l a limits on the same side of the centerUne Fifteen points in a row within the l o Umits on either side of the centerline (a.k.a., stratification rulesampling from two sources within a subgroup) Eight consecutive points outside the l a limits on both sides of the centerline (a.k.a., mixture rule sampling from two sources between subgroups) exists unless the plotted points for at least 25 consecutive subgroups fall within 3-sigma control limits. On the other hand, the number of subgroups needed to detect a lack of statistical control at the start may be as small as 4 or 5. Such a precaution against overlooking trouble may increase the risk of a false indication of lack of control. But it is a risk almost always worth taking in order to detect trouble early. What does this mean? If the objective of a control chart is to detect a process change, and that we want to know how to improve the process, then it would be desirable to assume a larger alpha (a) risk (smaller beta (P) risk) by using control limits smaller t h a n 3 standard deviations from the centerline. This would imply t h a t there would be more false signals of a process change if the process were actually in control. Conversely, if the alpha risk is too small, by using control limits larger than 2 standard deviations from the centerline, then we may not be able to detect a process change when it occurs which results in a larger beta ((3) risk. Typically, in a process improvement effort, it is desirable to consider a larger alpha risk with a smaller beta risk. However, if the primary objective is to control the process with a minimum of false alarms, then it would be desirable to have a smaller alpha risk with a larger beta risk. The latter situation is preferable if the user is concerned about the occurrence of too many false alarms, and is confident t h a t the control chart limits are the best approximation of chance cause variation. Once statistical control of the process has been established, occurrence of one plotted point beyond 3-sigma limits in 35 consecutive subgroups or 2 points in 100 subgroups need not be considered a cause for action.
8.
9.
There are other rules that can be applied to a control chart in order to detect nonrandomness, but those given here are the most common rules in practice. It is also important to understand what risks are involved when implementing control charts on a process. If we state that the process is in a state of statistical control, and present it as a hypothesis, then we can consider what risks are operative in any process investigation. In particular, there are two types of risk that can be seen in the following table:
DECISION ABOUT THE STATE OF THE PROCESS BASED ON DATA TRUE STATE OF THE PROCESS
Process is IN control No error is made Alpha (a) risk, or Type I error
Process is OUT of control Beta (P) risk, or Type II error No error is made
Process is IN control Process is OUT of control
For a set of data analyzed by the control chart method, when may a state of control be assumed to exist? Assuming subgrouping based on time, it is usually not safe to assume that a state of control
NOTE In a number of examples in PART 3,

fewer than 25 points are plotted. In most of these examples, evidence of a lack of control is found. In others, it is considered only that the charts fail to show such evidence, and it is not safe to assume a state of statistical control exists. A. Large Samples of Equal Size For samples of size n, the control chart lines are as shown in Table 2.
TABLE 2. Formulas for control chart lines. ^ CENTRAL CONTROL LINE LIMITS For averages X "Y+I ^ V-0.5 For standard deviations s J
ay
(2>
CONTROL NO STANDARD GIVEN 7. Introduction

Sections 7 to 17 cover the technique of analysis for control when no standard is given, as noted under A in Section 3. Here standard values of î, a, p', etc., are not given, hence values derived from the numerical observations are used in arriving at central lines and control limits. This is the situation that exists when the problem at hand is the analysis and presentation of a given set of experimental data. This situation is also met in the initial stages of a program using the control chart method for controlling quality during production. Available information regarding the quality level and variability resides in the data to be analyzed and the central lines and control limits are based on values derived from those data. For a contrasting situation, see Section 18.
J2n~2.5
" Formula 1 for control limits is an approximation based on Eq 70, Supplement A. It may be used for n of 10 or more. * Formula 2 for control limits is an approximation based on Eq 75, Supplement A. It may be used for n of 10 or more.
where
X = the grand average of the observed values of X for a / / s a m p l e s ,
= ( X , +X^ + +Xt)/k (3)
J = the average subgroup standard deviation,

= (5, + S 2 + -
h)/k
(4)
where the subscripts 1, 2, . . ., ^ refer to the k subgroups, respectively, all of size n. (For a discussion of this formula, see Supplement B, Note 3; also see Example 1). B. Large Samples of Unequal Size Use Equations 1 and 2 but compute X and J as follows X = the grand average of the observed values of X for all samples
8. Control Charts for Averages, X, and for Standard Deviations, sLarge Samples
This section assumes that a set of observed values of a variable X can be subdivided into k rational subgroups (samples), each subgroup containing n of more than 25 observed values.
Previous editions of this manual had used n instead of n - 0.5 in Equation 1, and 2 (n - 1) instead of 2n - 2.5 in Equation 2 for control limits. Both formulas are approximations, but the present ones are better for n less than 50. Also, it is important to note that the lower control limit for the standard deviation chart is the maximum of J -3a and 0 since negative values have no meaning. This idea also applies to the lower control limits for attribute control charts.
CONTROL CHART METHOD

l X i + 2 ^ 2 + - - - + ^Xj 1 + 2 "I 1 î "
CENTRAL LINE
TABLE 3. Formulas for control chart lines."
(5)
CONTROL LIMITS FORMULA USING FACTORS IN TABLE 6 X+A,s ALTERNATE FORMULA (7)
grand total of X values divided by their total number
For averages X For standard
^ '1
'' jn - 0.5
s = the weighted standard deviation
BxS and B^s
deviations 5
?+ 3 , ' 4lr, - 15
(8)'
(6)
where the subscripts 1, 2, . . . , k refer to the k subgroups, respectively. (For a discussion of this formula, see S u p p l e m e n t B, Note 3.) Then compute control limits for each sample size separately, using the individual sample size, n, in the formula for control limits (see Example 2). When most of the samples are of approximately equal size, computing and plotting effort can be saved by the procedure given in S u p p l e m e n t B, Note 4.
Alternate Formula 7 is an approximation based on Eq 70, Supplement A. It may be used for of 10 or more. The values of A3 in the tables were computed from Eqs 42 and 57 in Supplement A. Alternate Formula 8 is an approximation based on Eq 75, Supplement A. It may be used for n of 10 or more. The values of B3 and B4 in the tables were computed from Eqs 42, 61, and 62 in Supplement A.
X = the grand average of observed values of X for all samples, k (9)
and Si, S2, etc., refer to the observed standard deviations for the first, second, etc., samples and factors C4, A3, B3, and B4 are given in Table 6. For a discussion of Eq 9, see S u p p l e m e n t B, Note 3; also see Example 3. B. Small Samples of Unequal Size
9. Control Charts for A v e r a g e s X, a n d for S t a n d a r d D e v i a t i o n s , sSmall S a m p l e s

This section assumes that a set of observed values of a variable X is subdivided into k rational subgroups (samples), each subgroup containing n=25 or fewer observed values.
A. Small Samples of Equal Size For samples of size n, the control chart lines are shown in Table 3. The centerlines for these control charts are defined as the overall average of the statistics being plotted, and can be expressed as
For small samples of unequal size, use Equations 7 and 8 (or corresponding factors) for computing control chart lines. Compute X by Eq 5. Obtain separate derived values of J for the different sample sizes by the following working rule: Compute a the overall average value of the observed ratio s/c^ for the individual samples; then compute J = câ for each sample size n. As shown in Example 4, the computation can be simplified by combining in separate groups all samples having the same sample size n. Control limits may then be determined separately for each sample size. These difficulties can be avoided by planning the collection of data so that the samples are made of equal size. The factor C4 is given in Table 6 (see Example 4).
10. Control Charts for A v e r a g e s , X, a n d for R a n g e s , RSmall Samples

This section assumes t h a t a set of observed values of a variable X is subdivided into k rational subgroups (samples), each subgroup containing n- 10 or fewer observed values. The range, R, of a sample is the difference between the largest observation and t h e smallest observation. When n = 10 or less, simplicity and economy of effort can be obtained by using control charts for X a n d R in place of control charts for X a n d s. The range is not recommended, however, for samples of more than 10 observations, since it becomes rapidly less effective than the standard deviation as a detector of assignable causes as n increases beyond this value. In some circumstances it may be found satisfactory to use the control chart for ranges for samples up to n = 15, as when data are plentiful or cheap. On occasion it may be desirable to use t h e chart for ranges for even larger samples: for this reason Table 6 gives factors for samples as large as n. = 25.
where X is the grand average of observed values of X for all samples, R is the average value of range R for the k individual samples (2?i + i?2 + . . + Rk)/ k (12)
and the factors ^2, ^2, D3 and D4 are given in Table 6, and d^ in Table 49 (See Example 5).
B. Small Samples of Unequal Size For small samples of unequal size, use Formulas 10 and 11 (or corresponding factors) for computing control chart lines. Compute X by Eq 5. Obtain separate derived values of R for the different sample sizes by the following working rule: compute 6, the overall average value of the observed ratio R/d2 for the individual samples. Then compute R = did for each sample size n. As shown in Example 6, t h e computation can be simplified by combining in separate groups all samples having the same sample size n. Control limits may then be determined separately for each sample size. These difficulties can be avoided by planning the collection of data so t h a t the samples are made of equal size.
A. Small Samples of Equal Size For samples of size n, the control chart lines are as shown in Table 4.
11. S u m m a r y , Control Charts for X, s, and RNo S t a n d a r d Given

The most useful formulas and equations from Sections 7 to 10, inclusive, are collected in Table 5 and are followed by Table 6, which gives t h e factors used in these and other formulas.
TABLE 4. Formulas for control chart lines. CONTROL LIMITS FORMULA USING FACTORS IN TABLE 6
CENTRAL LINE For Averages
ALTERNATE FORMULA
X3 X XA^R D,R and D,R
X
For Ranges R
^ d^^n
12. Control Charts for Attributes Data

(10)
R 3 ^ d.
(11)
Although in what follows the fraction p is designated fraction nonconforming, the methods described can be applied quite generally and p may in fact be used to represent the ratio of the number of items.
TABLE 5. Formulas for control chart lines." CENTRAL LINE Averages using s Averages using R Standard deviations Ranges X X s CONTROL LIMITS X Aji (s as given by Eq 9) YA^J(Ras given by Eq 12)
B^s and B^s (J as given by Eq 9) D^ ^ and D^ ^ (R as given by Eq 12)
"Controlno standard given (|i, a, not given)small samples of equal size.
TABLE 6. Factors for computing control chart linesno standard given. CHART FOR AVERAGES CHART FOR STANDARD DEVL^TIONS FACTORS FOR CENTRAL LINE
Ci
CHART FOR RANGES FACTORS FOR CENTRAL LINE
FACTORS FOR CONTROL LIMITS OBSERVATIONS IN SAMPLE, n 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Over 25 "3/7/1-0,5 ''(4/!-4)/(4/i-3) 'l-3/V2n-2.5 ''l + 3/V2n-2.5 A, 1.880 1.023 0.729 0.577 0.483 0.419 0.373 0.337 0.308 0.285 0.266 0.249 0.235 0.223 0.212 0.203 0.194 0.187 0.180 0.173 0.167 0.162 0.157 0.153 A, 2.659 1.954 1.628 1.427 1.287 1.182 1.099 1.032 0.975 0.927 0.886 0.850 0.817 0.789 0.763 0.739 0.718 0.698 0.680 0.663 0.647 0.633 0.619 0.606
a
FACTORS FOR CONTROL LIMITS B, 0 0 0 0 0.030 0.118 0.185 0.239 0.284 0.321 0.354 0.382 0.406 0.428 0.448 0.466 0.482 0.497 0.510 0.523 0.534 0.545 0.555 0.565
c
FACTORS FOR CONTROL LIMITS

^3
B, 3.267 2.568 2.266 2.089 1.970 1.882 1.815 1.761 L716 1.679 1.646 1.618 1.594 1.572 1.552 1.534 1.518 1.503 1.490 1.477 1.466 1.455 1.445 1.435
d
0^2
D4 3.267 2.575 2.282 2.114 2.004 1.924 1.864 1.816 1.777 1.744 1.717 1.693 1.672 1.653 1.637 1.622 1.609 1.596 1.585 1.575 1.565 1.557 1.548 1.541
0.7979 0.8862 0.9213 0.9400 0.9515 0.9594 0.9650 0.9693 0.9727 0.9754 0.9776 0.9794 0.9810 0.9823 0.9835 0.9845 0.9854 0.9862 0.9869 0.9876 0.9882 0.9887 0.9892 0.9896 6
1.128 1.693 2.059 2.326 2.534 2.704 2.847 2.970 3.078 3.173 3.258 3.336 3.407 3.472 3.532 3.588 3.640 3.689 3.735 3.778 3.819 3.858 3.895 3.931
0 0 0 0 0 0.076 0.136 0.184 0.223 0.256 0.283 0.307 0.328 0.347 0.363 0.378 0.391 0.404 0.415 0.425 0.435 0.443 0.452 0.459

occurrences, etc. t h a t possess some given attribute to the total number of items under consideration. The fraction nonconforming, p, is particularly useful in analyzing inspection and test results that are obtained on a "go/ no-go" basis (method of attributes). In addition, it is used in analyzing results of measurements t h a t are made on a scale and recorded (method of variables). In the latter case, p may be used to represent the fraction of the total number of measured values falling above any limit, below any limit, between any two limits, or outside any two limits. The fraction p is used widely to represent the fraction nonconforming, that is, the ratio of the number of nonconforming units (articles, parts, specimens, etc.) to the total number of units under consideration. The fraction nonconforming is used as a measure of quality with respect to a single quality characteristic or with respect to two or more quality characteristics treated collectively. In this connection it is important to distinguish between a nonconformity and a nonconforming unit. A nonconformity is a single instance of a failure to meet some requirement, such as a failure to comply with a particular requirement imposed on a unit of product with respect to a single quality characteristic. For example, a unit containing departures from requirements of the drawings and specifications with respect to (1) a particular dimension, (2) finish, and (3) absence of chamfer, contains three defects. The words "nonconforming unit" define a unit (article, part, specimen, etc.) containing one or more "nonconformities" with respect to the quality characteristic under consideration. When only a single quality characteristic is under consideration, or when only one nonconformity can occur on a unit, the number of nonconforming units in a sample will equal the number of nonconformities in that sample. However, it is suggested that under these circumstances the phrase "number of nonconforming units" be used rather than "number of nonconformities." Control charts for attributes are usually based either on counts of occurrences or on the average of such counts. This means that a series of attribute samples may be summarized in one of these two principal forms of a control chart and, though they differ in appearance, both will produce essentially the same evidence as to the state of statistical control. Usually it is not possible to construct a second type of control chart based on the same attribute data which gives evidence different from that of the first type of chart as to the state of statistical control in the way the X and s (or X and R) control charts do for variables. An exception may arise when, say, samples are composed of similar units in which various numbers of nonconformities may be found. If these numbers in individual units are recorded, then in principle it is possible to plot a second type of control chart reflecting variations in the number of nonuniformities from unit to unit within samples. Discussion of statistical methods for helping to judge whether this second type of chart gives different information on the state of statistical control is beyond the scope of this Manual. In control charts for attributes, as in s and R control charts for small samples, the lower control limit is often at or near zero. A point above the upper control limit on an attribute chart may lead to a costly search for cause. It is important, therefore, especially when small counts are likely to occur, that the calculation of the upper limit accounts adequately for the magnitude of chance variation that may be expected. Ordinarily there is little

to justify the use of a control chart for attributes if the occurrence of one or two nonconformities in a sample causes a point to fall above the upper control limit. expected np is four or more. When n is less than 25 or when the expected np is less t h a n 1, the control chart for p may not yield reliable information on the state of control. The average fraction nonconforming p is defined as _ _ total # of nonconforming units in all samples total # of units in all samples = fraction nonconforming in the complete set of test results. (13)
NOTE To avoid or minimize this problem of small counts, it is best if the expected or estimated number of occurrences in a sample is four or more. An attribute control chart is least useful when the expected number of occurrences in a sample is less than one.
NOTE The lower control limit based on the formulas given may result in a negative value that h a s no meaning. In such situations, the lower control limit is simply set at zero. It is important to note that a positive non-zero lower control limit offers the opportunity for a plotted point to fall below this limit when the process quality level significantly improves. Identifying the assignable cause(s) for such points will usually lead to opportunities for process and quality improvements.
A.
Samples of Equal
Size.
For a sample of size n, the control chart lines are as follows in Table 7 (see Example 7).
TABLE 7. Formulas for control chart lines. CENTRAL LINE For values ofp CONTROL LIMITS
^3. P{^-P)
(14)
When p is small, say less than 0.10, the factor 1 - p may be replaced by unity for most practical purposes, which gives control limits for p by the simple relation
13. C o n t r o l C h a r t f o r F r a c t i o n Nonconforming p This section assumes that the total number of units tested is subdivided into k rational subgroups (samples) consisting of Ml, n2,---> % units, respectively, for each of which a value ofp is computed. Ordinarily the control chart of p is most useful when the samples are large, say when n is 50 or more, and when the expected number of nonconforming units (or other occurrences of interest) per sample is four or more, that is, the p3J^ (14a)
B. Samples of Unequal Size Proceed as for samples of equal size but compute control limits for each sample size separately. When the data are in the form of a series of k subgroup values of p and the

TABLE 8. Formulas for control chart lines. CENTRAL LINE For values of rep np CONTROL LIMITS np3Jnp(\-p) (16)
corresponding sample sizes n, p may be computed conveniently by the relation
p = -^^
2^-^
^^-^
(15)
for which it is a convenient practical substitute when all samples have the same size n. It makes direct use of the number of nonconforming units np, in a sample (np - the fraction nonconforming times the sample size.) For samples of size n, the control chart lines are as shown in Table 8, where np - total number of nonconforming units in all samples/number of samples (17)
where the subscripts 1, 2,..., k refer to the k subgroups. When most of the samples are of approximately equal size, computation and plotting effort can be saved by the procedure in S u p p l e m e n t B, Note 4 (see Example 8).
NOTE If a sample point falls above the upper control limit for p when np is less than 4, the following check and adjustment method is recommended to reduce t h e incidence of misleading indications of a lack of control. If the non-integral remainder of the product of n and the upper control limit value for p is one-half or less, the indication of a lack of control stands. If t h a t remainder exceeds one-half, add one to the product and divide the sum by n to calculate an adjusted upper control limit for p. Check for an indication of lack of control in p against this adjusted limit (see Example 7 and Example 8).
= the average number of nonconforming units in the k individual samples, and p = the value given by Eq 13. When p is small, say less than 0.10, the factor l~p may be replaced by unity for most practical purposes, which gives control limits for np by the simple relation np 3 ^jnp (18)
or, in other words, it can be read as the avg. number of nonconforming units + 3.^/average number of nonconform ing units where "average number of nonconforming units" means the average number in samples of equal size (see Example 7). When the sample size, n, varies from sample to sample, the control chart for p (Section 13) is recommended in preference to the control chart for np; in this case, a
14, Control Chart for N u m b e r s of N o n c o n f o r m i n g U n i t s np

The control chart for np, number of conforming units in a sample of size n, is the equivalent of the control chart for p,
7 1 CONTROL CHART METHOD

graphical presentation of values of np does not give an easily understood picture, since the expected values, np, (central line on the chart) vary with n, and therefore the plotted values of np become more difficult to compare. The recommendations of Section 13 as to size of n and expected np in a sample apply also to control charts for the numbers of nonconforming units. When only a single quality characteristic is under consideration, and when only one nonconformity can occur on a unit, the word "nonconformity" can be substituted for the words "nonconforming unit" throughout the discussion of this section but this practice is not recommended. NOTE If a sample point falls above the upper control limit for np when np is less than 4, the following check and adjustment procedure is to be recommended to reduce the incidence of misleading indications of a lack of control. If the non-integral remainder of the upper control limit value for np is one-half or less, the indication of a lack of control stands. If that remainder exceeds one-half, add one to the upper control limit value for np to adjust it. Check for an indication of lack of control in np against this adjusted limit (see Example 7). sample of units, the numerical value of nonconformities per unit, u, is equal to the number of nonconformities in all the units in the sample divided by the number of units in the sample. The control chart for the nonconformities per unit in a sample u is convenient for a product composed of units for which inspection covers more than one characteristic, such as dimensions checked by gages, electrical and mechanical characteristics checked by tests, and visual nonconformities observed by the eye. Under these circumstances, several independent nonconformities may occur on one unit of product and a better measure of quality is obtained by making a count of all nonconformities observed and dividing by the number of units inspected to give a value of nonconformities per unit, rather than merely counting the number of nonconforming units to give a value of fraction nonconforming. This is particularly the case for complex assemblies where the occurrence of two or more nonconformities on a unit may be relatively frequent. However, only independent nonconformities are counted. Thus, if two nonconformities occur on one unit of product and the second is caused by the first, only the first is counted. The control chart for nonconformities per unit (more especially the chart for number of nonconformities, see Section 16) is a particularly convenient one to use when the number of possible nonconformities on a unit is indeterminate, as for physical defects (finish or surface irregularities, flaws, pinholes, etc.) on such products as textiles, wire, sheet materials, etc., which are not continuous or extensive. Here, the opportunity for nonconformities may be numerous though the chances of nonconformities occurring at any one spot may be small.
15. Control Chart for Nonconformities per Unit u

In inspection and testing, there are circumstances where it is possible for several nonconformities to occur on a single unit (article, part, specimen, unit length, unit area, etc.) of product, and it is desired to control the number of nonconformities per unit, rather than the fraction nonconforming. For any given

Table 9. Formulas for control chart lines. CENTRAL LINE CONTROL LIMITS
For values of u
u3fn V
(20)
This section assumes that the total number of units tested is subdivided into k rational subgroups (samples) consisting of rii, n2,. . ., n^ units, respectively, for each of which a value of u is computed. The control chart for u is most useful when the expected nu is 4 or more. When the expected nu is less than 1, the control chart for u may not yield reliable information on the state of control. The average nonconformities per unit, M, is defined as _ total # nonconformities in all samples u= total # units in all samples (19) = nonconformities per unit in the complete set of test results The simplified relations shown for control limits for nonconformities per unit assume that for each of the characteristics under consideration the ratio of the expected number of nonconformities to the possible number of nonconformities is small, say less than 0.10, an assumption that is commonly satisfied in quality control work. For an explanation of the nature of the distribution involved, see S u p p l e m e n t B, Note 5.
For samples of equal size, a chart for the number of nonconformities, c, is recommended, see Section 16. In the special case where each sample consists of only one unit, that is, = 1, then the chart for u (nonconformities per unit) is identical with that chart for c (number of nonconformities) and may be handled in accordance with Section 16. In this case the chart may be referred to either as a chart for nonconformities per unit or as a chart for number of nonconformities, but the latter designation is recommended (see Example 9).
B. Samples of Unequal Size Proceed as for samples of equal size but compute the control limits for each sample size separately. When the data are in the form of a series of subgroup values of u and the corresponding sample sizes, u may be computed by the relation
H=_J_J 2_2 UL (21)
where as before, the subscripts 1, 2, ..., k refer to the k subgroups. Note that ni, ^2, etc., need not be whole numbers. For example, if u represents nonconformities per 1000 ft of wire, samples of 4000 ft, 5280 ft, etc., then
A. Samples of Equal Size For samples of size n (n = number of units), the control chart lines are as shown in Table 9.
73 CONTROL CHART METHOD the corresponding values will be 4.0, 5.28, etc., units of 1000 ft. When most of the samples are of approximately equal size, computing and plotting effort can be saved by the procedure in S u p p l e m e n t B, Note 4 (see Example 10).
TABLE 10. Formulas for control chart lines. CENTRAL LINE For values of c F CONTROL LIMITS c 3 V? (22)
where
NOTE If a sample point falls above the upper limit for u where riuis less t h a n 4, the following check and adjustment procedure is recommended to reduce the incidence of misleading indications of a lack of control. If the nonintegral remainder of the product of n and the upper control limit value for u is one half or less, the indication of a lack of control stands. If that remainder exceeds one-half, add one to the product and divide the sum by n to calculate an adjusted upper control limit for u. Check for an indication of lack of control in u against this adjusted limit (see Examples 9 and 10).
_ total number of nonconfirmities in all samples c= number of samples = average number of nonconformities per sample. (23) The use of c is especially convenient when there is no natural unit of product, as for nonconformities over a surface or along a length, and where the problem is to determine uniformity of quality in equal lengths, areas, etc., of product (see Example 9 and Example 11).
B. Samples of Unequal Size For samples of unequal size, first compute the average nonconformities per unit M, by Eq 19; then compute the control limits for each sample size separately as shown in Table 11.
TABLE 11. Formulas for control chart lines. CENTRAL LINE For values of c CONTROL LIMITS nu 3 Jnu
(24)
16. Control Chart for N u m b e r of Nonconformities, c

The control chart for c, the number of nonconformities in a sample, is the equivalent of the control chart for u, for which it is a convenient practical substitute when all samples have the same size n (number of units).
A. Samples of Equal Size For samples of equal size, if the average number of nonconformities per sample is c, the control chart lines are as shown in Table 10.
The control chart for u is recommended as preferable to the control chart for c when the sample size varies from sample to sample for reasons stated in discussing the control charts for p and np. The recommendations of Section 15 as to expected c = nu also applies to control charts for numbers of nonconformities.

TABLE 12. Formulas for control chart lines.
C O N T R O I ^ N O STANDARD GIVENATTRIBUTES DATA CENTRAL LINE CONTROL LIMITS APPROXIMATION
Fraction nonconforming p
15.3 B b )
np 3 ^np{l~p)
P 1 3 , ^
Number of nonconforming units
np
np
np 3
Nonconformities per unit, u
u 3
if
Number of nonconformities c samples of equal size samples of unequal size
c 3 Vc"
nu 3 -Jnu
C O N T R O L W I T H R E S P E C T TO A GIVEN STANDARD NOTE 18. I n t r o d u c t i o n If a sample point falls above the upper control limit for c when nu is less than 4, the following check and adjustment procedure is to be recommended to reduce the incidence of misleading indications of a lack of control. If the non-integral remainder of the upper control limit for c is onehalf or less, the indication of a lack of control stands. If that remainder exceeds one-half, add one to the upper control limit value for c to adjust it. Check for an indication of lack of control in c against this adjusted limit (see Examples 9 and 11). Sections 18 to 27 cover the technique of analysis for control with respect to a given standard, as noted under (B) in Section 3. Here, standard values of \i, a, p,' etc., are given, and are those corresponding to a given standard distribution. These standard values, designated as Ho, Oo, Po etc., are used in calculating both central lines and control limits. (When only [ig is given and no prior data are available for establishing a value of Oo, analyze data from the first production period as in Sections 7 to 10, but use Ho for the central line.) Such standard values are usually based on a control chart analysis of previous data (for the details, see S u p p l e m e n t B, Note 6), but may be given on the basis described in Section 2>B. Note that these standard values are set up before the detailed analysis of the data at hand is undertaken and frequently before the data to be analyzed are collected. In addition to the standard values, only the information regarding sample size or sizes
17. S u m m a r y , C o n t r o l C h a r t s f o r p, np, u, a n d c N o S t a n d a r d Given The formulas of Sections 13 to 16, inclusive, are collected as shown in Table 12 for convenient reference.

is required in order to compute central lines and control limits. For example, the values to be used as central lines on the control charts are for averages, for standard deviations. for ranges. for values of p, etc..
^0
controlling X at the standard level |i<,that h a s been given. In other cases, interest centers on controlling the conformance to specified minimum and maximum limits within which material is considered acceptable, sometimes established without regard to the actual variation experienced in production. Such limits may prove unrealistic when data are accumulated and an estimate of the standard deviation, say o* of the process is obtained therefrom. If the natural spread of the process (a band having a width of 60*), is wider than the spread between the specified limits, it is necessary either to adjust the specified limits or to operate within a band narrower than the process capability. Conversely, if the spread of the process is narrower than the spread between the specified limits, the process will deliver a more uniform product than required. Note that in the latter event when only maximum and minimum limits are specified, the process can be operated at a level above or below the indicated midvalue without risking the production of significant amounts of unacceptable material.
cô^
^2^0
Po
where factors C and d2, which depend only 4 on the sample size, n, are given in Table 16, and defined in S u p p l e m e n t A. Note that control with respect to a given standard may be a more exacting requirement than control with no standard given, described in Sections 7 to 17. The data must exhibit not only control but control at a standard level and with no more than standard variability. Extending control limits obtained from a set of existing data into the future and using these limits as a basis for purposive control of quality during production, is equivalent to adopting, as standard, the values obtained from the existing data. Standard values so obtained may be tentative and subject to revision as more experience is accumulated (for details, see S u p p l e m e n t B, Note 6).
NOTE Two situations t h a t are not covered specifically within this section should be mentioned. 1. In some cases a standard value of \i is given as noted above, but no standard value is given for a. Here a is estimated from the analysis of the data at hand and the problem is essentially one of
19. Control Charts for Averages X and for S t a n d a r d Deviation, s

For samples of size n, the control chart lines are as shown in Table 13. For samples of n greater than 25, replace C4 by (4n - 4)/(4n - 3). See Examples 12 and 13; also see S u p p l e m e n t B, Note 9.

TABLE 13. Formulas for control chart lines.^
CONTROL LIMITS CENTRAL LINE For averages X \in FORMULA USING FACTORS IN TABLE 16 Ho ^ ^ 0 ALTERNATE FORMULA Cn (26)
For comments on the use of t h e control chart for ranges, see Section 10 (also see Example 16).
j"o3-r
For standard ^ ^ ^ "0 Alternate Formula 26 is an approximation based on Eq 74, Supplement A. It may be used for n of 10 or more. The values of Bs and Be given in the tables are computed from Eqs 42, 59, and 60 in Supplement A. ^ ^ ^ ^ ti (26)"
V2-1.5
21. S u m m a r y , Control Charts for X, s, a n d R S t a n d a r d G i v e n The most useful formulas from Sections 19 and 20 are summarized as shown in Table 15 and are followed by Table 16 which gives the factors used in these and other formulas.
For samples of n = 2 5 or less, use Table 16 for factors A, B^, and Be- Factors C4, A, Bs and B^ are defined in S u p p l e m e n t A. See Examples 14 and 15.
22. C o n t r o l C h a r t s f o r A t t r i b u t e s Data The definitions of terms and t h e discussions in Sections 12 to 16, inclusive, on the use of the fraction nonconforming, p, number of nonconforming units, np, nonconformities per unit, u, and number of nonconformities, c, as measures of quality are equally applicable to t h e sections which follow and will not be repeated here. It will suffice to discuss the central lines and control limits when standards are given.
20. C o n t r o l C h a r t f o r R a n g e s R The range, R, of a sample is the difference between the largest observation and the smallest observation. For samples of size n, the control chart lines are as shown in Table 14.

CONTROL LIMITS FORMULA USING FACTORS IN TABLE 16
CENTRAL LINE
ALTERNATE FORMULA dzf'o * '^s'ô (27)
23. Control Chart for F r a c t i o n Nonconforming, p Ordinarily, the control chart for p is most useful when samples are large, say, when n is 50 or more and when the expected number of nonconforming units (or other occurrences of interest) per sample is four or more, that is, the expected values of np is four or more. When n is less than 25 or the expected np is less than 1, t h e control chart for p may not yield reliable information on t h e state of control even with respect to a given standard.
For range R
dô^
DIOQ and DiOo
Use Table 16 for factors d2, Di, and D2. Factors d^, d^, Di, and D2 are defined in S u p p l e m e n t A.
^Previous editions of this manual had 2{n 1) instead of 2fi - 1.5 in Alternate Formula 26. Both formulas are approximations, b u t the present one is better for n less t h a n 50

TABLE 15. Formulas for control chart lines CONTROL WITH RESPECT TO A GIVEN STANDARD (Ho, O GIVEN) o CENTRAL LINE Average X Standard Deviations s Range R CONTROL LIMITS Be O and B^ O o o A a and D^ o
C4O0
TABLE 16. Factors for computing control chart linesstandard given. CHART FOR AVERAGES FACTORS FOR CONTROL LIMITS OBSERVATIONS IN SAMPLE, n 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Over 25 (4n-4)/(4n - 3 ) A 2.121 1.732 1.500 1.342 1.225 1.134 1.061 1.000 0.949 0.905 0.866 0.832 0.802 0.775 0.750 0.728 0.707 0.688 0.671 0.655 0.640 0.626 0.612 0.600 3/Vn' CHART FOR STANDARD DEVIATIONS FACTOR FOR CENTRAL LINE
C4
CHART FOR RANGES FACTOR FOR CENTRAL LINE da 1.128 1.693 2.059 2.326 2.534 2.704 2.847 2.970 3.078 3.173 3.258 3.336 3.407 3.472 3.532 3.588 3.640 3.689 3.735 3.778 3.819 3.858 3.895 3.931
FACTORS FOR CONTROL LIMITS B, 0 0 0 0 0.029 0.113 0.179 0.232 0.276 0.313 0.346 0.374 0.399 0.421 0.440 0.458 0.475 0.490 0.504 0.516 0.528 0.539 0.549 0.559 b Be 2.606 2.276 2.088 1.964 1.874 1.806 1.751 1.707 1.669 1.637 1.610 1.585 1.563 1.544 1.526 1.511 1.496 1.483 1.470 1.459 1.448 1.438 1.429 1.420 c
FACTORS FOR CONTROL LIMITS Dr 0 0 0 0 0 0.205 0.388 0.547 0.686 0.811 0.923 1.025 1.118 1.203 1.282 1.356 1.424 1.489 1.549 1.606 1.660 1.711 1.759 1.805 D2 3.686 4.358 4.698 4.918 5.079 5.204 5.307 5.393 5.469 5.535 5.594 5.647 5.696 5.740 5.782 5.820 5.856 5.889 5.921 5.951 5.979 6.006 6.032 6.056
0.7979 0.8862 0.9213 0.9400 0.9515 0.9594 0.9650 0.9693 0.9727 0.9754 0.9776 0.9794 0.9810 0.9823 0.9835 0.9845 0.9854 0.9862 0.9869 0.9876 0.9882 0.9887 0.9892 0.9896 a
''(4n-4)/(4-3)-3/.v/2n-2.5 '(4n-4)/ (4n-3)+3/^2n-2.5
See Supplement B, Note 9, on replacing first term in footnotes b and c by unity.

For samples of size n, where po is the standard value of/), the control chart lines are as shown in Table 17 (see Example IT).
TABLE 17. Formulas for control chart lines. CENTRAL LINE For values of p po TABLE 18. Formulas for control chart lines. CONTROL LIMITS CENTRAL LINE For values of np npo CONTROL LIMITS npf,3^npu{l-po) (29)
For samples of size n, where po is the standard value of p , the control chart lines are as shown in Table 18.
;,3j^(^-^)
V
(28)
When Po is small, say less than 0.10, the factor 1 - Po may be replaced by unity for most practical purposes, which gives the simple relation for computing the control limits for p as
When Po is small, say less than 0.10, the factor 1 - po may be replaced by unity for most practical purposes, which gives the simple relation for computing the control limits for np as
(28a) Vn For samples of unequal size, proceed as for samples of equal size but compute control limits for each sample size separately (see Example 18). When detailed inspection records are maintained, the control chart for p may be broken down into a number of component charts with advantage (see Example 19). See the N O T E at the end of Section 13 for possible adjustment of the upper control limit when npo is less than 4. (Substitute npo for np.) See Examples 17, 18, and 19 for applications.
np o3V "Po
(30)
As noted in Section 14, the control chart for p is recommended as preferable to the control chart for np when the sample size varies from sample to sample. The recommendations of Section 23 as to size of n and the expected np in a sample also apply to control charts for the number of nonconforming units. When only a single quality characteristic is under consideration, and when only one nonconformity can occur on a unit, the word "nonconformity" can be substituted for the words "nonconforming unit" throughout the discussion of this article, but this practice is not recommended. See the N O T E at the end of Section 14 for possible adjustment of the upper control limit when npo is less than 4. (Substitute npo for np.) See Examples 17 and 18.
24. Control Chart for Number of Nonconforming Units, np

The control chart for np, number of nonconforming units in a sample, is the equivalent of the control chart for fraction nonconforming, p , for which it is a convenient practical substitute, particularly when all samples have the same size, n. It makes direct use of the number of nonconforming units, np, in a sample (np = the product of the sample size and the fraction nonconforming). See Example 17.
25. Control Chart for Nonconformities per Unit, u

For samples of size n {n = number of units), where Uo is the standard value of u, the control chart lines are as shown in Table 19.
79 CONTROL CHART METHOD 26. Control Chart for Number of Nonconformities, c

The control chart for c, number of nonconformities in a sample, is the equivalent of the control chart for nonconformities per unit for which it is a convenient practical substitute when all samples have the same size, n (number of units). Here c is the number of nonconformities in a sample. If the standard value is expressed in terms of number of nonconformities per sample of some given size, that is, expressed merely as CQ, and the samples are all of the same given size (same number of product units, same area of opportunity for defects, same sample length of wire, etc.), then the control chart lines are as shown in Table 20.

CENTRAL LINE CONTROL LIMITS
For values of u
: 3,Mi-
(31)
See Examples 20 and 21. As noted in Section 15, the relations given here assume that for each of the characteristics under consideration, the ratio of the expected to the possible number of nonconformities is small, say less than 0.10. If u represents "nonconformities per 1000 ft of wire," a "unit" is 1000 ft of wire. Then if a series of samples of 4000 ft are involved, uo represents the standard or expected number of nonconformities per 1000 ft, and n. = 4. Note that n need not be a whole number, for if samples comprise 5280 ft of wire each, n - 5.28, that is, 5.28 units of 1000 ft (see Example 11). Where each sample consists of only one unit, that is n= 1, then the chart for u (nonconformities per unit) is identical with the chart for c (number of nonconformities) and may be handled in accordance with Section 26. In this case the chart may be referred to either as a chart for nonconformities per unit or as a chart for number of nonconformities, but the latter practice is recommended. Ordinarily, the control chart for u is most useful when the expected nu is 4 or more. When the expected nu is less than 1, the control chart for u may not yield reliable information on the state of control even with respect to a given standard. See the NOTE at the end of Section 15 for possible adjustment of the upper control limit when TIUQ is less than 4. (Substitute nuo for MU.^See Examples 20 and 21.
TABLE 20. Formulas for control chart lines (C given). CENTRAL LINE For number of nonconformities, c CONTROL LIMITS
c^+iyfc^
(32)
Use of C is especially convenient when o there is no natural unit of product, as for nonconformities over a surface or along a length, and where the problem of interest is to compare uniformity of quality in samples of the same size, no matter how constituted (see Example 21). When the sample size, n, (number of units) varies from sample to sample, and the standard value is expressed in terms of nonconformities per unit, the control chart lines are as shown in Table 21.
TABLE 21. Formulas for control chart lines (uo given). CENTRAL LINE For values of c nuo CONTROL LIMITS nUgtsJnu^ (33)
TABLE 22. Formulas for control chart lines. CONTROL WITH RESPECT TO A GIVEN STANDARD (po, npo, u OR Co GIVEN) CENTRAL LINE CONTROL LIMITS APPROXIMATION
Fraction nonconforming, p Number of nonconforming units, np Nonconformities per unit, u Number of nonconformities, c Samples of equal size CQ given) Samples of unequal size (UQ given)
Po 3 npo
Po{^-Po)
Po 3
nPo 3V"Po(l-/'o) Mo 3
Co 3 ^
MQ 3 -JnUfj
Under these circumstances the control chart for u (Section 25) is recommended in preference to the control chart for c, for reasons stated in Section 14 in the discussion of control charts for p and for np. The recommendations of Section 25 as to the expected c - nu also applies to control charts for nonconformities. See the NOTE at the end of Section 16 for possible adjustment of the upper control limit when nuo is less than 4. (Substitute C = nuo for nu). See Example o 21.
CONTROL CHARTS FOR INDIVIDUALS
Sections 28 to 30^ deal with control charts for individuals, in which individual observations are plotted one by one. This type of control chart has been found useful more particularly in process control when only one observation is obtained per lot or batch of material or at periodic intervals from a process. This situation often arises when: (a) sampling or testing is destructive, (6) costly chemical analyses or physical tests are involved, and (c) the material sampled at any one time (such as a batch) is normally quite homogeneous, as for a well-mixed fluid or aggregate. The purpose of such control charts is to discover whether the individual observed values differ from the expected
27. S u m m a r y , Control Charts for p, np, u, and cStandard Given

The formulas of Sections 22 to 26, inclusive, are collected as shown in Table 22 for convenient reference.
To be used with caution if the distribution of individual values is marlcedly asymmetrical.

value by an amount greater than should be attributed to chance. When there is some definite rational basis for grouping the batches or observations into rational subgroups, as, for example, four successive batches in a single shift, the method shown in Section 29 may be followed. In this case, the control chart for individuals is merely an adjunct to the more usual charts but will react more quickly to a sharp change in the process than the X chart. This may be important when a single batch represents a considerable sum of money. When there is no definite basis for grouping data, the control limits may be based on the variation between batches, as described in Section 30. A measure of this variation is obtained from moving ranges of two observations each (the absolute value of successive differences between individual observations that are arranged in chronological order). A control chart for moving ranges may be prepared as a companion to the chart for individuals, if desired, using the formulas of Section 30. It should be noted that adjacent moving ranges are correlated, as they have one observation in common. The methods of Sections 29 and 30 may be applied appropriately in some cases where more than one observation is obtained per lot or batch, as for example with very homogeneous batches of materials, for instance, chemical solutions, batches of thoroughly mixed bulk materials, etc., for which repeated measurements on a single batch show the within-batch variation (variation of quality within a batch and errors of measurement) to be very small as compared with between-batch variation. In such cases, the average of the several observations for a batch may be treated as an individual observation. However, this procedure should be used with great caution; the restrictive conditions just cited should be carefully noted. The control limits given are sigma control limits in all cases. three
29. Control Chart for Individuals, XUsing Rational S u b g r o u p s

Here the control chart for individuals is commonly used as an adjunct to the more usual X and s, or X and R, control charts. This can be useful, for example, when it is important to react immediately to a single point that may be out of statistical control, when the ability to localize the source of an individual point that has gone out of control is important, or when a rational subgroup consisting of more than two points is either impractical or nonsensical. Proceed exactly as in Sections 9 to 11 (controlno standard given) or Sections 19 to 21 (controlstandard given), whichever is applicable, and prepare control charts for X and s, or for X and R. In addition, prepare a control chart for individuals having the same central line as the X chart but compute the control limits as shown in Table 23. Table 26 gives values of E2 and Eg, for samples of ?i = 10 or less. Values that are more complete are given in Table 50, S u p p l e m e n t A for n through 25 (see Examples 22 and 23).
30. Control Chart for Individuals, XUsing Moving R a n g e s

A. No Standard Given
Here the control chart lines are computed from the observed data. In this section the symbol, MR, is used to signify the moving

TABLE 23. Formulas for control chart lines. CHART FOR INDIVIDUALSASSOCIATED WITH CHART FOR sORR HAVING SAMPLE SIZE n CONTROL LIMITS FORMULA USING NATURE OF CENTRAL FACTORS ALTERNATE DATA LINE jj^ FORMULA TABLE 26 NO STANDARD GIVEN
Samples of equal size

based on s based on R Samples of unequal size: o computed from observed values of s per Section 9 or from observed values of R per SectionlO(6) X STANDARD GIVEN Samples of equal or unequal size \^ Ho+3(7 (37) X3(T X X X + EjS X + E^R X3J/c, XiRld^ (34) (35)
(36)
"See Example 4 for determination of 6 based on values of i and Example 6 for determination of a based on values of R
TABLE 24. Formulas for control chart lines. CHART FOR INDIVIDUALSUSING MOVING RANGES NO STANDARD GIVEN CENTRAL LINE CONTROL LIMITS For individuals X X E,MR = X 2.66MR For moving ranges of R D^MR = 3.27MR two observations _ -TTK /^ D^MR =0 range. The control chart lines are shown in Table 24 where X= the average observations, of the as j5 Standard Given
(33) (39)
individual
When [i^ and Oo are given, the control chart lines are as shown in Table 25. See Example 25.
MR =
the mean moving range, (see S u p p l e m e n t B, Note 7 for more general discussion) the average of the absolute values of successive differences between pairs of the individual observations, and
EXAMPLES
n 2 for determining E2, D3 and D4. êe Example 24.
31, Illustrative Examples Control, No Standard Given

Examples 1 to 11, inclusive, illustrate the ^^^ ^f ^^^ control chart method of
TABLE 25. Formulas for control chart lines. CHAET FOR INDIVIDUALSSTANDARD GIVEN CENTRAL CONTROL LIMITS LINE Ho3oo For individuals ^0 For moving ranges of two observations D20o= 3.69oo DiOo = 0
(40) (41)
TABLE 26. Factors for computing control limits. CHART FOR INDIVIDUALSASSOCIATED WITH CHART FOR s OR i? HAVING SAMPLE SIZE n OBSERVATIONS IN SAMPLES OF EQUAL SIZE (FROM WHICH s" OR iJ HAS BEEN DETERMINED) Factors for control limits
3
10 3.760 2.659 3.385 1.772 3.256 1.457 3.192 1.290 3.153 1.184 3.127 1.109 3.109 1.054 3.095 1.010 3.084 0.975
Ei
TABLE 27. Operating characteristic, daily control data Sample

1 2 3 4 5 6 7 8 9 10 Total Average
Sample Size
50 50 50 50 50 50 50 50 50 50 500 50
Average
Standard Deviation
35.1 34.6 33.2 34.8 33.4 33.9 34.4 33.0 32.8 34.8 340.0 34.0
5.35 4.73 3.73 4.55 4.00 4.30 4.98 5.30 3.29 3.77 44.00 4.40

analyzing data for control, when no standard is given (see Sections 7 to 17).
ForX X3
I - 3 4 . 0 + 1.9, Vn-0.5 32.1 and 35.9 "+1
"V2-2.5
- 1 10 + 1 31
Example 1: Control Charts for X and s, Large Samples of Equal Size (Section 8A) A manufacturer wished to determine if his product exhibited a state of control. In this case, the central lines and control limits were based solely on the data. Table 27 gives observed values of X and s for daily samples of n. = 50 observations each for ten consecutive days. Figure 2 gives the control charts for X and s.
3.06 and 5.74
ResultsThe charts give no evidence of lack of control. Compare with Example 12, in which the same data are used to test product for control at a specified level.
Example 2: Control Charts for X and s, Large Samples of Unequal Size (Section 8B) To determine whether there existed any assignable causes of variation in quality for an important operating characteristic of a given product, the inspection results given in Table 28 were obtained from ten shipments whose samples were unequal in size; hence, control limits were computed separately for each sample size.
TABLE 28. Operating characteristic, mechanical part.
Shipment Sample Size n 50 50 100 25 25 50 100 50 50 50 550 55 Average
X
381IK
4>
? 34
30
^-^Jk|
1^ ^^"TBTT^^TT
r^
'
'
'
'
'
10
Standard Deviation s 4.35 4.03 2.43 3.56 3.10 3.30 4.18 4.30 2.09 2.67
S?ts
2 4 6 8 Sample Number
10
1 2 3 4 5 6 7 8 9 10 Total Weighted average
bb.l 54.6 52.6 55.0 53.4 55.2 53.3 52.3 53.7 54.3 Znjf =29590.0 53.8
FIG. 2Control charts for X and s. Large samples of equal size, n = 50; no standard given.
= 1864.50 3.39
Central Lines ForX:X = 34.0 For s: 1 = 4.40 Control Limits re = 50
Figure 3 gives the control charts for X and s.

TABLE 29. Width in inches, specimens of Grade BB zinc.
Measured Values Average, Standard Range, Deviation, E X s 0.50030 0.49973 0.49952 0.50028 0.50063 0.60078 0.49985 0.49958 0.49943 0.49970 0.49998 0.00035 0.0008 0.00018 0.0005 0.00004 0.0001 0.00026 0.0007 0.00035 0.0010 0.00019 0.0006 0.00029 0.0007 0.00021 0.0006 0.00020 0.0005 0.00041 0.0010 0.00025 0.00064
Set
X,
X,
X,
X,
X,
0.6005 0.4999 0.4995 0.5003 0.5008 0.5006 0.4996 0.4996 0.4995 0.6000
Ae
0.5000 0.4998 0.4996 0.5004 0.6010 0.5009 0.4997 0.4997 0.4992 0.5000
1 2 3 4 5 6 7 8 9 10
Average
0.6006 0.4998 0.4995 0.4998 0.5000 0.6008 0.5000 0.4993 0.4995 0.4994
0.5000 0.4997 0.4996 0.5005 0.6005 0.5009 0.5001 0.4994 0.4995 0.4998
10
Group 1 0.5008 0.5000 0.4998 0.4994 0.4995 0.4995 0.6006 0.5002 0.5008 0.6007 Group 2 0.5010 0.5005 0.5002 0.4995 0.4999 0.4996 0.4997 0.4992 0.6000 0.4990
Shipment Numt>r
FIG. 3Control charts for X and s. Large samples of unequal size, n = 25, 50,100; no standard given.
Central Lines F o r X ; f = 53.8 F o r i : s =3.39 Control Limits ForX: X3 = 53.8 10.17
Vn-0.5
VH-0.5
of sets of specimens consisting of six specimens each used in atmosphere exposure tests sponsored by ASTM. In each of the two groups, the five sets correspond to five different milHngs that were employed in the preparation of the specimens. Figure 4 shows control charts for X and s.
n =25:51.7 and 55.9 n =50:52.4 and 55.2 n =100: 52.8and54.8 For.s: s 2i , ^ = 3.39 + V2-2.5 \l2n-2.5 =25:1.91and4.87 =50:2.36 and 4.42 =100: 2.67 and 4.11
0.501 ? > e < 0.500 0.499
10
i | aoooe
I 0.0004 f 0.0002
ResultsLack of control is indicated with respect to both X and s. Corrective action is needed to reduce the variabihty between shipments.
>Kft>^
4 6 Sat Number 8 10
Example 3: Control Charts for X and s, Small Samples of Equal Size (Section 9A) Table 29 gives the width in inches to the nearest 0.0001-in. measured prior to exposure for ten sets of corrosion specimens of Grade BB zinc. These two groups of five sets each were selected for illustrative purposes from a large number
FIG. 4Control chart for X and s. Small samples of equal size, n = 6 no standard given.
ResultsThe chart for averages indicates the presence of assignable causes of variation in width, X, from set to set, that is, from milling to milling. The pattern of points for averages indicates a systematic pattern of width values for the five

millings, a factor that required recognition in the analysis of the corrosion test results.
a = - L f ^ : l L , i ^ ] = 0.902 2ll 0.9213 0.9400 j

Central Lines
Central Lines For Z: X = 0.49998 For s: s = 0.00025
For H: X = 71.65 For i: = 4; ? = C4a = (0.9213)(0.902) = 0.831 = 5: ? = 40 = (0.9400)(0.902) = 0.848
Control Limits n =6
ForX: XA,s =
0.49998 (l.287)(0.00025) 0.49966 and 0.50030 Fori : S,? = (1.970X0.00025) = 0.00049 B,s = (0.030X0.00025)= 0.00001
Control Limits ForX: = 4: X A^l = 71.65 (l.628)(0.83l), 73.0, and 703 n = 5: X + ^3 J = 71.65 (l.427)(0.848), 72.9, and 70.4 Fori: n = 4:5,5 = (2.266X0.831) = 1.88 B3i=(0X0.83l)=0 = 5 :B,i = (2.089X0.848) = 1.77 B35=(0X0.848) = 0
Example 4: Control Charts for X and s, Small Samples of Unequal Size (Section 9B) Table 30 gives interlaboratory calibration check data on 21 horizontal tension testing machines. The data represent tests on No. 16 wire. The procedure is similar to that given in Example 3, but indicates a suggested method of computation when the samples are not equal in size. Figure 5 gives control charts for X and s.
ResultsThe calibration levels of machines were not controlled at a common level; the averages of six machines are above and the averages of five machines are below the control limits. Likewise, there is an indication that the variability within machines is not in statistical control, since three machines, Numbers 6, 7, and 8, have standard deviations outside the control limits.
Example 5: Control Charts for X and R, Small Samples of Equal Size (Section lOA) Same data as in Example 3, Table 29. Use is made of control charts for averages and ranges rather than for averages and standard deviations. Figure 6 shows control charts for X and R.
10 IS Mochint Number
20
FIG. 5Control chart for X and s. Small samples of unequal size, n = 4 no standard given.

TABLE 30. Interlaboratory calibration, horizontal tension testing machines.
Number of Machine 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Total Tests 5 5 5 5 5 5 4 5 5 5 5 5 5 5 5 4 5 5 5 5 5 103 1 73 70 74 70 70 65 72 69 71 71 71 70 73 74 72 75 68 71 72 68 69 2 73 71 74 70 70 65 72 70 71 71 71 71 74 74 72 75 69 71 73 69 69 Test Value 3 73 71 74 70 70 66 74 71 71 71 72 71 74 75 72 75 69 72 73 70 69 4 75 71 74 72 70 69 76 73 71 71 72 72 75 75 73 76 69 72 73 71 69 5 75 72 75 73 70 70 73 72 72 72 72 75 75 73 70 73 73 71 69 Average Standard Deviation s
n=4
Range R
n=4 H=5
X
73.8 71.0 74.2 71.0 70.0 67.0 73.5 71.2 71.2 71.2 71.6 71.2 74.2 74.6 72.4 75.3 69.0 71.8 72.8 69.8 69.0
n=5
1.10 0.71 0.45 1.41 0 2.35 1.91 1.79 0.45 0.45 0.55 0.55 0.84 0.55 0.55 0.50 0.71 0.84 0.45 1.30 0 2.41 15.34 5 1 4
2 2 1 3 0 5 4 1 1 1 2 2 1 1 2 2 1 3 0 34
weighted average X == 71.65
0.501
IX i5 't
Control Limits
>
o"
0.500
0.499
: V--^-J ^ _
_x
^
' 1
n=6 For X XA,R )64) = 0.49998 (0.483)(0. 00( = 0.50029 and 0.4996'
For R -.DJR =(2.004)(0.00064) =
^ _ .
1
.x .i_
2 0.0020 a: 0.0015 0.0010 o 0.0005 IE 0 2
to
0.00128
) D,R = (0)(0.00064) = ( > ^ . X "^ >s / , , .
-_l
4 6 Set Number
10
Example 6: Control Charts for X and R, Small Samples of Unequal Size (Section lOB) Same data as in Example 4, Table 8. In the analysis and control charts, the range is used instead of the standard deviation. The procedure is similar to that given in Example 5, but indicates a suggested method of computation when samples are not equal in size. Figure 7 gives control charts for X and R. a is determined from the tabulated ranges given in Example 4, using a similar procedure to that given in Example 4 for
FIG. 6Control charts for X and R. Small Samples of equal size, n = 6; no standard given.
ResultsThe results are practically identical in all respects with those obtained by using averages and standard deviations, Fig. 4, Example 3.
Central Lines For Z: Z = 0.49998 For R:^= 0.00064
DATA AND CONTROL CHART ANALYSIS

standard deviations where samples are not equal in size, t h a t is 1 ^ - + - ^ ^ 1 = 0.812 21U.059 2.326 (2.059)(0.812)=L67
?i = 5: R = djO =
(2.326)(0.812)=L89 Control Limits For 'X:n = 4:X A.^ = 71.65 (0.729)(1.67) 70.4 and 72.9 n = 5: X A2^ = 7L65 (0.577)(1.89) 70.6 and 72.7 Fori? : = 4 iDÂ =(2.282)(L67) = 3.8 A ^ = (0)(1.67) = 0 = 5. D , ^ = (2.114X1.89) = 4.0 D3^ = (0)(1.89) = 0
G--
10 (5 Mochine Number
FIG. 7Control charts for X and R. Small samples of unequal size, n = 4, 5; no standard given.
Example 7: Control Charts for p, Samples of Equal Size (Section ISA) and np, Samples of Equal Size (Section 14) Table 31 gives the number of nonconforming units found in inspecting a series of 15 consecutive lots of galvanized washers for finish nonconformities such as exposed steel, rough galvanizing. The lots were nearly the same size and a constant sample size oi n - 400 were used. The fraction nonconforming for each sample was determined by dividing the number of nonconforming units found, np, by the
ResultsThe results are practically identical in all respects with those obtained by using averages and standard deviations. Fig. 5, Example 4. Central Lines For Z: Z = 7L65 For R. n = 4: R = djO =
TABLE 31. Finish defects, galvanized washers.

NUMBER OF FRACTION NONCONFORMING NONCONFORMING UNITS np P 1 3 0 7 2 0 1 0 0.0025 0.0075 0 0.0175 0.0050 0 0.0025 0 NUMBER OF FRACTION SAMPLE NONCONFORMING NONCONFORMING SIZE UNITS n np P 400 400 400 400 400 400 400 6000 8 5 2 0 1 0 3 33 0.0200 0.0125 0.0050 0 0.0025 0 0.0075 0.0825
LOT No. No. No. No. No. 1 2 3 4 5
SAMPLE SIZE n 400 400 400 400 400 400 400 400
LOT No. 9 No. 10 No. No. No. No. No. 11 12 13 14 15
No. 6 No. 7 No. 8
Total

p3,
~4-p)
sample size, n, and is listed in the table. Figure 8 gives the control chart for p, and Fig. 9 gives the control chart for np.
^ 10.0055(0.9945) 0.0055 3. 400 0.0055 + 0.0111= 0 and 0.0166
10
Lot Number
Fig. 8Control chart for p. Samples of equal size, n=400; no standard given.
Note that these two charts are identical except for the vertical scale. (a) Control chart for p Central Line 33 : 0.0055 ^ 6000 0.0825 : 0.0055 15 Control Limits = 400 TABLE 32. Surface defects, galvanized hardware.
SAMPLE SIZE, n 580 550 580 640 880 880 640 550 580 880 800 800 580 580 550 NUMBER OF NONCONFORMING UNITS np 9 7 3 9 13 14 14 10 12 14 6 12 7 11 5 FRACTION NONCONFORMING P 0.0155 0.0127 0.0052 0.0141 0.0148 0.0159 0.0219 0.0182 0.0207 0.0159 0.0075 0.0150 0.0121 0.0190 0.0091
Fig. 9Control chart for np. Samples of equal size, n=400; no standard given.
ResultsLack of control is indicated; points for lots numbers 4 and 9 are outside the control limits.
(b) Control chart for np
LOT No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8 No. 9 No. 10 No. 11 No. 12 No. 13 No. 14 No. 15
LOT No. 16 No.l7 No.18 No. 19 No.20 No.21 No.22 No.23 No.24 No. 25 No.26 No.27 No.28 No.29 No.30 No.31 Total
SAMPLE SIZE n 330 330 640 580 550 510 640 300 330 880 880 800 580 880 880 330 19 510
NtnVIBER OF NONCONFORMING UNITS np 4 2 4 7 9 7 12 8 5 18 7 8 8 15 3 5 268
FRACTION NONCONFORMING P 0.0121 0.0061 0.0063 0.0121 0.0164 0.0137 0.0188 0.0267 0.0152 0.0205 0.0080 0.0100 0.0138 0.0170 0.0034 0.0152

Central Line
?i = 400 33 = 2.2 np - 15 _
>.m '
r
1 1
Control Limits
?i = 400 np3-,/n^ = 2.2 + 4.4 .02 0 and 6.6
'""""l.J-^'l^-J"''**
1,''TJ
\L.,-riJ
NOTE Since the value of np is 2.2, which is less than 4, the N O T E at the end of Section 13 (or 14) applies. The product of n and the upper control limit value for p is 400x0.0166 = 6.64. The non-integral remainder, 0.64, is greater than one-half, and so the adjusted upper control limit value for p is (6.64 -I- l)/400 = 0.0191. Therefore, only the point for Lot 9 is outside limits. For np, by the N O T E of Section 14, the adjusted upper control limit value is 7.6 with the same conclusion.
rr 1
ri*^*,
10
, 15 20 Ljot Number
n^t r-
25
30
FIG. 10Control chart for p. Samples of unequal size, n = 200 to 880; no standard given.
For n = 300 0.01374 3, 0.01374(0.98626) 300 0.013743(0.006720) = 0.0137410.02016 0 and 0.03390 For n = 880 0.01374 3 0.01374(0.98626) 880 0.01374 3(0.003924) = 0.01374 0.01177
Example 8: Control Chart for p, of Unequal Size (Section 13B)
Samples
Table 32 gives inspection results for surface defects on 31 lots of a certain type of galvanized hardware. The lot sizes varied considerably and corresponding variations in sample sizes were used. Figure 10 gives the control chart for fraction nonconforming p. In practice, results are commonly expressed in "percent nonconforming," using scale values of 100 times p.
Central Line 268 = 0.01374 P= 19 510
0.00197 and 0.02551 ResultsÂ state of control may be assumed to exist since 25 consecutive subgroups fall within 3-sigma control limits. There are no points outside limits, so that the N O T E of Section 13 does not apply.
Example 9: Control Charts for u, Samples of Equal Size (Section ISA) and c. Samples of Equal Size (Section 16A) Table 33 gives inspection results in terms of nonconformities observed in the inspection of 25 consecutive lots of burlap bags. Since the number of bags in each lot differed slightly, a constant sample size, n
Control Limits p + 3.
PJ^-P)

TABLE 33. Number of nonconformities in consecutive samples of ten units each^burlap bags. TOTAL NONCONFORMITIES IN SAMPLE SAMPLE
1 2 3 4 5 6 7 8 9 10 11 12 17 14 6 23 5 7 10 19 29 18 25 5
NONCONFORMITIES PER UNIT u

1.7 1.4 0.6 2.3 0.5 0.7 1.0 1.9 2.9 1.8 2.5 0.5
TOTAL NONCONFORMITIES IN SAMPLE SAMPLE

13 14 15 16 17 18 19 20 21 22 23 24 25 Total 11 18 13 22 6 23 22 9 15 20 6 24 375
NONCONFORMITIES PER UNIT u

0.8 1.1 1.8 1.3 2.2 0.6 2.3 2.2 0.9 1.5 2.0 0.6 2.4 37.5
= 10 w a s u s e d . All n o n c o n f o r m i t i e s w e r e counted although two or more n o n c o n f o r m i t i e s of t h e s a m e or different k i n d s o c c u r r e d on t h e s a m e b a g . T h e n o n c o n f o r m i t i e s p e r u n i t v a l u e for e a c h sample w a s d e t e r m i n e d by dividing t h e n u m b e r of n o n c o n f o r m i t i e s found b y t h e s a m p l e size a n d is l i s t e d i n t h e t a b l e . F i g u r e 11 gives t h e control c h a r t for u, a n d Fig. 12 gives t h e control c h a r t for c. Note t h a t these two charts are identical except for t h e v e r t i c a l scale.
Control Limits
71 = 10
3 1.50 + 3V0.150 = 1.5011.16 0.34 and 2.66
Sample Number FIG. 12Control chiart for c. Samples of equal size, n = 10; no standard given. 10 IS Sample Number (b)c FIG. 11Control chart for u. Samples of equal size, n = 10; no standard given.
Central Line 375 c = ^ ^ = 15.0 25
(a) u Central Line

37.5 25
1.5

TABLE 34. Number of nonconformities in samples from 20 successive lots of Type A machines.
TOTAL NONCONFORMITIES SAMPLE c 72 38 76 35 62 81 97 78 103 56 NONCONFORMITIES PER UNIT u 3.60 1.90 1.90 1.40 2.48 3.24 2.42 1.95 2.58 1.40 TOTAL NONCONFORMITIES SAMPLE c 47 55 49 62 71 47 41 52 128 84 1334 NONCONFORMITIES PER UNIT u 1.88 2.20 1.96 2.48 2.84 2.35 2.05 2.60 3.20 2.10
LOT No. No. No. No. No. 1 2 3 4 5
SAMPLE SIZE n 20 20 40 25 25 25 40 40 40 40
LOT No. No. No. No. No. No. No. No. No. No. 11 12 13 14 15 16 17 18 19 20
SAMPLE SIZE n 25 25 25 25 25 20 20 20 40 40 580
No. 6 No. 7 No. 8 No. 9 No. 10
Total
Control Limits K = 10
15.03Vl5 = 15.011.6 3.4 and 26.6
per unit value for each sample, number of nonconformities in sample divided by number of units in sample, was determined and these values are listed in the last column of the table. Figure 13 gives the control chart for u with control limits corresponding to the three different sample sizes.
ResultsPresence of assignable causes of variation is indicated by Sample 9. Since the value of nu is 15 (greater than 4), the N O T E at the end of Section 15 (or 16) does not apply. Example 10: Control Chart for u, of Unequal Size (Section 15B) Samples
Central Line M = 580
1334
= 2.30
Table 34 gives inspection results for 20 lots of different sizes for which 3 different sample sizes were used, 20, 25, and 40. The observed nonconformities in this inspection cover all of the specified characteristics of a complex machine (Type A), including a large number of dimensional, operational, as well as physical and finish requirements. Because of the large number of tests and measurements required as well as possible occurrences of any minor observed irregularities, the expectancy of nonconformities per unit is high, although the majority of such nonconformities are of minor seriousness. The nonconformities
Control Limits = 20
M 3 J - = 2.30 1.02,
Vn 1.28 and 3.32 n = 25
M3J-=2.300.91,
Vn 1.39and3.21 = 40
M 3 j - = 2.3010.72, Vn 1.58 and 3.02

4.0 r
10 Lot Number FIG. 13Control chart for u. Samples of unequal size, n = 20, 25, 40; no standard given.
ResultsLack of control of quality is indicated; plotted points for lot numbers 1, 6, and 19 are above the upper control limit and the point for lot number 10 is below the lower control limit. Of the lots with points above the upper control limit, lot no. 1 has the smallest value of nu (46), which exceeds 4, so t h a t the NOTE at the end of Section 15 does not apply.
maintained for "number of breakdowns in successive lengths of 1000 ft each." So many points were obtained during a short period of production by using the 1000-ft length as a unit and the expectancy in terms of number of breakdowns per length was so small t h a t longer unit lengths were tried. Table 35 gives (a) the "number of breakdowns in successive lengths of 5000 ft each," and (b) the "number of breakdowns in successive lengths of 10 000 ft each." Figure 14 shows the control chart for c where the unit selected is 5000ft, and Fig. 15 shows the control chart for c where the unit selected is 10 000 ft. The standard unit length finally adopted for control purposes was 10 000 ft for 'Tjreakdown."
Example 11: Control Charts for c, Samples of Equal Size (Section 16A) Table 35 gives the results of continuous testing of a certain type of rubber-covered wire at specified test voltage. This test causes breakdowns at weak spots in the insulation, which are cut out before shipment of wire in short coil lengths. The original data obtained consisted of records of the number of breakdowns in successive lengths of 1000 ft each. There may be 0, 1, 2, 3, ..., etc. breakdowns per length, depending on the number of weak spots in the insulation. Such data might also have been tabulated as number of breakdowns in successive lengths of 100 ft each, 500 ft each, etc. Here there is no natural unit of product (such as 1 in., 1 ft, 10 ft, 100 ft, etc.), in respect to the quality characteristic 'Tareakdown" since failures may occur at any point. Since the original data were given in terms of 1000-ft lengths, a control chart might have been
10 20 30 40 50 Succtssive Lengths of S 000 ft. Each FIG. 14Control chart for c. Samples of equal size, n =1 standard length of 5000 ft; no standard given.
60
(a) Lengths of 5 000 ft Each Central Line c = 60 Control Limits 6.233V623 0 and 13.72 = 3.12

TABLE 35. Number of breakdowns in successive lengths of 5 000 ft each and 10 000 ft each for rubber-covered wire. Number Number Number Number Number of of of of of Break- Length Break- Length Break- Length Break- Length Breakdowns No. downs No. downs No. downs No. downs
(a) Lengths of 5 000 f Each t
Length No.
1 2 3 4 5 6 7 8 9 10 11 12 Total 1 2 3 4 5 6
Total
0 1 1 0 2 1 3 4 5 3 0 1
13 14 15 16 17 18 19 20 21 22 23 24
1 1 2 4 0 1 1 0 6 4 3 2
25 26 27 28 29 30 31 32 33 34 35 36
0 0 9 10 8 8 6 14 0 1 2 4
37 38 39 40 41 42 43 44 45 46 47 48
5 7 1 3 3 2 0 1 5 3 4 3
49 50 51 52 53 54 55 56 57 58 59 60 60
5 4 2 0 1 2 5 9 4 2 5 3 187 9 2 3 14 6 8 187
1 1 3 7 8 1
7 8 9 10 11 12
(b) Lengths of 10 000 ft Each 19 2 13 0 14 19 20 6 21 1 15 16 22 1 16 20 23 10 17 1 24 6 5 18
12 4 5 1 8 7
25 26 27 28 29 30 30
(a) ResultsPresence of assignable causes of variation is indicated by length numbers 27, 28, 32, and 56 falling above the upper control limit. Since the value of c=nu is 3.12 (less than 4), the N O T E at the end of Section 16 does apply. The nonintegral remainder of the upper control limit value is 0.42. The upper control limit stands, as do the indications of lack of control.
(b) Lengths of 10 000 ft Each Central Line 187 ,,.,

= 6.23 30
c =
Control Limits C3^| = 6.233V6.23 0 and 13.72
5 10 15 20 25 30 Succsie Ltngtht of 10.000 ft. eoch FIG. 15Control chart for c. Samples of equal size, n = ^ standard length of 10 000 ft; no standard given.
(b) ResultsPresence of assignable causes of variation is indicated by length numbers 14, 15, 16, and 28 falling above the upper control limit. Since the value of c i s 6.23 (greater than 4), the N O T E at the end of Section 16 does not apply.
95 CONTROL CHART MFTHOn
32. Illustrative ExamplesControl With Respect to a Given Standard

Examples 12 to 21, inclusive, illustrate the use of the control chart method of analyzing data for control with respect to a given standard (see Sections 18 to 27).
33.2 and 36.8 . 4/3-4 , JQ For s I"; 7Fo3 = 418 127, 4n-3;'" ^/2n-15 2.91 and 5.45
38
IK at 34
Example 12: Control Charts for X and s, Large Samples of Equal Size (Section 19) A manufacturer attempted to maintain an aimed-at distribution of quality for a certain operating characteristic. The objective standard distribution which served as a target was defined by standard values: jJo = 35.00 lb., and OQ = 4.20 lb. Table 36 gives observed values of X a n d s for daily samples of n = 50 observations each for ten consecutive days. These data are the same as used in Example 1 and presented as Table 27. Figure 16 gives control charts for X and s. Central Lines For X : ^ = 35.00 For s:c^= 4.20 Control Limits n = 50
On For X:|Xo 3-7^= 35.00 + 1.8, Jn
\ *m
^
I I i
I i
i.,_i.
10
î: 4 c o
liV
2
^
^ 1 _
^ . I
.1
r V
P&"
10
Sample HumlMr
FIG. 16Control charts for X and & Large samples of equal size, n = 50; {i,, o given.
ResultsLack of control at standard level is indicated on the eighth and ninth days. Compare with Example 1 in which the same data were analyzed for control without specifying a standard level of quality.
TABLE 36. Operating characteristic, dally control data. Sample 1 2 3 4 5 6 7 8 9 10 Sample Size, n 50 50 50 50 50 50 50 50 50 50 Average, x 35.1 34.6 33.2 34.8 33.4 33.9 34.4 33.0 32.8 34.8 Standard Deviation, s 5.35 4.73 3.73 4.55 4.00 4.30 4.98 5.30 3.29 3.77

Control Limits Example 13: Control Charts for X and s, Large Samples of Unequal Size (Section 19) For a product, it was desired to control a certain critical dimension, the diameter, with respect to day to day variation. Daily sample sizes of 30, 50, or 75 were selected and measured, the number taken depending on the quantity produced per day. The desired level was ô = 0.20000 in. with ao = 0.00300 in. Table 37 gives observed values of X and s for the samples from 10 successive days' production. Figure 17 gives the control charts for X and s. TABLE 37. Diameter in inches, control data.
Sample 1 2 3 4 5 6 7 8 9 10 Sample Size n 30 50 50 30 75 75 75 50 50 30 Average
X
For X: M-03-7^ n. = 30 0.2000 3 : 5 ^ = V30 0.20000 0.00164 0.19836 and 0.20164 n = 50 0.19873 and 0.20127 n, = 75 0.19896 and 0.20104 For s: câ^ 3 -Jin-1.5
?i = 30
Standard Deviation s 0.00330 0.00292 0.00326 0.00358 0.00313 0.00306 0.00299 0.00335 0.00221 0.00397
li^lo.003003^:?22B:
ii7J 4s%l
0.0029710.00118 0.00180 and 0.00415 n = 50 0.00389 and 0.00208
71 = 75
0.20133 0.19886 0.20037 0.19965 0.19923 0.19934 0.19984 0.19974 0.20095 0.19937
0.00225 and 0.00373
0.20200
IK
"<r^-^*i
r^L-J" 0.19800 0.00500
2 4
JC 0.20000 o J
^K
L . _ _ ^ _ _
ResultsThe charts give no evidence of significant deviations from standard values.
s * s
E
10
Example 14: Control Chart for X and s, Small Samples of Equal Size (Section 19) Same product and characteristic as in Example 13, but in this case it is desired to control the diameter of this product with respect to sample variations during each day, since samples of 10 were taken at definite intervals each day. The desired level is Ho = 0.20000 in. with OQ = 0.00300 in. Table 38 gives observed values of X and s for 10 samples of 10 each taken during a single day. Figure 18 gives the control charts for X and s.
Is
tz 0.00300 0.00100
.---u
~ ^ ^ -
'^"7
10
f ~- - - - - w - J ^
4 8 6 S o m p i t Numbtr
FIG. 17Control charts for X and s. Large samples of unequal size, n = 30, 50, 70; ^g,ao given.
Central Lines For X : Ho = 0.20000 Fors: OQ = 0.00300

TABLE 38. Control data for one day's product.
Sample 1 2 3 4 5 6 7 8 9 10 Sample Size n 10 10 10 10 10 10 10 10 10 10 Average 0.19838 0.20126 0.19868 0.20071 0.20050 0.20137 0.19883 0.20218 0.19868 0.19968 Standard Deviation s 0.00350 0.00304 0.00333 0.00337 0.00159 0.00104 0.00299 0.00327 0.00431 0.00356
Example 15: Control Chart for Xand s. Small Samples of Unequal Size (Section 19) A manufacturer wished to control the resistance of a certain product after it had been operating for 100 h, where [IQ = 150 i^ and Oo = 7.5 Q from each of 15 consecutive lots, he selected a random sample of 5 units and subjected them to the operating test for 100 h. Due to mechanical failures, some of the units in the sample failed before the completion of 100 h of operation. Table 39 gives the averages and standard deviations for the 15 samples together with their sample sizes. Figure 19 gives the control charts forX and s.
)T0 |X 160
ISO E < 140 o
Central lines F o r X : \io = 0.20000 n = lO For s: C4O0 = (0.9727)(0.00300) = 0.00292 Control Limits n = 10 For X:
(XQ
-LJ-JT
^ ^ J ~
^ô =
0.20000 (0.949)(0.00300), 0.19715 and 0.20285 For s: Be^ô = (1.669)(0.00300) = 0.00501 B5O0 = (0.276)(0.00300) = 0.00083
130 u c o
I I I ,
10
12 14
SS
0.20300
5
' '
-JL
I 0.20000 >
O.I9S0O
/^v^^^T^C^"
1,.
10
12
14
1 ,
Lot Number FIG. 19Control charts for X and s. Small samples of unequal size, n = 3 , 4,5; \i,,a given.
10
I I 0.00400 I.I 0.00200

2 4 6 8 SompI* Number
10
p.
0.00600
Central Lines F o r X : /! =150 =3 Mo ^(To= 150 1.732(7.5) 137.0 and 163.0 =4 / X o ^ a = 150 1.500(7.5) 138.8andl61.2
FIG.18Control charts for Xand s. Small samples of equal size, n = 10; )i,a,given.
ResultsNo lack of control indicated.

H=5 M ^<7o= 150 1.342(7.5) o 139.9 and 160.1

T A B L E 39. R e s i s t a n c e i n o h m s after 100-h operation, lot b y lot control d a t a . Sample Size Sample Standard Deviation Sample 12.20 9.75 11.20 7.43 4.32 8.65 9.23 7.24 9 10 11 12 13 14 15 Sample Size Standard Deviation
Average
Average
1
154.6 143.4 160.8 152.7 136.0 147.3 161.7 151.0
X
156.2 137.5 153.8 143.4 156.0 149.8 138.2 8.92 3.24 6.85 7.64 10.18 8.86 7.38
For s: oo = 7.5 =3 040-0= (0.8862X7.5) =6.65 n=4: C4(7o= (0.9213X7.5)= 6.91 n, = 5 c,c7o= (0.9400X7.5) =7.05 For s: oo = 7.5 n = 3 :5gC7o =(2.276X7.5)= 17.07 5^(70= (0X7-5) =0 n = 4 :B,c7o =(2.088X7.5) = 15.66 55CT= (0X7.5)= 0 n = 5:B,a^ =(l .964X7.5)= 14.73 5,(70= (0X7.5) =0 ResultsEvidence of lack of control is indicated since samples from lots Numbers 5 and 10 have averages below their lower control limit. No standard deviation values are outside their control limits. Corrective action is required to reduce t h e variation between lot averages.
consecutive lots. Since the sample size n is less than 10, actually 5, he elected to use control charts for X and R rather t h a n for X and s. Figure 20 gives the control charts for X and R.
T A B L E 40, Operating characteristic, lot by lot control data. Sample Size Lot No. No. No. No. No. No. 6 No. 7 No. 8 No. 9 No. 10 36.0 31.4 39.0 35.6 38.8 41.6 36.2 38.0 31.4 29.2 Average Range 6.6 0.5 15.1 8.8 2.2 3.5 9.6 9.0 20.6 21.7
Example 16: Control Charts for X and R, Small Samples of Equal Size (Section 19 and 20) Consider the same problem as in Example 12 where |Lio = 35.00 lb and Oo = 4.20 lb. The manufacturer wished to control variations in quality from lot to lot by taking a small sample from each lot. Table 40 gives observed values of X and R for samples of n = 5 each, selected from ten
4
Lot
6
Number
FIG. 20Control charts for X and R. Small samples of equal size, n = 5; |ig, O given. g

Central Lines For X: Ho = 35.00 =5 For R: dÔo = 2.326(4.20)= 9.i Control Limits n =5 For X: Ho Aco 35.00 + (l.342)(4.20) 29.4 and 40.6 For R: dâ^ = (4.918)(4.20) ^20.7 AiOo = (0)(4.20) = 0 ResultsLack of control a t t h e s t a n d a r d level is i n d i c a t e d b y r e s u l t s for lot n u m b e r s 6 a n d 10. Corrective action is r e q u i r e d b o t h w i t h r e s p e c t to a v e r a g e s a n d w i t h r e s p e c t t o v a r i a b i l i t y w i t h i n a lot. (a)p Central Line
Po = 0.0040
Control Limits = 400 p,3. 0^040^ Po(^-Po) 00040(09960)
V 400 0.0040 0.0095 0 and 0.0135
Example of Equal of Equal
17: Control Charts for p, Size (Section 23) and np, Size (Section 24)
Samples Samples
5 10 Lot Number FIG. 22Control chart for np. Samples of equal size, n = 400; p given.
C o n s i d e r t h e s a m e d a t a a s i n Example 7, T a b l e 3 1 . T h e m a n u f a c t u r e r w i s h e s to c o n t r o l h i s p r o c e s s w i t h r e s p e c t to finish on g a l v a n i z e d w a s h e r s a t a level s u c h t h a t t h e fraction n o n c o n f o r m i n g po - 0.0040 (4 nonconforming washers per thousand). T a b l e 31 of Example 7 gives o b s e r v e d v a l u e s of " n u m b e r of n o n c o n f o r m i n g u n i t s " for finish n o n c o n f o r m i t i e s s u c h a s e x p o s e d steel, r o u g h g a l v a n i z i n g i n s a m p l e s of 400 w a s h e r s d r a w n from 15 successive lots. F i g u r e 2 1 s h o w s t h e control c h a r t for p, a n d Fig. 2 2 gives t h e c o n t r o l c h a r t for np. I n p r a c t i c e , only o n e of t h e s e control c h a r t s w o u l d b e u s e d since, e x c e p t for c h a n g e of scale, t h e t w o c h a r t s a r e identical.
(b)np Central Line npg = 0.0040 (400) = 1.6
Control limits Extract formula:

?i = 4 0 0
"/'o y n p { l ~ p , ) = 1.613^1.6(0.996) = 1.63Vl.5936 = 1.6 3(1.262) 0 and 5.4 Simplified approximate formula: = 400 Since po is small, replace Eq 29 by Eq 30
? 0.02
5
Lot
10
Numbtr
1.63VL6 =
FIG. 21Control ciiart for p. Samples of equal size, n = 400; Pggiven.
1.6 3(1.265) 0 and 5.4

ResultsLack of control of quality is indicated with respect to the desired level; lot numbers 4 and 9 are outside control limits. characteristics at a level such t h a t the fraction nonconforming po = 0.0020 (2 nonconforming units per thousand). Table 41 gives observed values of "number of nonconforming units" for this item found in samples drawn from successive lots. Sample sizes vary considerably from lot to lot and, hence, control limits are computed for each sample. Equivalent control limits for "number of nonconforming units," np, are shown in column 5 of the table. In this way, the original records showing "number of nonconforming units" may be compared directly with control limits for np. Figure 23 shows the control chart for p .
NOTE Since the value of npo is 1.6, less t h a n 4, the NOTE at the end of Section 13 (or 14) applies, as mentioned at the end of Section 23 (or 24). The product of n and the upper control limit value for p is 400 X 0.0135 = 5.4. The nonintegral remainder, 0.4, is less t h a n one-half. The upper control limit stands as does the indication of lack of control to po. For np, by the NOTE of Section 14, the same conclusion follows.
Example 18: Control Chart for p (Fraction Nonconforming), Samples of Unequal Size (Section 23e) The manufacturer wished to control the quality of a type of electrical apparatus with respect to two adjustment
FIG. 23Control chart for p. Samples of unequal size, to 2500; Po given.
TABLE 41. Adjustment irregularities, electrical apparatus.

SAMPLE SIZE n 600 1300 2000 2500 1550 2000 1550 780 260 2000 1550 950 950 950 35 330 200 600 1300 780 NUMBER OF NONCONFORMING UNITS 2 2 1 1 5 2 0 3 0 15 7 2 5 2 0 3 0 4 8 4 FRACTION NON-CONFORMING p UPPER CONTROL LIMIT FOR np UPPER CONTROL LIMIT FOR J3
LOT No. No. No. No. No. No. No. No. No. No. No. No. No. No. No. No. No. No. No. No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
0.0033 0.0015 0.0005 0.0004 0.0032 0.0010 0.0000 0.0038 0.0000 0.0075 0.0045 0.0021 0.0053 0.0021 0.0000 0.0091 0.0000 0.0067 0.0062 0.0051
4.5 7.4 10.0 11.7 8.4 10.0 8.4 5.3 2.7 10.0 8.4 6.0 6.0 6.0 0.9 3.1 2.3 4.5 7.4 5.3
0.0075 0.0057 0.0050 0.0047 0.0054 0.0050 0.0054 0.0068 0.0103 0.0050 0.0054 0.0063 0.0063 0.0063 0.0247 0.0094 0.0115 0.0075 0.0057 0.0068

Central Line for p Po= 0.0020 Control Limits for p
A 3^
A(1-/'O)
Example 19: Control Chart for p (Fraction Rejected), Total and Components, Samples of Unequal Size (Section 23) A control device was given a 100 percent inspection in lots varying in size from about 1800 to 5000 units, each unit being tested and inspected with respect to 23 essentially independent characteristics. These 23 characteristics were grouped into three groups designated Groups A, B, and C, corresponding to three successive inspections. A unit found nonconforming at any time with respect to any one characteristic was immediately rejected; hence units found nonconforming in, say, the Group A inspection were not subjected to the two subsequent group inspections. In fact, the number of units inspected for each characteristic in a group itself will differ from characteristic to characteristic if nonconformities with respect to the characteristics in a group occur, the last characteristic in the group having the smallest sample size. Since 100 percent inspection is used, no additional units are available for inspection to maintain a constant sample size for all characteristics in a group or for all the component groups. The fraction nonconforming with respect to each characteristic is sufficiently small so that the error within a group, although rather large between the first and last characteristic inspected by one inspection group, can be neglected for practical purposes. Under these circumstances, the number inspected for any group was equal to the lot size diminished by the number of units rejected in the preceding inspections. Part 1 of Table 42 gives the data for 12 successive lots of product, and shows for each lot inspected the total fraction rejected as well as the number and fraction rejected at each inspection station. Part 2 of Table 42 gives values of
For n = 600 :
0.0020 3
0.002(0,998) 600
0.0020 3(0.001824) 0 and 0.0075 (same procedure for other values of n)
Control Limits for np Using Eq 30 for np,
For = 600: 1.2 3 Vr2= 1.2 3(1.095), 0 and 4.5 (same procedure for other values of n) ResultsLack of control and need for corrective action indicated by results for lots Numbers 10 and 19.
NOTE The values of npo for these lots are 4.0 and 2.6, respectively. The NOTE at the end of Section 13 (or 14) applies to lot number 19. The product of n and the upper control limit value for p is 1300 x 0.0057 = 7.41. The nonintegral remainder is 0.41, less t h a n one-half. The upper control limit stands, as does the indication of lack of control at po. For np, by the NOTE of Section 14, the same conclusion follows.

Po, fraction rejected, at which levels the manufacturer desires to control this device, with respect to all twenty-three characteristics combined and with respect to the characteristics tested and inspected at each of the three inspection stations. Note that the po for all characteristics (in terms of nonconforming units) is less than the sum of the Po values for the three component groups, since nonconformities from more t h a n one characteristic or group of characteristics may occur on a single unit. Control limits, lower and upper, in terms of fraction rejected are listed for each lot size using the initial lot size as the sample size for all characteristics combined and the lot size available at the beginning of inspection and test for each group as the sample size for t h a t group.
TABLE 42. Inspection data for 100 percent inspectioncontrol device.

Observed Number of Rejects and Fraction Rejected All Groups Combined Total Rejected Lot Lot No. 1 No .2 No. 3 No. 4 No. No. No. No. 5 6 7 8 Size, n 4814 2159 3089 3156 2139 2588 2510 4103 2992 3545 1841 2748 Number 914 359 565 626 434 503 487 803 547 643 353 418 Fraction 0.190 0.166 0.183 0.198 0.203 0.194 0.194 0.196 0.183 0.181 0.192 0.152 Lot Size, n 4814 2159 3089 3156 2139 2588 2510 4103 2992 3545 1841 2748 Group A Rejected Fraction Number 311 128 195 233 146 177 143 318 208 172 97 141 0.065 0.059 0.063 0.074 0.068 0.068 0.057 0.078 0.070 0.049 0.053 0.051 Lot Size, n 4503 2031 2894 2923 1993 2411 2367 3785 2784 3373 1744 2607 Group B Rejected Number 253 105 149 142 101 151 116 242 130 180 119 114 Fraction 0.056 0.052 0.051 0.049 0.051 0.063 0.049 0.064 0.047 0.053 0.068 0.044 Lot Size, n 4250 1926 2745 2781 1892 2260 2251 3543 2654 3193 1625 2493 Group C Reje!Cted Number Fraction 350 126 221 251 187 175 228 243 209 291 137 163 0.082 0.065 0.081 0.090 0.099 0.077 0.101 0.069 0.079 0.091 0.084 0.065
No. 9 No. 10 No. 11 No. 12
Central Lines and Control Limits, Based on Standard p^ Values All Groups Combined Group A Central Lines P= Lot No. No. No. No. No. No. No. No. No. No. No. No. 1 2 3 4 5 6 7 8 9 10 11 12 0.197 0.205 0.201 0.200 0.205 0.203 0.203 0.198 and and and and and and and and 0.163 0.155 0.159 0.160 0.155 0.157 0.157 0.162 0.159 0.160 0.153 0.158 0.081 0.086 0.084 0.084 0.086 0.085 0.085 0.082 0.084 0.083 0.088 0.085 and and and and and and and and and and and and 0.180 0.070 Control Limits 0.059 0.054 0.056 0.056 0.054 0.055 0.055 0.058 0.056 0.057 0.052 0.055 0.060 0.064 0.062 0.062 0.065 0.063 0.064 0.061 0.062 0.061 0.066 0.063 and and and and and and and and and and and and 0.040 0.036 0.038 0.038 0.035 0.037 0.036 0.039 0.038 0.039 0.034 0.037 0.093 0.099 0.096 0.095 0.099 0.097 0.097 0.094 0.096 0.094 0.100 0.096 and and and and and and and and and and and and 0.067 0.061 0.064 0.065 0.061 0.063 0.063 0.066 0.064 0.066 0.060 0.064 0.050 0.080 Group B Group C
0.201 and 0.200 and 0.207 and 0.202 and
103 CONTROL CHART METHOD Group C: n = 4250
Total
u- 0.14
6 8 Lot Number
10
12
n 0.080(0.920) 0.0803 4250 0.080 + 3(0.0042) 0.067 and 0.093
P,^-
Group B
6 8 10 12 Lot Number 0.10
0.021 ' 2 4
Group C
6 8 10 12 Lot Number
1 1 0.06
2 4 6 8 10 12 Lot Number
FIG. 24Control charts for p (fraction rejected) for total and components. Samples of unequal size, n = 1625 to 4814; Po given.
Figure 24 shows four control charts, one covering all rejections combined for the control device and three other charts covering the rejections for each of the three inspection stations for Group A, Group B, and Group C characteristics, respectively. Detailed computations for the over-all results for one lot and one of its component groups are given. Central Lines See Table 42 Control Limits See Table 42 For Lot Number 1 Total: n, = 4814
PQ
ResultsLack of control is indicated for all characteristics combined; lot number 12 is outside control limits in a favorable direction and the corresponding results for each of the three components are less than their standard values, Group A being below the lower control limit. For Group A results, lack of control is indicated since lot numbers 10 and 12 are below their lower control limits. Lack of control is indicated for the component characteristics in Group B, since lot numbers 8 and 11 are above their upper control limits. For Group C, lot number 7 is above its upper limit indicating lack of control. Corrective measures are indicated for Groups B and C and steps should be taken to determine whether the Group A component might not be controlled at a smaller value of po, such as 0.06. The values of npo for lot numbers 8 and 11 in Group B and lot number 7 in Group C are all larger t h a n 4. The NOTE at the end of Section 13 does not apply.
Example 20: Control Chart for u. of Unequal Size (Section 25)
Samples
+ 3 Po (i-ô)
0.1803
0.180(0.820)
4814 0.180 3(0.0055) 0.163 and 0.197
It is desired to control the number of nonconformities per billet to a standard of 1.000 nonconformities per unit in order that the wire made from such billets of copper will not contain an excessive number of nonconformities. The lot sizes varied greatly from day to day so that a sampling schedule was set up giving three different samples sizes to cover the range of lot sizes received. A control program was instituted using a control chart for

TABLE 43. Lot by lot inspection results for copper billets in terms of number of nonconformities and nonconformities per unit. NUMBER OF NONCONFORMITIES c 75 138 212 444 508 312 168 266 119 NONCONFORMITIES PER UNIT u 0.750 1.380 1.060 1.110 1.270 0.780 0.840 1.330 1.190 SAMPLE SIZE n 100 100 400 400 200 200 3500 NUMBER OF NONCONFORMITIES c 130 58 480 316 162 178 3566 1.019 NONCONFORMITIES PER UNIT u 1.300 0.580 1.200 0.790 0.810 0.890
LOT No. No. No. No. No. No. No. No. No. 1 2 3 4 5 6 7 8 9
SAMPLE SIZE n 100 100 200 400 400 400 200 200 100
LOT No. No. No. No. No. No. 10 11 12 13 14 15
Total Overall"
''=3566/3500 = 1.019.
nonconformities per unit with reference to the desired standard. Table 43 gives data in terms of nonconformities and nonconformities per unit for 15 consecutive lots under this program. Figure 25 shows the control chart for u.
= 200
+3 =
V n
1.0003Jl-00 = V 200
1.000 3(0.0707) 0.788 and 1.212
i C _
n = 400 u,3. 1.000 3 1.000 400 1.000 3(0.0500) 0.850 and 1.150
6 8 (0 Lot Number
12
14
FIG. 25Control chart for u. Samples of unequal size, n = 100, 200, 400; n given.
Central Line W =1.000 o Control Limits n = 100 u,3^ = V n 1.000
1.000 3.,
100 1.000 3(0.100) 0.700 and 1.300
ResultsLack of control of quality is indicated with respect to the desired level since lot numbers 2, 5, 8, and 12 are above the upper control limit and lot numbers 6, 11, and 13 are below the lower control hmit. The overall level, 1.019 nonconformities per unit, is slightly above the desired value of 1.000 nonconformities per unit. Corrective action is necessary to reduce the spread between successive lots and reduce the average number of nonconformities per unit. The values of npo for all lots are at least 100 so that the NOTE at end of Section 15 does not apply.

TABLE 44. Daily inspection results for T5rpe D motors in terms of nonconformities per sample and nonconformities per unit.
SAMPLE SIZE n 25 25 25 25 25 25 25 25 25 25 250 25.0 NUMBER OF NONCONFORMITIES c 81 64 53 95 50 73 91 86 99 60 752 75.2 NONCONFORMITIES PER UNIT u 3.24 2.56 2.12 3.80 2.00 2.92 3.64 3.44 3.96 2.40 30.08 3.008
LOT No. No. No. No. No. 1 2 3 4 5
No. 6 No. 7 No. 8 No. 9 No. 10 Total Average
Example 21: Control Charts for c. Samples of Equal Size (Section 26) A Type D motor is being produced by a manufacturer t h a t desires to control the number of nonconformities per motor at a level of Uo= 3.000 nonconformities per unit with respect to all visual nonconformities. The manufacturer produces on a continuous basis and decides to take a sample of 25 motors every day, where a day's product is treated as a lot. Because of the nature of the process, plans are to control the product for these nonconformities at a level such that CQ 75.0 nonconformities and nu^ - CQ. Table 44 gives data in terms of number of nonconformities, c, and also the number of nonconformities per unit, u, for 10 consecutive days. Figure 26 shows the control chart for c. As in Example 20, a control chart may be made for u, where the central line is UQ = 3.000 and the control limits are
Central Line c^ = nu =3.000x25 = 75.0 Control Limits n = 25
75.03V75.0 = 75.03(8.66) 49.02 and 100.98
Lot Number FIG. 26Control chart for c. Sample of equal size, n = 25; C given. o
3.000 + 3
3.000 25
3.000 3(0.3464) 1.96 and 4.04
ResultsNo significant deviations from the desired level. There are no points outside limits so that the N O T E at the end of Section 16 does not apply. In addition, co = 75, larger than 4.
33. Illustrative E x a m p l e s - -Control Chart for I n d i v i d u a l s

Examples 22 to 25, inclusive, illustrate the use of the control chart for individuals, in which individual observations are plotted one by one. The examples cover the two general conditions: (a) control, no standard given; and (b) control with respect to a given standard (see Sections 28 to 30).
High silicon content was known to aid in the production of sound castings, but the specification set a maximum of 1.00 percent silicon for a heat, and all shoes from a heat exceeding this specification were rejected. It was important, therefore, to detect any trouble with silicon control before even one heat exceeded the specification. Since the heats of metal were well stirred, within-heat variation of silicon content was not a useful basis for control limits. However, each 8-h shift used the same materials, equipment etc., and the quality depended largely on the care and efficiency with which they operated so that the five heats produced in an 8-h shift provided a rational subgroup. Data analyzed in the course of an investigation and before standard values were established are shown in Table 45
Example Individuals, Subgroups,
22: Control Chart for X Using Rational Samples of Equal Size, No
Standard Given Based on X and MR (Section 29) In the manufacture of manganese steel tank shoes, five 4-ton heats of metal were cast in each 8-h shift, the silicon content being controlled by ladle additions computed from preliminary analyses.
TABLE 45. Silicon content of heats of manganese steel, percent. Heat Size, n Day Monday Shift 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 15 1 0.70 0.83 0.86 0.80 0.64 0.68 0.80 0.65 0.64 0.77 0.72 0.73 0.79 0.85 0.67 2 0.72 0.68 0.78 0.78 0.66 0.64 0.63 0.81 0.70 0.83 0.67 0.66 0.70 0.80 0.78 3 0.61 0.83 0.71 0.68 0.79 0.71 0.69 0.68 0.66 0.88 0.77 0.72 0.63 0.78 0.81 4 0.75 0.71 0.70 0.70 0.81 0.69 0.62 0.84 0.65 0.70 0.74 0.73 0.70 0.85 0.84 5 0.73 0.73 0.90 0.74 0.68 0.81 0.75 0.66 0.93 0.64 0.72 0.71 0.88 0.62 0.96 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 Sample Average, X 0.702 0.756 0.790 0.740 0.716 0.706 0.698 0.728 0.716 0.764 0.724 0.710 0.740 0.780 0.812 11.082 0.7388 Range, R 0.14 0.15 0.20 0.12 0.17 0.17 0.18 0.19 0.29 0.24 0.10 0.07 0.25 0.23 0.29 2.79 0.186
Tuesday
Wednesday
Thursday
Friday
Total Average

0.90
ResultsNone of the charts give evidence of lack of control.
0.60 Shift 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 Mon. Tues. Wed. Thurs. Fri.

0) "-
0.40
c E 0.20 c tr o o
- < B
: ^'K^ ^ v / ^
Example 23: Control Chart for Individuals, XUsing Rational Subgroups, Standard Given, Based on joo and Oo (Section 29) In the hand spraying of small instrument pins held in bar frames of 25 each, coating thickness and weight had to be delicately controlled and spray-gun adjustments were critical and had to be watched continuously from bar to bar. Weights were measured by careful weighing before and after removal of the coating. Destroying more t h a n one pin per bar was economically not feasible, yet failure to catch a bar departing from standards might result in the unsatisfactory performance of some 24 assembled instruments. The standard lot size for these instrument pins was 100 so t h a t initially control charts for average, and range were set up with n = A.lt was found that the variation in thickness of coating on the 25 pins on a single bar was quite small as compared with the between-bar variation. Accordingly, as an adjunct to the control charts for average and range, a control chart for individuals, X, at the sprayer position was adopted for the operator's guidance. Table 46 gives data comprising observations on 32 pins taken from consecutive bar frames together with 8 average and range values where n = 4. It was desired to control the weight with an average ô= 20.00 mg and Oo = 0.900 mg. Figure 28 shows the control chart for individual values X for coating weights of instrument pins together with the control charts for X and R for samples where n = 4.
Shift 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 IWon. Tues. Wed. Thurs. Fri.
0.50 Shift 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 Mon. Tues. Wed. Thurs. Fri.
FIG. 27Control charts for X, R, and X. Samples of equal size, n = 5; no standard given.
and control charts for X, MR, and X are shown in Fig. 27.
Central Lines For X: X = 0.7388 Fori?: ^ = 0 . 1 8 6 F o r Z : X = 0.7388 Control Limits re = 5
For X:XA^R
0.7388(0.577)(0.186) 0.631 and 0.846 For a Z)4^ = (2.115X0.186)= 0.393 D3^ = (0X0.186)=0
ForZ: 'XEJ^
0.7388 (1.290) (0.186) 0.499 and 0.979

TABLE 46. Coating weights of instrument pins, milligrams. Sample, = 4 Individual Observation, X Individual I 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18.5 21.2 19.4 16.5 17.9 19.0 20.3 21.2 19.6 19.8 20.4 20.5 22.2 21.5 20.8 20.3 Sample Average, X 18.90 Range, R 4.7 Individual 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Total Average Individual Observation, X 19.1 20.6 20.8 21.6 22.8 22.2 23.2 23.0 19.0 20.5 20.3 19.2 20.7 21.0 20.5 19.1 652.7 20.40 Sample Sample, n = 4 Average, X 20.52 Range, R 2.5
19.60
3.3
22.80
1.0
20.08
0.9
19.75
1.5
21.20
1.9
20.32
1.9
163.17 20.40
17.7 2.21
25
Central Line For X : / i o = 20.00
lU
II
o IS
IK
Kf^y__ tc\
I 1 . I . I I I ... H 1 I I I 1 ! . I I 1 I
Control Limits ForZ:Ai3(7o = 20.00 3(0.900) 17.3 and 22.7 Central Lines For X:|Io= 20.00 Fori?: d^<s^= (2.059) (0.900) = 1.85 Control Limits
12 16 ) 24 26 Individuol Numbtr
^ H M .j
32
23
^
_ _
/j^
6?
20
^ ^ 1 " * *
_ J
1,.
1 1... J -
'
5 6
71 = 4
ii:
0
6 r
For X:HoAOo = 20.00 (1.500)(0.900) 18.65 and 21.35 For R: D20o= (4.698) (0.900) = 4.23 DiCJo= (0) (0.900) = 0 ResultsÂll t h r e e c h a r t s s h o w l a c k of control. A t t h e o u t s e t , b o t h t h e c h a r t for
3 4 5 6 Sample Number
FIG. 28Control charts forX, X, and R. Small samples of equal size, n _ 4; (j, ag given.
109 CONTROL CHART METHOD ranges and the chart for individuals gave indications of lack of control. Subsequently, for Sample 6 the control chart for individuals showed the first u n i t in the sample of 4 to be outside its upper control limit, thus indicating lack of control before the entire sample was obtained.
6.0 5.0 4.0 = 1.0
.v^^ v ^
5 10 15 20 25
A V \ A A,
;
0.5
Example 24: Control Charts for Individuals, X, and Moving Range, MR, of Two Observations, No Standard Given Based on X and MR the Mean Moving Range (Section 30A) A distilling plant was distilling and blending batch lots of denatured alcohol in a large tank. It was desired to control the percentage of methanol for this process. The variability of sampling within a single lot was found to be negligible so it was decided feasible to take only one observation per lot and to set control limits based on the moving range of successive lots. Table 47 gives a summary of the methanol content, X, of 26 consecutive lots of the denatured alcohol and the 25 values of the moving range, MR, the range of successive lots with n 2. Figure 29 gives control charts for individuals, X, and the moving range, MR.
.J\^^v
hP \K ?A v.v s
20 25
10 15 Lot Number
Fig. 29Control charts for X and MR. No standard given; based on moving range, where n=2.
Central Lines 128.1 = 4.927 For X : X = 26 For R:R = = 0.288 25 Control Limits
=2 For X: X EJM = X 2.mOMR 4.927 (2.660X0.288) 4.2 and 5.7 For R:DjdR=i^.26l\Q.2U)=Q3A D 3 M = (0X0.288)=0
TABLE 47. Methanol content of successive lots of denatured alcohol and moving range for n = 2.
LOT No. No. No. No. No. No. No. No. No. No. No. No. No. 1 2 3 4 5 6 7 8 9 10 11 12 13 PERCENTAGE OF METHANOL X 4.6 4.7 4.3 4.7 4.7 4.6 4.8 4.8 5.2 5.0 5.2 5.0 5.6 MOVING RANGE MR LOT No. No. No. No. No. No. No. No. No. No. No. No. No. 14 15 16 17 18 19 20 21 22 23 24 25 26 PERCENTAGE OF METHANOLX 5.5 5.2 4.6 5.5 5.6 5.2 4.9 4.9 5.3 5.0 4.3 4.5 4.4 128.1 MOVING RANGE MJR 0.1 0.3 0.6 0.9 0.1 0.4 0.3 0 0.4 0.3 0.7 0.2 0.1 7.2
0.1 0.4 0.4 0 0.1 0.2 0 0.4 0.2 0.2 0.2 0.6
Total

ResultsThe trend pattern of the individuals and their tendency to crowd the control limits suggests t h a t better control may be attainable. Example 25: Control Charts for Individuals, X, and Moving Range, MR, of Two Observations, Standard Given Based on |io and OQ (Section SOB) The data are from the same source as for Example 24, in which a distilling plant was distilling and blending batch lots of denatured alcohol in a large tank. It was desired to control the percentage of water for this process. The variability of sampling within a single lot was found to be negligible so it was decided to take only one observation per lot and to set control limits for individual values, X, and for the moving range, MR, of successive lots with n = 2 where |a,o = 7.800 percent and OQ = 0.200 percent. Table 48 gives a summary of the water content of 26 consecutive lots of the denatured alcohol and the 25 values of the moving range, R. Figure 30 gives control charts for individuals, X, and for the moving range, MR.
9.0 8.0 E 2 7.0
_i_L
I I I
10
15
20
25
10 15 Lot Number
Fig. 30Control charts for X and moving range, MR, wliere n = 2. Standard given; based on |j.o and Oo-
Central Lines For X: Ho =7.800

K= 2 For R: dô^ = (L128XO.200) = 0.23
Control Limits ForZ:|io30 = 7.800 3(0.200) 7.2 and 8.4

K= 2
For R. D2O0 = (3.686)(0.200) = 0.74

DIOQ = (0)(0.200) = 0
TABLE 48. Water content of successive lots of denatured alcohol and moving range forre= 2.
Lot No. 1 No. 2 No. 3 No. 4 No. 5 No. 6 No. 7 No. 8 No. 9 No. 10 No. 11 No. 12 No. 13 Percentage of Water, X 8.9 7.7 8.2 7.9 8.0 8.0 7.7 7.8 7.9 8.2 7.5 7.5 7.9 Moving Range, MR 1.2 0.5 0.3 0.1 0 0.3 0.1 0.1 0.3 0.7 0 0.4 Lot No. No. No. No. No. No. No. No. No. No. No. No. No. 14 15 16 17 18 19 20 21 22 23 24 25 26 Percentage of Water, X 8.2 8.2 7.5 7.5 7.8 8.5 7.5 8.0 8.5 8.4 7.9 8.4 7.5 207.1 26 7.965 Moving Range, MR 0.3 0 0.7 0 0.3 0.7 1.0 0.5 0.5 0.1 0.5 0.5 0.9 10.0 25 0.400
Total Number of values Average
Ill CONTROL CHART MFTHOn ResultsLack of control at desired levels is indicated with respect to both the individual readings and the moving range. These results indicate corrective measures should be taken to reduce the level in percent and to reduce the variation between lots. = 4- 3 2 1 = 24. If A is odd, ^/2) is the : product of all half-integers from k/2 down to 1/2, multiplied by VTT; for example, ii k = 7, so (7/2)! = (7/2) (5/2) (3/2) (1/2) 4^ = 11.6317.
d^=f
where
[-([-a^y-a;;]dx^
(43)
SUPPLEMENT A Mathematical Relations and Tables of Factors for Computing Control C h a r t Lines
Scope S u p p l e m e n t A presents mathematical relations used in arriving at the factors and formulas of PART 3. In addition, S u p p l e m e n t A presents approximations to C4, I/C4, -B3, -B4, -B5, and Be for use when needed. Finally, a more comprehensive tabulation of values of these factors is given in Tables 49 and 50, including reciprocal values of C4 and ^2, and values of da. where ! = , f e '" '^^dx, and n = sample size. ^/2K '-'"
d, = ^ 2 [jl-o(-(\-o^T
+ (o^-c^r]dxdx, -d
(44)
a, = , I e
dx
1
s/lTt'
'e-^^'''^dx
Factors d, d^, and d^, (values for n = 2 to 25, inclusive, in Table 49) The relations given for factors C4, ^2, and da are based on sampling from a universe having a Normal distribution (see Ref. 1, p. 184).
n = sample size, and c?2 = average range for a Normal law distribution with standard deviation equal to unity. (In his original paper, Tippett [9] used w for the range and w for ^2.). The relations just mentioned for C4, dz, and da are exact when the original universe is Normal but this does not limit their use in practice. They may for most practical purposes be considered satisfactory for use in control chart work although the universe is not Normal. Since the relations are involved and thus difficult to compute, values of C4, o?2, and ds for n = 2 to 25, inclusive, are given in Table 49. All values listed in the table were computed to enough significant figures so that when rounded off in accordance with standard practices the last figure shown in the table was not in doubt.
n-2
(42)
n-\
n-3
where the symbol {k/2)\ is called "k/2 factorial" and satisfies the relations (-1/2)! = VTT, 0! = 1, and (t/2)! = (fc/2)[((A; - 2)/2)!] for A = 1, 2, 3, ... . If A: is even, ^/2)! is : simply the product of all integers from k/2 down to 1; for example, if ^ = 8, (8/2)! = 4!

Standard and c Deviations of X, s, R, p, np, u, the sampling distributions of X, s, R, p, etc., for subgroups (samples) of size n. They are not the standard deviations which might be computed from the subgroup values of X, s, R, p, etc., plotted on the control charts but are computed by formula from t h e quantities listed in Table 51.
The standard deviations of X, s, R, p, etc., used in setting 3-sigma control limits and designated a^, Og, CR, Op, etc., in P A R T 3, are t h e standard deviations of
TABLE 49. Factors for computing control chart lines. CHART FOE AVERAGES OBSERVATIONS IN SAMPLE, n 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Over 25 FACTORS FOR CONTROL LIMITS A
CHART FOR STANDARD DEVIATIONS FACTORS FOR CENTRAL LINE

C4
CHART FOR RANGES FACTORS FOR CENTRAL LINE B, 2.606 2.276 2.088 1.964 1.874 1.806 1.751 1.707 1.669
^2
I CONTROL
FACTORS FOR LIMITS

Bt
FACTORS FOR CONTROL LIMITS d , 0.853 0.888 0.880 0.864 0.848 0.833 0.820 0.808 0.797 0.787 0.778 0.770 0.763 0.756 0.750 0.744 0.739 0.733 0.729 0.724 0.720 0.716 0.712 0.708 0 0 0 0 0 0.205 0.388 0.547 0.686 0.811 0.923 1.025 1.118 1.203 1.282 1.356 1.424 1.489 1.549 1.606 1.660 1.711 1.759 1.805 A D, 3.686 4.358 4.698 4.918 5.079 5.204 5.307 5.393 5.469 5.535 5.594 5.647 5.696 5.740 5.782 5.820 5.856 5.889 5.921 5.951 5.979 6.006 6.032 6.056 A 0 0 0 0 0 0.076 0.136 0.184 0.223 0.256 0.283 0.307 0.328 0.347 0.363 0.378 0.391 0.404 0.415 0.425 0.435 0.443 0.452 0.459 A 3.267 2.575 2.282 2.114 2.004 1.924 1.864 1.816 1.777 1.744 1.717 1.693 1.672 1.653 1.637 1.622 1.609 1.596 1.585 1.575 1.565 1.557 1.548 1.541
A,
lie.
B3
4 0 0 0 0 0.029 0.113 0.179 0.232 0.276
lick
2.121 1.880 2.659 1.732 1,023 1.954 1.500 0.729 1.628 1.342 0.577 1.427 1.225 1.134 1.061 1.000 0.949 0.905 0.866 0.832 0.802 0.775 0.750 0.728 0.707 0.688 0.671 0.655 0.640 0.626 0.612 0.600 0.483 0.419 0.373 0.337 0.308 0.285 0.266 0.249 0.235 0.223 0.212 0.203 0.194 0.187 0.180 0.173 0.167 0.162 0.157 0.153 1.287 1.182 1.099 1.032 0.975 0.927 0.886 0.850 0.817 0.789 0.763 0.739 0.718 0.698 0.680 0.663 0.647 0.633 0.619 0.606
a
0.7979 0.8862 0.9213 0.9400 0.9515 0.9594 0.9650 0.9693 0.9727 0.9754 0.9776 0.9794 0.9810 0.9823 0.9835 0.9845 0.9854 0.9862 0.9869 0.9876 0.9882 0.9887 0.9892 0.9896
b
1.2533 1.1284 1.0854 1.0638 1.0510 1.0424 1.0363 1.0317 1.0281 1.0253 1.0230 1.0210 1.0194 1.0180 1.0168 1.0157 1.0148 1.0140 1.0132 1.0126 1.0120 1.0114 1.0109 1.0105
C
0 0 0 0 0.030 0.118 0.185 0.239 0.284 0.321 0.354 0.382 0.406 0.428 0.448 0.466 0.482 0.497 0.510 0.523 0.534 0.545 0.555 0.565
d
3.267 2.568 2.266 2.089 1.970 1.882 1.815 1.761 1.716 1.679 1.646 1.618 1.594 1.572 1.552 1.534 1.518 1.503 1.490 1.477 1.466 1.455 1.445 1.435
e
1.128 1.693 2.059 2.326 2.534 2.704 2.847 2.970 3.078 3.173 3.258 3.336 3.407 3.472 3.532 3.588 3.640 3.689 3.735 3.778 3.819 3.858 3.895 3.931
0.8862 0.5908 0.4857 0.4299 0,3946 0.3698 0.3512 0.3367 0.3249 0.3152 0.3069 0.2998 0.2935 0.2880 0.2831 0.2787 0.2747 0.2711 0.2677 0.2647 0.2618 1.2592 0.2567 0.2544
0.313 1.637 0.346 1.610 0.374 1.585 0.399 1.563 0.421 1.544 0.440 0.458 0.475 0.490 0.504 1.526 1.511 1.496 1.483 1.470
0.516 1.459 0.528 1.448 0.539 1.438 0.549 1.429 0.559 1.420 f
g
3/V
NOTESValues of all factors in this Table were recomputed in 1987 by A. T. A. Holden of the Rochester Institute of Technology. The computed values of d^ and dj as tabulated agree with appropriately rounded values from H. L. Harter, in Order Statistics and Their Use in Testing and Estimation, Vol. 1, 1969, p. 376. "s/Vw-o.s ''(4n-4)/(4-3) "(4-3)/(4-4) ''l-3/V2n-2.5 'l+3/V2-2.5 ^(4-4)/(4-3)-3/V2-1.5 ^ (4 - 4)/(4n - 3)+3/-v/2n-1.5 See Supplement B, NOTE 9, on replacing first term in Footnotes b, c, f, and g by unity

TABLE 50. Factors for computing control limits chart for individuals. CHART FOR INDIVIDUALS FACTORS FOR CONTROL LIMITS OBSERVATIONS IN SAMPLE, n 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Over 25 E, 2.659 1.772 1.457 1.290 1.184 1.109 1.054 1.010 0.975 0.946 0.921 0.899 0.881 0.864 0.849 0.836 0.824 0.813 0.803 0.794 0.785 0.778 0.770 0.763 3/d^ E, 3.760 3.385 3.256 3.192 3.153 3.127 3.109 3.095 3.084 3.076 3.069 3.063 3.058 3.054 3.050 3.047 3.044 3.042 3.040 3.038 3.036 3.034 3.033 3.031 3
TABLE 51. Basis of standard deviations for control limits. Standard Deviation Used in Computing Control Chart X s R P np u c etc., are standard values. 3-Sigma Limits is Computed From ControlNo Standard Given s ori? sorR s ori? P np u c Control-- S t a n d a r d Given Oo Oo Oo Po npo "o Co
N O T E X , R , etc., are computed averages of subgroup values; OQ Po.

The standard deviations o_ and O g The expression under the square root sign in (47) can be rewritten as the reciprocal of a sum of three terms obtained by applying Stirling's formula (see Eq (12.5.3) of [10]) simultaneously to each factorial expression in (47). The result is
(48)
computed in this way are unaffected by any assignable causes of variation between subgroups. Consequently, the control charts derived from them will detect assignable causes of this type. The relations in Eqs. 45 to 55, inclusive, which follow, are all of the form standard deviation of the sampling distribution is equal to a function of both the sample size, n, and a universe value o, p', u', or cf. In practice a sample estimate or standard value is substituted for o, p', u' or c'. The quantities to be substituted for the cases "no standard given" and "standard given" are shown below immediately after each relation. Average, X
(45)
o, =
^}2n~l.5 + P
where Pn is a relatively small positive quantity which decreases toward zero as n increases. For no standard given, substitute j / c ^ or R/d^ for o; for standard given, substitute Oo for o. For control chart purposes, these relations may be used for distributions other than normal. The exact relation of Eq 46 or Eq 47 is used in PART 3 for control chart analyses involving Os and for the determination of factors B3 and B4 of Table 6, and of B5 and Be of Table 16. Range, R
where a is the standard deviation of the universe. For no standard given, substitute 5/C4 or ^ / j ^ for o, or for standard given, substitute OQ for o. Equation 45 does not assume a Normal distribution (see Ref 1, pp. 180 and 181). Standard Deviation, s
2 C4
OR
= dÔ
(49)
where o is the standard deviation of the universe. For no standard given, substitute jjc^ or R/d^ for o; for standard given, substitute Oofor a. The factor da given in Eq 44 represents the standard deviation for ranges in terms of the true standard deviation of a Normal distribution. Fraction Nonconforming, p
<7. = c r ^
(46)
or by substituting the expression for C 4 from Equation (42) and noting {{n l)/2)x((n-3)/2)! = ( ( n - l)/2)!. \
/(I-/)
^P
(50)
n-l
^"-M.fî.
.(47)
where p' is the value of the fraction nonconforming for the universe. For no standard given, substitute p for p' in (50);
115 CONTROL CHART METHOD for standard given, substitute po for p'. When p' is so small that the factor (1 - p') may be neglected, the following approximation is used The number of nonconformities found on any one unit may be considered to result from an unknown but large (practically infinite) number of causes where a nonconformity could possibly occur combined with an unknown but very small probability of occurrence due to any one point. This leads to the use of the Poisson distribution for which the standard deviation is the square root of the expected number of nonconformities on a single unit. This distribution is likewise applicable to sums of such numbers, such as the observed values of c, and to averages of such numbers, such as observed values of u, the standard deviation of the averages being 1/n times that of the sums. Where the number of nonconformities found on any one unit results from a known number of potential causes (relatively a small number as compared with the case described above), and the distribution of the nonconformities per unit is more exactly a multinomial distribution, the Poisson distribution, although an approximation, may be used for control chart work in most instances.
^.=
(51)
Number of Nonconforming Units, np ^, = > / ( ! - / ) (52)
where p' is the value of the fraction nonconforming for the universe. For no standard given, substitute p for p'; and for standard given, substitute p for p'. When p' is so small that the term (1 - p') may be neglected, the following approximation is used (53) The quantity np has been widely used to represent the number of nonconforming units for one or more characteristics. The quantity np has a binomial distribution. Equations 50 and 52 are based on the binomial distribution in which the theoretical frequencies for np 0, 1, 2, . . ., ?i are given by the first, second, third, etc. terms of the expansion of the binomial [(1 - p')\n where p' is the universe value. Nonconformities per Unit, u (54)
N u m b e r of Nonconformities, c Oc = Vi/ = 4c' (55)
<T =
where n is the number of units in sample, u ' i s the value of nonconformities per unit for the universe, and c' is the number of nonconformities in samples of size n for the universe. For no standard given, substitute c=nu for c'; for standard given, substitute c'o nu'o for c'. The distribution of the observed values of c is discussed above.
where n is the number of units in sample, and u ' is the value of nonconformities per unit for the universe. For no standard given, substitute u f o r u'; for standard given, substitute UQ for u'.
F a c t o r s for C o m p u t i n g Control Limits Note that all these factors are actually functions of n only, the constant 3 resulting from the choice of 3-sigma limits.

Averages Individuals
Es=-
(67)
A=
(56)
Jn
(68)
d.
A,= (57)
c^4n
Approximations to Control Chart F a c t o r s for S t a n d a r d D e v i a t i o n s (58) dj'Jn At times it may be appropriate to use approximations to one or more of the control chart factors c^, l/c^, B^, B^, 55,and B^ (see S u p p l e m e n t B, Note 8). The theory leading to Eqs 47 and 48 also leads to the relation (59) (60) (61) 2-2.5 l + (o.046875 + e^)/n'] (69) -^[ 2n-\.5 where Q is a small positive quantity which decreases towards zero as n increases. Eq 69 leads to the approximation 2n-2.5 4n-5 2-1.5 H H - 3
NOTE4= ^ / c , ,
A^=A/d^
Standard deviations B^^c.-Syjl^,
B,=c,+3^ll^,
B,=\-^<
54=1 + V l ^
NOTE-B,=Bjc,,B,=Bjc,.
(62)
C4-
(70)
Ranges
)] = (^2 ~ 3 J3
(63) (64) (65)
which is accurate to 3 decimal places for n of 7 or more, and to 4 decimal places for n of 13 or more. The corresponding approximation for I/C4 is 2-1.5 2-2.5 4n-3 \4n-5
Dj-dj-
M}
A =1-3^
l/c,=
(71)
A=1+3-^
(66)
NOTE>3 = Di/di, D, = Djdi
which is accurate to 3 decimal places for n of 8 or more, and to 4 decimal places for n of 14 or more. In many applications, it is sufficient to use the slightly simpler and slightly less accurate approximation c^=(4n-4)/(4n-3), (72)
which is accurate to within one unit in the third decimal place for /i of 5 or more, a n d

to within one unit in the fourth decimal place for n of 16 or more (See Ref 2, p. 34). The corresponding approximation to I/C4 is l/c, = (4n-3)/(4-4), (73)
53=1--==^=
A/2-1.5
(78)
which has accuracy comparable to that of Eq72.
B, = \ + Jln-l.S
(79)
NOTE The approximations to C4 in Eqs 70 and 72 have the exact relation where V 4 - 5 _ 4 - 4 V4-3 4 - 3 ^ '
(4-4)
With a few exceptions the approximations in Eqs 76, 77, 78 and 79 are accurate to 3 decimal places for n of 13 or more. The exceptions are all one unit off in the third decimal place. That degree of inaccuracy does not limit the practical usefulness of these approximations when n is 25 or more. (See S u p p l e m e n t B, Note 8.) For other approximations to B^, and BQ, see S u p p l e m e n t B, Note 9. Tables 6, 16, 49, and 50 of PART 3 give all control chart factors through n 25. The factors C4, I/C4, 5, Be, B3, and B^ may be calculated for larger values of n accurately to the same number of decimal digits as the tabled values by using Eqs 70, 71, 76, 77, 78 and 79 respectively. If three-digit accuracy suffices for C4 or I/C4, Eqs 72 or 73 may be used for values of n larger than 25.
The square root factor is greater than .998 for ?i of 5 or more. For ri of 4 or more an even closer approximation to C4 than those of Eq 70 and 72 is (4n45)/(4n-35) While the increase in accuracy over Eq 70 is immaterial, this approximation does not require a square-root operation.
From Eqs 70 and 71
7r^=W2n-li
and
(74)
Supplement B Explanatory Notes
I J ^
^1/42^^25
(75)
Note 1 As explained in detail in Supplement A a J and Cs are based (1) on variation of
If the approximations of Eqs 72, 74 and 75 are substituted into Eqs 59, 60, 61 and 62, the following approximations to the Bfactors are obtained: 4-4 3 (76)
4n-3 -Jln-l.S
4n-4 4n-3 3 -Jln-l.S (77)
individual values within subgroups and the size TI of a subgroup for the first use (A) ControlNo Standard Given, and (2) on the adopted standard value of <7 and the size TI of a subgroup for the second use (B) Control with Respect to a Given Standard. Likewise, for the first use, Op is based on the average value of p, designated p, and n, and for the second use from po ai^d n. The method for determining c^ is outlined in Supplement

A. For purpose (A) the estimated from the data. as must be It should be pointed out that the problem of choosing a control chart criterion for use in "ControlNo Standard Given" is not essentially a problem in estimation. The criterion is by nature more a test of consistency of the data themselves and must be based on the data at h a n d including some which may have been influenced by the assignable causes which it is desired to detect. The final justification of a control chart criterion is its proven ability to detect assignable causes economically under practical conditions. When control has been achieved and standard values are to be based on the observed data, the problem is more a problem in estimation, although in practice many of the assumptions made in estimation theory are imperfectly met and practical considerations, sampling trials, and experience are deciding factors. In both cases, data are usually plentiful and efficiency of estimation a minor consideration.
Note 2 This is discussed fully by Shewhart [1]. In some situations in industry in which it is important to catch trouble even if it entails a considerable amount of otherwise unnecessary investigation, 2-sigma limits have been found useful. The necessary changes in the factors for control chart limits will be apparent from their derivation in the text and in S u p p l e m e n t A. Alternatively, in process quality control work, probability control limits based on percentage points are sometimes used (See Ref. 2, pp. 15-16).
Notes From the viewpoint of the theory of estimation, if normality is assumed, an unbiased and efficient estimate of the standard deviation within subgroups is
1 l(n,-l)sf+--- + in,-iy
+ n,-k
(80) Note 4
where C4 is to be found from Table 6, corresponding to n, = ?Xi + . . . + Uk-k + \. Actually, C4 will lie between .99 and unity if '^l + . . . + Mt - A + I is as large as 26 or f ; more as it usually is, whether Ui, n^, etc. be large, small, equal, or unequal. Equations 4, 6, and 9, and the procedure of Sections 8 and 9, "Control No Standard Given," have been adopted for use in PART 3 with practical considerations in mind, Eq 6 representing a departure from that previously given. From the viewpoint of the theory of estimation they are unbiased or nearly so when used with the appropriate factors as described in the text and for Normal distributions are nearly as efficient as Equation 80.
If most of the samples approximately equal size^, effort saved by first computing and approximate control limits based
are of may be plotting on some
^ According to Ref. 13, p. 18, "If the samples to be used for a p-chart are not of the same size, then it is sometimes permissible to use the average sample size for the series in calculating the control limits." As a rule of thumb, the authors propose that this approach works well as long as, "the largest sample size is no larger than twice the average sample size, and the smallest sample size is no less than half the average sample size." Any samples, whose sample sizes are outside this range, should either be separated (if too big) or combined (if too small) in order to make them of comparable size. Otherwise, the only other option is to compute control limits based on the actual sample size for each of these affected samples.
119 CONTROL CHART METHOD typical sample size, such as the most frequent sample size, standard sample size, or the average sample size. Then, for any point questionably near the limits, the correct limits based on the actual sample size for the point should be computed and also plotted, if the point would otherwise be shown in incorrect relation to the limits. on a painted surface, the underlying distribution may often usefully be assumed to be the Poisson distribution. (The proportion of nonconformities expected in the sample, p, is indeterminate and usually small; and the possible number of occurrences of nonconformities in the sample, n, is also indeterminate and usually large; but the product np is finite. For the sample this np value is c.) (For example, see Equation 22 in PART 3.) For characteristics of types (a) and (6) the fraction p is almost invariably small, say less t h a n 0.10, and under these circumstances the Poisson distribution may be used as a satisfactory approximation to the binomial. Hence, in general, for all these three types of characteristics, taken individually or collectively, we may use relations based on the Poisson distribution. The relations given for control limits for number of nonconformities (Sections 16 and 26) have accordingly been based directly on the Poisson distribution, and the relations for control limits for nonconformities per unit (Sections 15 and 25), have been based indirectly thereon.
Notes
Here it is of interest to note the nature of the statistical distributions involved, as follows. (a) With respect to a characteristic for which it is possible for only one nonconformity to occur on a unit, and, in general, when the result of examining a unit is to classify it as nonconforming or conforming by any criterion, the underlying distribution function may often usefully be assumed to be the binomial, where p is the fraction nonconforming and n is the number of units in the sample (for example, see Equation 14 in PART 3). {b) With respect to a characteristic for which it is possible for two, three, or some other limited number of defects to occur on a unit, such as poor soldered connections on a unit of wired equipment, where we are primarily concerned with the classification of soldered connections, rather than units, into nonconforming and conforming, the underlying distribution may often usefully be assumed to be the binomial, where p is the ratio of the observed to the possible number of occurrences of defects in the sample and n is the possible number of occurrences of defects in the sample instead of the sample size (for example, see Equation 14 in this part, with n defined as number of possible occurrences per sample). (c) With respect to a characteristic for which it is possible for a large but indeterminate number of nonconformities to occur on a unit, such as finish defects
Note 6 In the control of a process, it is common practice to extend the central line and control limits on a control chart to cover a future period of operations. This practice constitutes control with respect to a standard set by previous operating experience and is a simple way to apply this principle when no change in sample size or sizes is contemplated. When it is not convenient to specify the sample size or sizes in advance, standard values of \i, a, etc. may be derived from past control chart data using the relations

;U = X = X (if individual chart) R s MR . . . . . . _^. a= or = (it md. chart) d^ c^ c?2 p=p np = np u^ =u c,=c
where the values on the right-hand side of the relations are derived from past data. In this process a certain amount of arbitrary judgment may be used in omitting data from subgroups found or believed to be out of control.
Note 7 It may be of interest to note that, for a given set of data, the mean moving range as defined here is the average of the two values of R which would be obtained using ordinary ranges of subgroups of two, starting in one case with the first observation and in the other with the second observation. The mean moving range is capable of much wider definition [11] but that given here has been the one used most in process quality control. When a control chart for averages and a control chart for ranges are used together, the chart for ranges gives information which is not contained in the chart for averages and the combination is very effective in process control. The combination of a control chart for individuals and a control chart for moving ranges does not possess this dual property; all the information in the chart for moving ranges is contained, somewhat less explicitly, in the chart for individuals.
2based on an empirical and economic judgment, and (2) on data that may be appreciably affected by measurement error. In addition, the assumed theory on which these factors are based cannot be applied with unerring precision. Somewhat cruder approximations to the exact theoretical values are quite useful in many practical situations. The form of approximation, however, must be simple to use and reasonably consistent with the theory. The approximations in PART 3, including S u p p l e m e n t A, were chosen to satisfy these criteria with little loss of numerical accuracy. Approximate formulas for the values of control chart factors are most often useful under one or both of the following conditions: (1) when the subgroup sample size n exceeds the largest sample size for which the factor is tabled in this Manual; or (2) when exact calculation by computer program or by calculator is considered too difficult. Under one or both of these conditions the usefulness of approximate formulas may be affected by one or more of the following: (a) there is unlikely to be an economically justifiable reason to compute control chart factors to more decimal places than given in the tables of this Manual; it may be equally satisfactory in most practical cases to use an approximation having a decimal-place accuracy not much less than that of the tables, for instance, one having a known maximum error in the same final decimal place; (b) the use of factors involving the sample range in samples larger than 25 is inadvisable; (c) a computer (with appropriate software) or even some models of pocket calculator may be able to compute from an exact formula by subroutines so fast that little or nothing is gained either by approximating the exact formula or by storing a table in memory; (d) because some approximations suitable for large sample sizes are unsuitable for small ones, computer programs using
Notes
The tabled values of control chart factors in this Manual were computed as accurately as needed to avoid contributing materially to rounding error in calculating control limits. But these limits also depend: (1) on the factor 3or perhaps

approximations for control chart factors may require conditional branching based on sample size.
REFERENCES
[1] Shewhart, W. A., Economic Control of Quality of Manufactured Product, Van Nostrand, New York, 1931; republished by ASQC Quality Press, Milwaukee, WI, 1980. American National Standards Zl.1-1985 (ASQC Bl-1985), "Guide for Quality Control Charts," Zl.2-1985 (ASQC B2-1985), "Control Chart Method of Analyzing Data," Zl.3-1985 (ASQC B3-1985), "Control Chart Method of Controlling Quality During Production," American Society for Quality Control, Nov. 1985, Milwaukee, WI, 1985. Simon, L. E., An Engineer's Manual Statistical Methods, Wiley, New York, 1941. of
Note 9 The value of c^ rises towards unity as n increases. It is then reasonable to replace C by unity if control limit calculations can 4 thereby be significantly simplified with little loss of numerical accuracy. For instance, Equations (4) and (6) for samples of 25 or more ignore C factors in the 4 calculation of J. The maximum absolute percentage error in width of the control limits on X or s is not more than 100 (1 C4) %, where C4 applies to the smallest sample size used to calculate J. Previous versions of this Manual gave approximations to B5 and Be which substituted unity for C4 and used 2{n -1) instead of 2 -1.5 in the expression under the square root sign of Eq 74. These approximations were judged appropriate compromises between accuracy and simplicity. In recent years three changes have occurred: (a) simple, accurate and inexpensive calculators have become widely available; (b) closer but still quite simple approximations to B5 and B^ have been devised; and (c) some applications of assigned standards stress the desirability of having numerically accurate limits. (See Examples 12 and 13). There t h u s appears to be no longer any practical simplification to be gained from using the previously published approximations for 5 and BQ. The substitution of unity for C4 shifts the value for the central line upwards by approximately (25/fi)%; the substitution of 2{n - 1) for 2n - 1.5 increases the width between control limits by approximately {12ln)%. Whether either substitution is material depends on the application.
[2]
[3]
[4]
British Standard 600:1935, Pearson, E. S., "The Application of Statistical Methods to Industrial Standardization and Quality Control;" British Standard 600 R:1942, Dudding, B. P. and Jennett, W. J., "Quality Control Charts," British Standards Institution, London, England. Bowker, A. H. and Lieberman, G. L., Engineering Statistics, 2nd ed., Prentice-Hall, Englewood Cliffs, N.J., 1972. Burr, I. W., Engineering Statistics and Control, McGraw-Hill, New York, 1953. Quality
[5]
[6]
[7]
Duncan, A. J., Quality Control and Industrial Statistics, 5th ed., Irwin, Homewood, IL, 1986. Grant, E. L. and Leavenworth, R. S., Statistical Quality Control, 5th ed., McGraw-Hill, New York, 1980. Tippett, L. H. C , "On the Extreme Individuals and the Range of Samples Taken from a Normal Population," Biometrika, Vol. 17, 1925, pp. 3 6 4 387.
[8]
[9]
[10] Cramer, H., Mathematical Methods of Statistics, Princeton University Press, Princeton, NJ, 1946. [11] Hoel, P. G., ' T h e Efficiency of the Mean Moving Range," The Annals of Mathematical Statistics, Vol. 17, No. 4, Dec. 1946, pp. 475-482. [12] Ott, E. R., Schilling, E. G.,. and Neubauer, D. v . . Process Quality Control, 3rd ed., McGrawHill, New York, N.Y., 2000.

[13] Small, B. B., ed., Statistical Quality Control Handbook, AT&T Technologies, Indianapolis, IN, 1984. Crosier, R. B., "Multivariate Generalizations of Cumulative Sum Quality-Control Schemes," Technometrics, Vol. 30, 1988, pp. 291-303. Goel, A. L. and Wu, S. M., "Determination of A. R. L. and A Contour Nomogram for CUSUM Charts to Control Normal Mean," Technometrics, Vol. 13, 1971, pp. 221-230. Johnson, N. L. and Leone, F. C , "Cumulative Sum Control ChartsMathematical Principles Applied to Their Construction and Use," Industrial Quality Control, J u n e 1962, pp. 1 5 21; July 1962; pp. 29-36; and Aug. 1962, pp. 2 2 28. Johnson, R. A. and Bagshaw, M., "The Effect of Serial Correlation on the Performance of CUSUM Tests," Technometrics, Vol. 16, 1974, pp. 103-112. Kemp, K. W., "The Average Run Length of the Cumulative Sum Chart When a V-Mask is Used," Journal of the Royal Statistical Society, Series B, Vol. 23, 1961, pp. 149-153. Kemp, K. W., "The Use of Cumulative Sums for Sampling Inspection Schemes," Applied Statistics, Vol. 11, 1962, pp. 16-31. Kemp, K. W., "An Example of Errors Incurred by Erroneously Assuming Normality for CUSUM Schemes," Technometrics, Vol. 9, 1967, pp. 4 5 7 464. Kemp, K. W., "Formal Expressions Which Can Be Applied in CUSUM Charts," Journal of the Royal Statistical Society, Series B, Vol. 33, 1971, pp. 331-360. Lucas, J. M., "The Design and Use of V-Mask Control Schemes," Journal of Quality Technology, Vol. 8, 1976, pp. 1-12. Lucas, J. M. and Crosier, R. B., "Fast Initial Response (FIR) for Cumulative Sum Quantity Control Schemes," Technometrics, Vol. 24, 1982, pp. 199-205. Page, E. S., Technometrics, Used more for control purposes than data presentation. This selection of papers illustrates the variety and intensity of interest in control chart methods. They differ widely in practical value. "Cumulative Sum Vol. 3, 1961, pp. 1-9. Charts,"
SELECTED PAPERS ON CONTROL CHART TECHNIQUES'

A. G e n e r a l Alwan, L. C. and Roberts, H. V., "Time-Series Modeling for Statistical Process Control," Journal of Business & Economic Statistics, Vol. 6, 1988, pp. 393-400. Barnard, G. A., "Control Charts and Stochastic Processes," Journal of the Royal Statistical Society, Series B, Vol. 21, 1959, pp. 239-271. Ewan, W. D. and Kemp, K. W., "Sampling Inspection of Continuous Processes with No Autocorrelation Between Successive Results," Biometrika, Vol. 47,1960, p. 363. Freund, R. A., "A Reconsideration of t h e Variables Control Chart," Industrial Quality Control, Vol. 16, No. 11, May 1960, pp. 35-41. Gibra, I. N., "Recent Developments in Control Chart Techniques," Journal of Quality Technology, Vol. 7, 1975, pp. 183-192. Vance, L. C , "A Bibliography of Statistical Quality Control Chart Techniques, 1970-1980," Journal of Quality Technology, Vol. 15, 1983, pp. 59-62.
B. Cumulative Sum (CUSUM) Charts

Crosier, R. B., "A New Two-Sided Cumulative Sum Quality-Control Scheme," Technometrics, Vol. 28, 1986, pp. 187-194.
Vance, L., "Average Run Lengths of Cumulative Sum Control Charts for Controlling Normal Means," Journal of Quality Technology, Vol. 18, 1986, pp. 189-193.

Woodall, W. H. and Ncube, M. M., "Multivariate CUSUM Quality-Control Procedures," Technometrics, Vol. 27, 1985, pp. 285-292. Woodall, W. H., ' T h e Design of CUSUM Quality Charts," Journal of Quality Technology, Vol. 18, 1986, pp. 99-102. Jaehn, A. H., "Improving QC Efficiency with Zone Control Charts," ASQC Quality Congress Transactions, Minneapolis, MN, 1987. Langenberg, P. and Iglewicz, B., "Trimmed X a n d R Charts," Journal of Quality Technology, Vol. 18, 1986, pp. 151-161. Page, E. S., "Control Charts with Warning Lines," Biometrika, Vol. 42, 1955, pp. 243-254. Reynolds, M. R., Jr., Amin, R. W., Arnold, J. C , and Nachlas, J. A., " X Charts with Variable Sampling Intervals," Technometrics, Vol. 30, 1988, pp. 181-192. Roberts, S. W., "Properties of Control Chart Zone Tests," The Bell System Technical Journal, Vol. 37, 1958, pp. 83-114. Roberts, S. W., "A Comparison of Some Control Chart Procedures," Technometrics, Vol. 8, 1966, pp. 411-430.
C. Exponentially Weighted Moving Average (EWMA) Charts

Cox, D. R., "Prediction by Exponentially Weighted Moving Averages and Related Methods," Journal of the Royal Statistical Society, Series B, Vol. 23, 1961, pp. 414-422. Crowder, S. V., "A Simple Method for Studying RunLength Distributions of Exponentially Weighted Moving Average Charts," Technometrics, Vol. 29, 1987, pp. 401-408. Hunter, J. S., "The Exponentially Weighted Moving Average," Journal of Quality Technology, Vol. 18, 1986, pp. 203-210. Roberts, S. W., "Control Chart Tests Based on Geometric Moving Averages," Technometrics, Vol. 1, 1959, pp. 239-250.
E. Special Applications of Control Charts

Case, K. E., "The p Control Chart Under Inspection Error," Journal of Quality Technology, Vol. 12, 1980, pp. 1-12.
D. Charts Using Various Methods

Beneke, M., Leemis, L. M., Schlegel, R. E., and Foots, F. L., "Spectral Analysis in Quality Control: A Control Chart Based on the Periodogram," Technometrics, Vol. 30, 1988, pp. 63-70. Champ, C. W. and Woodall, W. H., "Exact Results for Shewhart Control Charts with Supplementary Runs Rules," Technometrics, Vol. 29, 1987, pp. 393-400. Ferrell, E, B., "Control Charts Using Midranges and Medians," Industrial Quality Control, Vol. 9, 1953, pp. 30-34. Ferrell, E. B., "Control Charts for Log-Normal Universes," Industrial Quality Control, Vol. 15, 1958, pp. 4-6. Hoadley, B., Quality Technical 1979, pp. "An Empirical Bayes Approach to Assurance," ASQC 33rd Annual Conference Transactions, 14-16 May 257-263. Freund, R. A., "Acceptance Control Charts," Industrial Quality Control, Vol. 14, No. 4, Oct. 1957, pp. 13-23. Freund, R. A., "Graphical Process Control," Industrial Quality Control, Vol. 18, No. 7, J a n . 1962, pp. 15-22. Nelson, L. S., "An Early-Warning Test for Use with the Shewhart p Control Chart," Journal of Quality Technology, Vol. 15, 1983, pp. 68-71. Nelson, L. S., "The Shewhart Control Chart-Tests for Special Causes," Journal of Quality Technology, Vol. 16, 1984, pp. 237-239.
F. Economic Design of Control Charts

Banerjee, P. K. and Rahim, M. A., "Economic Design of X-Control Charts Under WeibuU Shock

Models," Technometrics, 414. Vol. 30, 1988, pp. 407-
Duncan, A. J., "Economic Design of X Charts Used to Maintain Current Control of a Process," Journal of the American Statistical Association, Vol. 51, 1956, pp. 22S-242. Lorenzen, T. J. and Vance, L. C , "The Economic Design of Control Charts: A Unified Approach," Technometrics, Vol. 28, 1986, pp. 3-10. Montgomery, D. C , "The Economic Design of Control Charts: A Review and Literature Survey," Journal of Quality Technology, Vol. 12, 1989, pp. 75-87. Woodall, W. H., "Weakness of the Economic Design of Control Charts," (Letter to the Editor, with response by T. J. Lorenzen and L. C. Vance), Technometrics, Vol. 28, 1986, pp. 408-410.
MNL7A-EB/Feb. 2002
125 APPENDIX
APPENDIX
List of Some Related Publications on Quality Control ASTM Standards
E29-93a (1999), Standard Practice for Using Significant Digits in Test Data to Determine Conformance with Specifications El22-00 (2000), Standard Practice for Calculating Sample Size to Estimate, With a Specified Tolerable Error, the Average for Characteristic of a Lot or Process J u r a n , J. M. and Godfrey, A. B., Juran's Quality Control Handbook, 5 * ed., McGraw-Hill, New York, 1999.* Mood, A. M., Graybill, F. A., and Boes, D. C , Introduction the Theory of Statistics, 3rd ed., McGraw-Hill, New York, 1974. Moroney, M. J., Facts from Figures, 3rd ed.. Penguin, Baltimore, MD, 1956. Ott, EUis, Schilling, E. G. and Neubauer, D. V., Process Quality Control, 3""^ ed., McGraw-Hill, New York, 2000.* Rickmers, A. D. and Todd, H. N., StatisticsAn Introduction, McGraw-Hill, New York, 1967.* Selden, P. H., Sales Process Engineering, ASQ Quality Press, Milwaukee, 1997. Shewhart, W. A., Economic Control of Quality of Manufactured Product, Van Nostrand, New York, 1931.* Shewhart, W. A., Statistical Method from the Viewpoint of Quality Control, Graduate School of the U.S. Department of Agriculture, Washington, DC, 1939.* Simon, L. E., An Engineer's Manual of Statistical Methods, Wiley, New York, 1941.* Small, B. B., ed.. Statistical Quality Control Handbook, AT&T Technologies, Indianapohs, IN, 1984.* Snedecor, G. W. and Cochran, W. G., Statistical Methods, 8 * ed., Iowa State University, Ames, lA, 1989. Tippett, L. H. C , Technological Applications of Statistics, Wiley, New York, 1950. Wadsworth, H. M., Jr., Stephens, K. S., and Godfrey, A. B., Modern Methods for Quality Control and Improvement, Wiley, New York, 1986.* Wheeler, D. J. and Chambers, D. S., Understanding Statistical Process Control, 2^^ ed., SPC Press, Knoxville, 1992.*
Texts
Bennett, C. A. and Franklin, N. L., Statistical Analysis in Chemistry and the Chemical Industry, New York, 1954. Bothe, D. R., Measuring Process Capability, McGraw-Hill, New York, 1997. Bowker, A. H. and Lieberman, G. L., Engineering Statistics, 2''ed., Prentice-Hall, Englewood Cliffs, N J 1972.* Box, G. E. P., Hunter, W. G., and Hunter, J. S., Statistics for Experimenters, Wiley, New York, 1978. Burr, I. W., Statistical Quality Control Methods, Marcel Dekker, Inc., New York, 1976.* Carey, R. G., and Lloyd, R. C , Measuring Quality Improvement in Healthcare: A Guide to Statistical Process Control Applications, ASQ Quality Press, Milwaukee, 1995.* Cramer, H., Mathematical Methods of Statistics, Princeton University Press, Princeton, NJ, 1946. Dixon, W. J. and Massey, F. J., Jr., Introduction to Statistical Analysis, 4* ed., McGraw-Hill, New York, 1983. Duncan, A. J., Quality Control and Industrial Statistics, 5th ed., Richard D. Irwin, Inc., Homewood, IL, 1986.* Feller, William, An Introduction to Probability Theory and Its Application, S^d ed., Wiley, New York, vol. 1, 1970, vol. 2, 1971. Grant, E. L. and Leavenworth, R. S., Statistical Quality Control, 7ti> ed., McGraw-Hill, New York, 1996.* Guttman, I., Wilks, S. S., and Hunter, J. S., Introductory Engineering Statistics, S'^ ed., Wiley, New York, 1982. Hald, A., Statistical Theory and Engineering Applications, Wiley, New York, 1952. Hoel, P. G., Introduction to Mathematical Statistics, 5"^ ed., Wiley, New York, 1984. Jenkins, L., Improving Student Learning: Applying Deming's Quality Principles in Classrooms, ASQ Quality Press, Milwaukee, 1997.*
Journals
Annals of Statistics Applied Statistics (Royal Statistics Society, Series C) Journal of the American Statistical Association Journal of Quality Technology* Journal of the Royal Statistical Society, Series B Quality Engineering* Quality Progress* Technometrics*
*With special reference to quality control
Copyright
2002 by A S T M International
www.astm.org
MNL7A-EB/Feb. 2002
127 INDEX
Index
Central lines formulas for standard given, 75-80 standard not given, 64-74 mathematical relations and tables of factors for computing, 111-118 Central tendency measures of, 23-24 See also specific measures of central tendency Chance causes, 59 Chebyshev's inequality, 28-29 Classifying observations, into cells, 14, 17 Coefficient of kurtosis. See g^ (sample coefficient of kurtosis); zi (population coefficient of kurtosis) Coefficient of skewness. See g^ (sample coefficient of skewness); t i (population coefficient of skewness) Coefficient of variation. See cv (sample coefficient of variation); C F (population coefficient of variation) Computers, and presentation of data and control chart analysis, 3 Condensing data, 7-41 Confidence limits. See limits of uncertainty (for X) Constancy of cause system, lack of, detecting, 59 Control chart method, 56-124 control chart lines formulas for (standard given), 75-80 formulas for (standard not given), 64-74 mathematical relations and tables of factors for computing, 111-118 control limits, 64-80 criteria of control, 61-64, 118 essential features of a control chart, 60 general technique for using, 59-60 importance of, 3 for individuals, 80-81 purpose of, 80-81 using moving ranges, 81, 117-118 example (standard given), 110-111 example (standard not given), 109-110 using rational subgroups, 60, 81 example (standard given), 107-109 example (standard not given), 106-107 literature on, 122-124 purpose of, 57-59 and rational subgroups, 59 standard given, 74-80 for attributes data, 76-80 examples, 100, 106, 107 for c (number of nonconformities), 79-80 example (equal size samples), 106, 107 forp (fraction nonconforming), 77, 78, 80 example (unequal size samples), 94-98 forp (fraction rejected), example (unequal size samples), 93-95 for np (number of nonconforming units), 78-80 example (equal size samples), 94-97
a, defined, 42 A, defined, 76, 116 American Society for Testing and Materials, purposes and objectives of, 1 Approximations to control chart factors, 116-117, 120 Arithmetic mean. See [i (population average or universe mean); X (sample arithmetic mean) Assignable causes defined, 56, 59 determining if present, 60-61 eliminating, 60 Attributes data control charts (standard given), 74-80 examples, 95-105 control charts (standard not given), 64-74 examples, 82-94 Average. See [i (population average or universe mean); X (sample arithmetic mean) Average deviation, defined, 25 Average quality, essential information for, 39
B Bs, defined, 116, 117 B^, defined, 116, 117 Ss, defined, 76, 77, 116, 117, 121 Be, defined, 76, 77, 116, 117, 121 Bimodal frequency distributions, 24, 34 Breaking strength, presentation of data for, 915, 42, 47-48, 50-51
c (number of nonconformities) control chart (standard given), 79-80 example, 105 control chart (standard not given), 73, 74 examples, 91-94 defined, 56 standard deviation ( a ^ ) of, 115 C4, 57, 75-77, 111, 112, 114, 116-117 Causes of variation in quality detecting lack of constancy of, 59 See also assignable causes; chance causes Cells boundary selection for, 14, 15, 17 classifying observations into, 14, 17 defined, 14 frequency for, defined, 14 graphic representation of, 18-19 number of, 14 relative frequency for, defined, 14 tabular presentation of, 17-18
Copyright
2002 by A S T M International
www.astm.org
128 INDEX
purpose of, 59, 75, 76 summary charts of formulas, 78, 80 for u (nonconformities per unit), 78, 80 example (unequal size samples), 105, 106 for X and R, 77, 78 example (small samples of equal size), 94, 97 for X and s example (large samples of equal size), 93, 94,95 example (large samples of unequal size), 94, 95,96 example (small samples of equal size), 93, 96,97 example (small samples of unequal size), 97 standard not given, 64-74 for attributes data, 66-74 examples, 82-94 for c (number of nonconformities), 73, 74 example (equal size samples), 91-94 for np (number of nonconforming units), 66-72, 74 example (equal size samples), 82, 83 ior p (fraction nonconforming), 64-66, 74 example (equal size samples), 88-90 example (unequal size samples), 90 purpose of, 59, 61 summary charts of formulas, 64, 74 for u (nonconformities per unit), 71-73 example (equal size samples), 91-92 example (unequal size samples), 92-94 for X and R (small samples), 63-65 example (equal size samples), 80-82 example (unequal size samples), 81-82 for X and s (large samples), 64-65, 67 example (equal size samples), 82-83 example (unequal size samples), 83-85 for X and s (small samples), 65, 67 example (equal size samples), 83, 84 example (unequal size samples), 85 terminology and technical background, 56-59 Controlled conditions data not obtained under, 5, 39 evidence for, 40 and statistical control, 40 and X and s, 29-30, 38, 39 Costly testing, and control chart method, 80 Cumulative frequency distributions, 14-15 graphic representation of, 19, 20 cv (sample coefficient of variation) amount of information contained in, 33-34 defined, 5-6, 24-25 standard deviation (s) vs., 33-34 CF(population coefficient of variation), 6 D D,, defined, 77, 112, 116 d^ defined, 57,77, 111, 112 D^ defined, 77, 112, 116 da, defined. 111, 112 Ds, defined, 112, 116 Dt, defined, 112, 116 Data breaking up into rational subgroups, 60 condensing, 7-41 homogeneous, 8, 29 interpretation of, factors affecting, 39-40 objects of gathering, 1 physical, examples of, 7, 8-9 raw, table of, 8-9 type considered in manual, 6 See also information; observations; presentation of data Defectives, nonconformities distinguished from, 68 Definitions for Part 1 of manual, 5-6 for Part 2 of manual, 42 for Part 3 of manual, 56-58 summary of changes in, 3 Density, defined, 27 Destructive sampling/testing, and control chart method, 78 Dispersion and essential information, 40 measures of, 24-25 See also specific measures of dispersion Distributions. See frequency distributions E E^ defined, 116 E^ defined, 116 E[f(X)], defined, 6 Engineering data. See data Errors of measurement essential information for, 39 and frequency distributions, 7 type of data gathered for, 6-7 See also sampling errors Essential information, 39-42 defined, 39, 42 functions of data containing, 39-40 loss of, 40 presentation of, 40 and relationship between two variables, 40-42 expected value of a function, defined, 6
/(observed frequency), defined, 5 Factors for computing control chart lines approximations to, 116-117, 120 tables of, 113-115, 117 Fraction nonconforming, See p (sample fraction nonconforming); p' (population fraction nonconforming) Fraction rejected. Seep (sample fraction rejected) Frequency cell, defined, 15 observed, defined, 5 See alsop '(population relative frequency); p (sample relative frequency or proportion) Frequency distributions amount of information contained in, 26-35 bimodal, 23, 34 characteristics of, 22 and errors of measurement, 7 estimating observed percentages of, 34
129 INDEX
functions of, 22-25 graphic representation of, 14, 16-17 examples of, 14, 15, 17 objective, example of, 38 observed defined, 37 example of, 38 overview of, 6-7, 22 tabular, 14-17 unimodal, 22, 34 See also grouped frequency distributions; ungrouped frequency distributions Functions, expected value of, defined, 6 defined, 29-30 functions for representing, 30-37 See also data; essential information; observations Interpretation of data, factors affecting, 40-41
K k, defined for Part 2 of manual, 42 for P a r t 3 of manual, 57 /ti, 31-32 ku, 31-32 Kurtosis defined, 22 illustrated, 25-26 See also g^ (sample coefficient of kurtosis); Xj (population coefficient of kurtosis)
G gi (sample coefficient of skewness) amount of information contained in, 30-31, 33-35, 37 computation of, 28-29 defined, 5, 25-26 and essential information, 37 2 (sample coefficient of kurtosis) amount of information contained in, 30-31, 33-35, 37 computation of, 28-29 defined, 5, 26 and essential information, 37 %i (population coefficient of skewness), defined, 6 Xj (population coefficient of kurtosis), defined, 6 Geometric mean, defined, 24 "Go no-go" basis, a n d p (fraction nonconforming), 68 Graphic representation of frequency distributions, 16, 20-21 examples of, 17, 19, 21, 22 Grouped frequency distributions, 13-21 and cells boundary selection for, 15 defined, 14 number of, 15-16 and classifying observations, 16 cumulative frequency distributions, 16-17, 21, 22 defined, 14 graphic representation of, 16, 19, 20-21, 22 and large numbers of observations, 16, 22 and small numbers of observations, 16 "stem and l e a f diagrams, 21-22 tabular presentation of, 17-20
Lack of constancy, of cause system, detecting, 59 Large samples and computing X s, gi and gs, 28 control charts for X and s (standard given), 77 examples, 93, 94, 95 control charts for X and s (standard not given), 64-65, 67 examples, 82-85 and essential information, 39 and grouped frequency distributions, 5, 14, 20 recommendations for presentation of, 5 Leptokurtic frequency distribution, 23, 26 Limits, percentages of observations outside specified. See p (sample fraction nonconforming) Limits of uncertainty (for X), 42-43 cautions about using, 47-48 computation of, 42-45, 46-47 one-sided, 48 presentation of data for, 45-49 f o r p ' (population fraction nonconforming), 53, 54 for o (population standard deviation), 51-53 problem defined, 42-43 theoretical background for, 43 Lots, defined, 56
M H Homogeneous data, 8, 29 Homogeneous materials, and control chart method, 78, 79 Hypotheses, and interpretation of data, 39 Manual history of development of, 1-2 purpose of, 1 Mean (arithmetic). See (J, (population average or universe mean); X (sample arithmetic mean) Mean (geometric), defined, 23 Median, defined, 24 Mesokurtic frequency distribution, 23, 26 Minus or plus limits of uncertainty. See limits of uncertainty (for X) Mode, defined, 24 Moving ranges and control charts for individuals, 81-82 examples, 110-111 H (population average or universe mean)
Individuals, control charts for. See control chart method, for individuals Information prior, and interpretation of data, 39 relevant, presentation of, 39-40 total
130 INDEX
defined, 3, 6, 42, 58 and limits of uncertainty (for X ) , 42-44 standard value of, 58, 60, 79 N n (number of observed values) defined, 5, 42, 57 and essential information, 39-40, 41 and range (R), 40 Nonconformities defectives distinguished from, 66 statistical distributions involving, 119 See also c (number of nonconformities); p (sample fraction nonconforming); np (number of nonconforming units); u (nonconformities per unit); p' (population fraction nonconforming) Normal law, and "controlled conditions," 29 Normal population, and expected value of range (K), 3940 Normal variables, deviation of (k), 42 Notation. See symbols np (number of nonconforming units) control chart (standard given), 78-80 example, 94-97 control chart (standard not given), 66-72, 74 example, 82-84 defined, 58, 66 standard deviation (ap) of, 115 Number of observed values. See n (number of observed values) O o (value of standard deviation), defined, 42 Objective frequency distributions. See frequency distributions Observations amount of information contained in, 30-31 function for representing, 30-37 classifying into cells, 14, 17 number of {n) defined, 5, 42, 57 and essential information, 39-40, 41 and range (E), 40 order of, 58 and range (i?), 40 of relationships between two variables, 40-42 symbol for observed values {X), 6, 42, 58 See also data; information; large samples; small samples Observed average. See X (sample arithmetic mean) Observed frequency, defined, 5 Observed frequency distributions. See frequency distributions One-sided limits of uncertainty (X ), 48 "Ordered stem and leaf diagram, 21 Order of observations, 58 examples, 82-83 defined, 5, 27, 42, 58, 66 essential information for, 39, 40 standard deviation (a^) of, 58, 115, 116 standard value of, 58, 60, 79 p (sample fraction rejected), control charts (standard given), example, 93-95 p (sample relative frequency or proportion) amount of information contained in, 27 defined, 5, 27, 58 graphic representation of, 17, 27 P (probability), defined, 42 p' (population fraction nonconforming) defined, 42, 58 presenting limits of uncertainty for, 53-55 standard value of, 58, 60, 79 p ' (population relative frequency) defined, 6 standard value of, 58 Peaks multiple, 23, 38 single, 27, 38 Pearson curves, 29, 30, 31, 32 Pearson, Karl, 25, 29 Percentages of observations Percentages of observations outside specified limits. See p (sample fraction nonconforming) Physical data, examples of, 7, 8-9 Platykurtic frequency distribution, 23, 26 Plus or minus limits of uncertainty. See limits of uncertainty (for X) Poisson distribution, and nonconformities, 118 Population average. See ) (population average or X universe mean) Population coefficient of kurtosis. See ta (population coefficient of kurtosis) Population coefficient of skewness. See x^ (population coefficient of skewness) Population coefficient of variation. See V (population coefficient of variation) Population fraction nonconforming. See p' (population fraction nonconforming) Population relative frequency. Seep' (population relative frequency) Population standard deviation. See a (population standard deviation); a (estimated population standard deviation) Population variance. See (P (population variance) Position, defined and illustrated, 27 Presentation of data recommendations for essential information, 39-40 general, 5 limits of uncertainty (for X ), 45-52 relevant information, 40-41 Prior information, and interpretation of data, 40 "Probable error," defined, 43 Probable inference, and interpretation of data, 40 Q Quality characteristics average, 39 defined, 6
p (sample fraction nonconforming) control charts (standard given), 77, 78, 80 examples, 94-98 control charts (standard not given), 64-66, 74
131 INDEX
representation of, 6-7, 23 variability of, 39, 58-59 Quality control charts. See control chart method Quality control methods, types employed in manual, 1 Sample coefficient of kurtosis. See g^ (sample coefficient of kurtosis) Sample coefficient of skewness. See gj (sample coefficient of skewness) Sample coefficient of variation. See v (sample coefficient of variation) Sample fraction nonconforming. See p (sample fraction nonconforming) Sample fraction rejected. Seep (sample fraction rejected) Sample Sample Sample Sample mean. See X (sample arithmetic mean) range. See R (sample range) relative frequency density, defined, 24 relative frequency or proportion. See p (sample relative frequency or proportion) Sample standard deviation. See s (sample standard deviation) Sample variance. See s^ (sample variance) Samples defined, 56 random, 29 Sampling errors, 36 See also errors of measurement Sampling method, and interpretability of data, 38 Shewhart, W. A., 2, 29 Short method, for computing X , s, g^, and g^, 28-29 a (population standard deviation) defined, 3, 6, 58 presenting limits of uncertainty for, 53-54 standard value of, 58, 60, 61, 69, 112, 113, 115 a (estimated population standard deviation), defined, 42, 43 a ' . See a (population standard deviation) a^ (population variance), defined, 6 Skewness defined, 25-26 illustrated, 22 see also gi (sample coefficient of skewness); Xi (population coefficient of skewness) Small samples and classifying observations into cells, 17 control charts for X and R (standard given), 7678 example, 94, 111 control charts for X and R (standard not given), 63-65 examples, 81-82 control charts for X and s (standard given), 75, 76 examples, 93-96 control charts for X and s (standard not given), 63,65 examples, 79-81 and grouped frequency distributions, 17 Spread, defined and illustrated, 25 Standard deviations approximations to control chart factors for, 116118, 120 of sampling distributions, 58, 61, 114-115 See also s (sample standard deviation); a (population standard deviation); 6 (estimated population standard deviation) Standard values, 58, 60, 61, 74, 111-112, 117-118
R R (sample range) control charts (for individuals) using moving ranges, 81, 117-118 examples, 109-110 using rational subgroups, 59, 80, 81 example, 105-108 control charts (standard given) for I and R, 77-78 example, 94 control charts (standard not given) X and R (small samples), 63-65 examples, 81-82 defined, 6, 25, 39, 58 and size of n, 40 standard deviation (Oj) of, 58, 112 variance in expected value of, 39-40 Random samples, 29 Range. See R (sample range) Rational subgroups breaking up data into, 59 and control charts for individuals, 80-81, 83 examples, 106-109 defined, 56 size of, 60 Raw data, table of, 8-9 Rejections. Seep (sample fraction rejected) Relationship, between two variables, 40-41 Relative frequency for cells, 14 See also p (sample relative frequency or proportion); p' (population relative frequency) Relative frequency density, defined, 27 Relevant information, presentation of, 40-41 Root-mean-square deviation. See S(,,, (sample rootmean-square deviation)
s (sample standard deviation) amount of information contained in, 29-34, 35 coefficient of variation (v) vs., 33-34 computation of, 26-29, 58 control charts (standard given), 76-78 examples, 93-96 control charts (standard not given) for X and s (large samples), 64-65, 67 examples, 82-84 for X and s (small samples), 65, 67 examples, 83-85 defined, 2, 6, 25, 42, 58 and essential information, 35, 36, 37 formulas for, 25 standard deviation ( o j of, 58, 112, 115 s^ (sample variance), defined, 6, 24 Sample average. See X (sample arithmetic mean)
132 INDEX
summary of changes in notation for, 3 Standards, and control charts. See control chart method, standard given; control chart method, standard not given Statistical control and control chart method, 58-59, 61-62 defined, 43 Statistical variables, relationship between two, 40 "Stem and leaf diagrams, 20-21 Subgroups defined, 56 See also rational subgroups Supplement A (Part 2), 51-53 Supplement A (Part 3), 111-117 Supplement B (Part 2), 53-55 Supplement B (Part 3), 117-121 Symbols in Part 1 of manual, 5-6 in Part 2 of manual, 42 in Part 3 of manual, 56-58 summary of changes in, 3 table of, 8-9 Unimodal frequency distributions, 22, 38 Units, defined, 56 Universe mean. See n (population average or universe mean) Universe standard deviation. See a (population standard deviation)
Variability of quality causes of, 59-60 essential information for, 39, 42 Variables method of, and p (fraction nonconforming), 68 relationship between two, 40-42 Variance. See s^ (sample variance); a^ (population variance)
X
Tabular grouped frequency distributions, 14-17, 39 Tensile strength, presentation of data for, 36-38 3-sigma control limits. See control chart method, control limits Time, subgrouping with respect to, 63 Timeorder, 29 Tolerance limits, 33 Total information defined, 29-30 functions for representing, 30-39 Transverse strength, presentation of data for, 8-20 2-sigma control limits, 116 U u (nonconformities per unit) control chart (standard given), 79, 80 example, 110-111 control chart (standard not given), 71-73 examples, 91-94 defined, 58, 68 standard deviation (c!) of, 115 Uncertainty. See limits of uncertainty (for X) Ungrouped frequency distributions, 7-8 graphic representation of, 10 as representation of total information, 26
X (observed values), defined, 6, 42, 58 X (sample arithmetic mean) amount of information contained in, 32-37, 39 computation of, 26-29, 58 control charts (for individuals) using rational subgroups, 79 examples, 103-106 control charts (standard given) for X and R. 77, 78 example, 95, 98 for X and s, 75-77 examples, 91-96 control charts (standard not given) for X and R (small samples), 63-65 examples, 81-82 for X and s (large samples), 64-65, 67 examples, 82-84 for X and s (small samples), 65, 67 examples, 83-85 defined, 6, 23, 43, 58 and essential information, 36, 37, 38 standard deviation (Oj) of, 58, 115, 118 See also limits of uncertainty (for X) X See (i (population average or universe mean)

ASTM Data and Control

Uploaded by

Copyright:

Available Formats

ASTM Data and Control

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

ASTM Data and Control

Uploaded by

Copyright:

Available Formats

What is the purpose of this manual?

What is the purpose of this manual?

What types of data are considered in this manual?

What types of data are considered in this manual?

Manual on Presentation of Data and Control Chart Analysis

COMMITTEE E-11 ON QUALITY AND STATISTICS

Ittirary at Congress Cataloging-in-Publication Data

Printed in Bridgeport, NJ February 2002

30. Control Chart for Individuals, XUsing Ranges

103 105 106

DATA AND CONTROL CHART ANALYSIS

(f, p', u', c'

( = universe values) |io, Oo, po, Uo, C o

Xg,Oo', po', Uo', Co'

DATA AND CONTROL CHART ANALYSTS

GLOSSARY OF SYMBOLS U S E D PARTI

DATA AND CONTROL CHART ANALYSTS

Firsi Type n Chvobsmafm l/iMffs mtachfhing V

Second Type One n Observaiions thing

FIG. 1Two general types of data.

2. Type of Data Considered

8 DATA AND CONTROL CHARTANALYSIS

TABLE 1. Three groups of original data.

900 890 970

700 920 860 950

800 990 870 660

990 880 750 970

780 760 910 990 870

910 870 810

960 870 910

980 830 460

800 710 880

10 DATA AND CONTROL CHART ANALYSIS

1000 Transverse Strength, psi.

UNGROUPED WHOLE NUMBER DISTRIBUTION 5. Ungrouped Distribution

6. Empirical Percentiles and Order Statistics

(a) Transverse Strength, psi (data of Table 1 (a))

12 DATA AND CONTROL CHART ANALYSTS

GROUPED FREQUENCY DISTRIBUTIONS

14 DATA AND CONTROL CHART ANALYSTS

10. Number of Bins

11. Rules for Constructing Bins

9. Choice of Bin Boundaries

' Excel is a trademark of Microsoft Corporation.

Total Number of observations = 270

NOTE"Number of observations" should be recorded with tables of relative frequencies.

Using 12 cells, (Table III [a])

Fig. 4Illustrating increased irregularity with larger number of cells, or bins.

12. Tabular Presentation

1125 to 1275 to 1425 to 1575 to

1725 to 1875 1875 to 2025

18 DATA AND CONTROL CHART ANALYSIS

13. Graphical Presentation

80 60 40 20 0 80 60 40 20 0 500 30 20 10 Frequency - Histogram 0 (Columns erected on cells) 30 20

Transverse Strength, psi. Frequency I I I 11 6l38l80l83l39ll7l2 I 2 lo I 1 I C e l l B o u n d r i e s S ^ g g m ^ | |

14. Cumulative Frequency Distribution

20 DATA AND CONTROL CHART ANALYSIS

15. "Stem and Leaf Diagram

The 'TJOX" is formed by the 2 5 * and 75'*^

1.323 1.4678 1.540 1.6030