Work Measurement: New Principles and Procedures 9780231899864

Table of contents :
Preface
Contents
Tables
Figures
1. Introduction: Industrial Experimental Studies
2. Designing a Research Program on Industrial Productivity
3. The Problem of Process Standardization
4. Case Studies of Local Stability
5. Studies of Local Stability: Implications
6. Case Studies of Grand Stability
7. Studies of Grand Stability: Implications
8. Developing Standard Data for Operation Elements and Motions
9. The Relationships among Operation Elements
10. Comparing the Work Methods of Different Operators
11. The Implications of the Likelihood Ratio Results
12. Motion Standard Data and Related Questions
13. Obtaining Useful Measurements and Estimates
14. Measuring and Estimating Delay Factors
Bibliography
Index

Citation preview

Work Measurement

Work Measurement NEW PRINCIPLES AND PROCEDURES

By Adam Abruzzi ASSISTANT PKOFE8SOB STEVEN8 INSTITUTE OF TECHNOLOGY

COLUMBIA UNIVERSITY PRESS NEW Y O R K , MORNING SIDE HEIGHTS

1952

COPYRIGHT COLUMBIA

UNIVERSITY

1952

PRESS,

PUBLISHED IN GREAT BRITAIN, BY

GEOFFREY

CUMBERLEGE,

NEW

YORK

CANADA, AND I N D I A

OXFORD UNIVERSITY

LONDON, T O R O N T O ,

AND

PRESS

BOMBAY

MANUFACTURED IN T H E UNITED STATES OF AMERICA

TO HAZEL W. ABRUZZI

Preface I T will soon become apparent to the reader that most of the procedures now used for work measurement do not perform their intended functions. Yet nothing of any consequence has been done to correct the situation. Of inventions there have been many, but almost all of them have been concerned with procedural minutiae which the authors apparently hoped would give them a wide audience and clientele. This has led to a large number of widely advertised prescriptions for the measurement of work. It turns out that the differences among the prescriptions resolve themselves into quarrels about who should treat the patient; nobody seems to be very much concerned about the fact that the treatment may be fatal. In our day it has become fashionable to use the label "scientific" to help advertise one's notions. True to fashion, those who write the prescriptions on work measurement continually remind us that theirs is the only truly scientific prescription. Curiously enough, none of them has taken the trouble to describe the scientific method or to show how it is related to what they are offering in its name. One also hears a great deal about the fact that the present procedures "work" in practice. The trouble with this notion is that what "works" here is what management and labor decide will be allowed to work or will be made to work. This does not make the procedures scientific, nor does it mean that other procedures will not work much better. This, I think, is the essence of the matter. We have two—and only two—alternatives open to us. On the one hand, we may continue to use the present procedures; they will continue to "work" in the same sense that collective bargaining "works."

viii

Preface

This book shows, in fact, that these procedures are intimately associated with the bargaining process. The other alternative is to adopt the viewpoint taken in this book. According to this viewpoint the procedures of work measurement should be based on the rules of the scientific method, not on the bargaining process. It seems to me that we can no longer tolerate taking empirical action about productivity problems on the basis of the subjective judgments of individuals acting as advocates of vested viewpoints. Instead, we need objective principles and procedures so that the estimates we make and the action we take will be sound in a scientific sense. This book is intended to supply such principles and procedures. I should like to acknowledge the generosity of the following publishers for permission to quote from their copyrighted works as indicated in the text: D. Van Nostrand Company; United States Department of Agriculture Graduate School; AppletonCentury-Crofts, Inc.; and The Controller of H. M. Stationery Office. Acknowledgment is also made to the following journals which have published or are about to publish articles on some of the material contained in the book: Advanced Management; Industrial arid Labor Relations Review; Mechanical Engineering; and Personnel. I should also like to express my sincere appreciation to Dr. William Gomberg for his invaluable assistance and counsel. I am also extremely grateful to his associates, Mr. Louis Rolnick and Mrs. Nancy Rolnick. They were extremely helpful in making some of the studies, as were Mr. E. Rosa, Mr. C. Dobula, Mr. P. Flatow, and Dr. P. Koditschek. I should also like to thank Professor Robert Bechhofer, of Columbia University, for a number of helpful suggestions regarding the statistical treatment of some of the data. Unfortunately, it is impossible to identify here any of the many people in the two plants who made possible most of the studies described in the book. I hope they will accept this general acknowledgment of their kindness and interest; I am especially grateful to the workers who acted as willing and friendly study-subjects.

ix

Preface

Without question, my greatest debt is to Hazel W. Abruzzi, who spent many tedious hours with computation and typing chores. ADAM ABRUZZI

Castle Point, Hoboken, N.J. July, 1951

Contents 1. INTRODUCTION: INDUSTRIAL EXPERIMENTAL STUDIES 2 . D E S I G N I N G A R E S E A R C H P R O G R A M ON

3

INDUSTRIAL

PRODUCTIVITY

11

3 . T H E P R O B L E M O F P R O C E S S STANDARDIZATION

30

4 . C A S E STUDIES OP LOCAL STABILITY

39

5 . S T U D I E S OF L O C A L S T A B I L I T Y : I M P L I C A T I O N S

62

6 . C A S E STUDIES OF G R A N D STABILITY

75

7 . S T U D I E S OF G R A N D S T A B I L I T Y : I M P L I C A T I O N S 8. DEVELOPING

STANDARD

DATA

FOR

OPERATION

106 ELE-

MENTS AND M O T I O N S

120

9 . T H E R E L A T I O N S H I P S AMONG O P E R A T I O N E L E M E N T S 1 0 . COMPARING THE W O R K M E T H O D S OF

129

DIFFERENT

OPERATORS

157

1 1 . T H E IMPLICATIONS O F T H E L I K E L I H O O D R A T I O R E S U L T S

180

1 2 . M O T I O N STANDARD D A T A AND R E L A T E D Q U E S T I O N S

196

1 3 . O B T A I N I N G U S E F U L M E A S U R E M E N T S AND E S T I M A T E S

220

1 4 . M E A S U R I N G AND E S T I M A T I N G D E L A Y F A C T O R S

246

BIBLIOGRAPHY

273

INDEX

281

Tables 1. Stop-watch Readings Obtained from Operation 1 2. Ratio-test Results on the Data for Operator ID 3. The Production-rate Estimates Based on Studies Made on Operator 13^4. 4. The Production-rate Estimates Based on Operations in the Two Plants 5. The Relation between Skill and Uniformity 6. The Production-rate Estimates Obtained from Three Inexperienced Workers 7. The Relation between Delays and Net Production Rates 8. Data Showing How a Reduction in the Mean Cycle Time Affects Delays 9. The Period Breakdown Used in the Studies of Grand Stability 10. Ratio-test Results on the Daily Means for Operation 1 11. The Additive-test Data on Operation 1 12. Test Results on the Period Means for Operation 1 13. The Analysis-of-variance Data on Garment Sizes 34, 36, and 38 (Operation 1) 14. Test Results on the Daily Means for Operation 4 15. The Analysis-of-variance Data on Garment Sizes 32, 34, and 36 (Operators 4A and 4C) 16. Test Results on the Daily Means for Operation 6 17. The Analysis-of-variance Data on Garment Sizes 32, 34, and 36 (Operation 6) 18. Test Results on the Daily Means for Operation 7 19. Test Results on the Pooled Daily Means of the Seven Operations

39 44 57 63 64 65 66 68 78 85 87 90 91 98 100 102 103 104 105

xiv

Tables

20. The Relation between the Level and the Uniformity of Grand Production Rates 21. The Relation between the Level and the Uniformity of Local Production Rates 22. The Relation between the Mean Cycle Time and the Relative Aggregate Efficiency 23. Criteria Reported for Defining Operation Elements 24. The Independence-test Matrices from the Element Data of Operator ID 25. The Reduced Numerator Matrix from the Data of Operator ID 26. The Independence-test Matrices from the Element Data of Operator 1H 27. The Numerator Matrix from the Element Data of Operator 1 *A 28. The Independence-test Data from Studies Made on Operation 13 29. The Independence-test Data from the Study Made on Operator 21A 30. The Independence-test Data from Other Studies Made in Plants A and B 31. The Independence-test Matrices from the Grouped Data of Operator ID 32. Independence-test Results from the Grouped Data of Four Operators on Operation 3 33. Independence-test Results from the Second Grouping of the Data of Operators 3N and 3F 34. Independence-test Results from the Grouped Data of Four Studies on Operation 13 35. Independence-test Results from the Second Grouping of the Data of Four Studies on Operation 13 36. Independence-test Results from the Grouped D a t a of Three Studies Made in Plant B 37. The Summarized Data for the Cases in Which Independence Was Not Established 38. The Summarized Data for the Cases in Which Independence Was Established

109 110 111 121 134 135 137 139 141 142 143 149 150 151 152 152 153 154 155

Tables 39. The Denominator Matrix in the L(yc) Test on the Element Data of Operators 21 and 2F 40. The Relation between the Mean and the Degree of Variability of Element-group Times 41. Element Data of Operator 3H Showing That the Relative Degree of Variability Increases as the Mean Time Decreases 42. The Motion Descriptions and Times for the First Two Cycles in Motion Study A 43. The Wink-counter Readings in Motion Study A 44. The Mean Values of the Motion and Motion-group Data in Motion Study C 45. Data from the Three Motion Studies Showing the Relation between the Mean and the Degree of Variability 46. Data on the Snap-back and Continuous Stop-watch Methods 47. Data on the Film Wink-counter and Stop-watch Methods 48. Data on the Three Measurement Methods Considered 49. Factory Data on the Stop-watch and Marsto-chron Methods 50. Factory Data on the Stop-watch and Marsto-chron Methods, Using a Simplified Element Structure 51. The Data from an Experiment by Three Stop-watch Observers 52. The Period Breakdown Used in the Delay Study in Plant B 53. The Period Breakdown Used in the Delay Study in Plant A 54. A Comparison of the Delay Estimates and the Allowances in Plant A 55. The Analysis-of-variance Data on the Period Delay Percentages in Plant A

xv 174 180 193 198 200 203 210 230 236 238 241 242 243 259 264 267 268

Figures 1. The Percentage of Operations Standardized in Advance by Survey Respondents 2. A Comparison of the Relative Output under Three Types of Wage Payment Plans 3. The Mean and Range Charts on Local Cycle Times for Operator ID 4. The Mean and Range Charts on Local Times for Elements 1 and 2 for Operator ID 5. The Mean and Range Charts on Local Cycle Times for Operator 1H 6. The Mean and Range Charts on Local Cycle Times for Operator I*A, Showing the Separate and Combined X , R, and Control Limit Values 7. The Mean and Range Charts on Local Cycle Times for Operator 27 8. The Mean and Range Charts on Local Cycle Times for Operator 11.A, Showing the Separate and Combined X, R, and Control Limit Values 9. The Mean and Range Charts on Local Cycle Times for Operator 134—First Study 10. The Mean and Range Charts on Local Cycle Times for Operation 21 11. The Mean and Range Charts on Local Cycle Times for Operation 22 12. The Group Mean and Range Charts for Operation 1, Showing Both the Outer 2s* and the Inner 3-sigma Control Limits for the Means 13. The Mean and Range Charts on the Data for Operation 1 Arranged According to the Operator

31 34 42 44 45

50 51

54 56 59 60

80 88

xviii

Figures

14. The Mean and Range Charts on the Data for Operation 1 Arranged According to the Work Period, Showing Both the Outer 2s* and the Inner 3-sigma Control Limits for the Means 89 15. The Group Mean and Range Charts for Operation 2, Showing Both the Outer 2s* and the Inner 3-sigma Control Limits for the Means 92 16. The Mean and Range Charts on the Data for Operation 2 Arranged According to the Operator 93 17. The Mean and Range Charts on the Data for Operation 2 Arranged According to the Work Period, Showing Both the Outer 2s* and the Inner 3-sigma Control Limits for the Means 94 18. The Group Mean and Range Charts for Operation 3, Showing Both the Outer 2s* and the Inner 3-sigma Control Limits for the Means 95 19. The Group Mean and Range Charts for Operation 4, Showing Both the Outer 2s* and the Inner 3-sigma Control Limits for the Means 97 20. The Pooled Means of Operators 4A and 4C Arranged According to the Garment Size 99 21. The Mean and Range Charts for Operation 6, Showing Both the Outer 3s* and the Inner 3-sigma Control Limits for the Means 102 22. Chart for Estimating the Sample Sizes Required to Obtain Maximum Confidence Intervals of ± 5 percent for Given Coefficient of Variation Values 184 23. Control Charts on the Total Delay Percentages Obtained in Plant B, Plotted by the Period and by the Day 262 24. Control Chart on the Total Delay Percentages Obtained in Plant A, Plotted by the Period 265 25. Control Charts on the Unavoidable and Personal Delay Percentages Obtained in Plant A, Plotted by the Period 266

Work Measurement

1. Introduction: Industrial Experimental Studies T H E problem of making industrial experimental studies has been examined by a fairly large number of writers. Among these is Luther Gulick, who writes that both dynamic and static analyses are required to classify the variables involved according to their importance.1 Extensive documentation is also required to obtain generalizations and hypotheses within a single system of definition. James Gillespie, another student of this question, presents a series of descriptive rules for the application of the scientific method to this field.2 According to him, science consists of the ordered knowledge of natural phenomena and their relations. This knowledge depends on the use of the rational approach and "reflective thinking" as opposed to rationalization and snap judgments. In doing scientific work the first step should be a clear statement of the objectives; the problem is then subdivided into its component parts. The observations themselves should be as free as possible from bias and prejudice, and all hypotheses developed from the observations should be verified by direct experiment. T. N. Whitehead, who analyzed the data obtained in the widely publicized Western Electric experiments, considers that a scientific investigation should begin with a careful definition of the problem. Primitively, this definition describes "an area of comparative ignorance within which the investigator wishes to exercise increased discrimination." 1 In the specific field of in1 Gulick, "Science, Values and Public Administration," in Papera on the Science of Administration, chap. xi. * Gillespie, The Principles of Rational Industrial Management. » Whitehead, The Industrial Worker, I, xi, 99.

4

Introduction

dustrial productivity, Whitehead adds, "men and women must be examined under conditions which are sufficiently typical of their daily experiences, and yet which permit of an orderly investigation." Shewhart's Views and Results.—Walter Shewhart's views deserve special attention here, for he is probably the outstanding contemporary student of this subject.4 Applied science, he says, has even more exacting requirements than has pure science. This is because the applied scientist must take direct action on the basis of his results, often with important economic consequences. "Invariably," Shewhart adds, "each practical rule of action, so far as it has been adopted as a result of reasoning, is based upon some abstract concept or group of concepts." Before it can finally be accepted, however, it must be verified experimentally. In fact, scientific statements have a definite meaning only if they can be verified at least theoretically by experimental measurements; the process of verification also requires that the method of measurement be completely specified in advance. Shewhart also discusses the fundamental problem of how far formal statistical methods can be used in applied science. Such applications, he insists, cannot be made unless empirical criteria are available for deciding when a set of observed data constitutes a random sample. This statement strikes at the heart of the whole question of making experimental inferences. It means that a profound difference exists between formal theory and empirical applications of that theory. In theoretical work it is sufficient to assume randomness and proceed to develop estimates and test procedures. In empirical applications the problems are much more difficult, for empirical criteria must first be established for deciding when a sample can be considered a random sample. It is only when these criteria are satisfied that the formal theory can safely be applied. In Shewhart's view all scientific estimates must be accompanied by an explicit statement about their precision. As an example he cites Planck's constant from physics. The value of this constant is given in terms of a range of variation, within 4

Shewhart, Statistical Method from the Viewpoint of Quality Control.

Introduction

5

which future observations are expected to lie if they are made under the same conditions. Shewhart then turns to the question of verification and finds that it is made up of two distinct components: practical verification and theoretical verification. Practical verification is obtainable in terms of experimental observations that are always finite in number. Theoretical verification, however, can be conceived formally only as the result theoretically attainable from an infinite number of observations. If a scientific statement is to be practically verifiable, the number of observations must be specified, as well as the function (or functions) of those observations. This is not enough, however. It is also necessary to specify the interval within which the functional value must he in order that the statement be deemed true. PRINCIPLES OF MODERN EXPERIMENTAL INFERENCE

Gulick, Gillespie, Whitehead, and most other writers on this subject provide only descriptive procedures for making industrial experimental studies. The single major exception is the small group of writers on the special subjects of product quality and technological research, of whom Shewhart is the leading example. The principal objection to most of the procedures offered is that they fail to take into account the powerful principles of experimental inference developed in recent years. These principles are described in detail by a number of writers on the philosophy of science, such as C. West Churchman.6 Churchman supplies the following series of scientific axioms: (1) "the purpose of all scientific activity, taken collectively, is to reduce error to zero, i.e., to become absolutely precise"; (2) a limited number of observations is never sufficient to give an answer to a scientific question, since there always exists an error of observation; (3) specific methods exist for determining the magnitude of the error of observation; (4) a zero error means that a complete answer has been obtained. Such a complete answer can only be conceived in formal terms as the end-product of a never1 Churchman, Theory of Experimental Inference and "Probability Theory I, II, III."

6

Introduction

ending series of successive approximations; (5) a complete formal theory of statistics is necessary for indefinite progress in science, although it is not sufficient; (6) the unified action of students in all the disciplines concerned will be required to obtain approximate solutions to scientific problems. Morris R. Cohen and Ernest Nagel have also given extensive treatment to the problems of experimental inference.6 Like Whitehead, they consider scientific investigations to begin with a careful definition of the problem. Once the problem has been defined, irrelevant material must be separated from the relevant, which must then be examined in an exploratory fashion to develop tentative hypotheses. All hypotheses should be capable of predicting in some sense what will happen, which is the same as saying that they should be verifiable. Whenever a choice remains between two or more hypotheses, the process of experimentation must be continued until a single hypothesis is accepted. At first, hypotheses can be framed only in terms of qualitative differences. In the interest of precision, it is essential to indicate as clearly as possible the degree of these differences. These hypotheses can gradually be expressed in quantitative terms; ultimately, this process will yield acceptable hypotheses specifying the numerical values of the variable considered on a continuous scale. Cohen and Nagel, Churchman, and others also reject the classical notion that scientists are interested primarily in an abstract search for truth, independent of social values. Frank Hartung makes this point in more positive terms.7 He says that the very decision to follow scientific procedures involves making a value judgment. Value judgments also enter into scientific work in other crucial ways, for, as Hartung points out, it must always be decided which problems should be worked on and which goals are desirable. More Specific Problems.—First it must be decided which procedure will insure valid results. The modern viewpoint on this key problem is well represented by the comments of Norbert Wiener and Arturo Rosenblueth. 8 Their main thesis is that empirical • Cohen and Nagel, An Introduction to Logic and Scientific Method. 1 Hartung, "On the Contribution of Sociology to the Physical Sciences." • Wiener and Rosenblueth, "The Role of Models in Science."

Introduction

7

phenomena can be understood only by making use of models. These models are of two kinds: the formal model and the material model. These two kinds of models are closely related, for all material models are ultimately based on formal models. These models are also related in that a complex system is represented by a simpler system assumed to have similar properties. Wiener and Rosenblueth also make an instructive analogy between models and "closed boxes." Only a limited number of variables involved in a complex system can be put into one "closed box" at one time. Scientific progress can be considered a successive opening of these boxes, which include more and more variables. Finally, the boxes (and the models) can be made to include as many variables of the original complex system as is desired. In general, formal models have a mathematical structure. The problem of applying formal models in empirical science is therefore equivalent to the problem of applying mathematical models. The latter problem has been considered by Eugen Altschul and Erwin Biser; their starting point is the classical belief that a science is exact only to the extent to which mathematical equations can be applied to its subject matter.9 In recent years, they assert, this belief has largely been rejected, even in the science of physics, where it originated. The essential point is that a mathematical model cannot completely encompass the variables in a given empirical situation; at best, such a model can handle only a selected number of these variables. Experimental studies can scarcely be carried on without solving the problem of designing suitable scales of measurement. According to S. S. Stevens, the process of measurement can be defined "as the assignment of numerals to objects or events according to rules." 10 Scales of measurement can be constructed only when a unique correspondence has been established between certain characteristics of these objects or events and the properties of the number system. The practical value of establishing such a correspondence lies in the fact that the number system can then be used to represent the result of certain em9

Altschul and Biser, "The Validity of Unique Mathematical Models in Science." Stevens, "On the Theory of Scales of Measurement."

10

8

Introduction

pirical operations and hence to simplify the problem of obtaining definitive information. Modern scientists are becoming increasingly aware of the need for interdisciplinary cooperation for optimum results in experimental work. This viewpoint is given specific attention by Paul Olmstead, who says that the natural scientist needs to know more about what social scientists have done about measuring "intangibles." 11 The natural scientist also needs and should seek the cooperation of social scientists, for the human element is always present to some degree in his work. SPECIFICATIONS FOR MAKING INDUSTRIAL EXPERIMENTS

The preceding sections emphasize the need for a series of specifications in industrial experimental work based on the modern theory of experimental inference. A series of such specifications is given below. In the following chapters these specifications are used in two ways: (1) to determine the scientific validity of current time-study procedures and (2) to plan and analyze the experimental studies considered. To do effective industrial experimental work, it is first necessary to define the fundamental concepts insofar as possible in terms of empirical operations (that is, operationally). Meeting this requirement is especially important in the field of industrial productivity, where many fundamental concepts are defined in a nonoperational manner and are subject to a variety of interpretations. After operational definitions have been established, the broad objectives of the investigation should be clearly stated. In this statement a sharp distinction should be made among economic, sociological, physiological, psychological, and technological objectives. The specific objectives of the investigation should also be clearly stated. Thus, the required order of accuracy and of precision should be specified in estimation problems. Similarly, the risks to be tolerated with regard to making wrong decisions should be specified in making tests of hypotheses. 11 Olmstead, "Some Thoughts on 'What the Natural Scientist Needs from the Social Scientist.1 ''

Introduction

9

Wherever necessary, the investigation should include a preliminary phase for the purpose of designing appropriate scales and methods of measurement. These scales and methods of measurement should be capable of meeting the accuracy and precision requirements of the main problem to be investigated. A list of rules should also be developed, specifying the empirical operations necessary for making observations under the proposed scales and methods of measurements. The main investigation should be based on the formal model whose assumptions are best satisfied by the empirical conditions of the immediate problem. The main investigation should also be carefully designed so as to insure that optimum results are obtained at a minimum cost. This implies that the estimating or testing procedure must be decided in advance, together with such essential details as the sampling procedure and the sample size. Investigations on industrial problems are of three principal types. The first type includes laboratory studies of technological problems, such as evaluating the comparative advantages of different measuring instruments. The second type includes situational studies made under conditions characteristic of the problem in question; situational studies should be made on problems whose results might be substantially distorted if they were made under laboratory conditions. The third type of investigation combines both the preceding types: the preliminary work to be done under laboratory conditions and the confirmatory work under actual operating conditions. Here the laboratory serves as an economical agent for formulating preliminary estimates or hypotheses which can then be tested situationally. The estimates obtained in an investigation should include a statement specifying how they can be verified by other observers. This means that the verification process requires the estimating procedure to be independent of the observer insofar as possible. The verification process also requires that the main statistical properties of the estimates be presented so that the results obtained by different observers can be compared. The most important statistical property of an estimate is precision, which varies inversely with the variability of the observed

10

Introduction

data. It should be reported because it specifies the range of future observations (with the same measurement method). The range is defined in terms of some predetermined probability value and is established on the basis of the mean value (or some other statistic) computed from the observed data. A high degree of precision is no guarantee that the estimate is correct or accurate. The accuracy of a given estimate can be determined only by comparing it with the estimate obtained by the standard measurement method. Like its precision, the accuracy of an estimate specifies the range of future observations again at some predetermined probability value. The essential difference is that the range in accuracy statements is based upon the "true" value, estimated as the mean value, let us say, of the data obtained by the standard measurement method. The results obtained from tests of hypotheses should be presented in operational terms, thus providing other observers with the empirical means for verifying and extending the results of these tests. Each test result should also be accompanied by a statement of the probability of rejecting a true hypothesis or, alternatively, of accepting a false hypothesis. The assumptions of a given formal model are never fully satisfied in applications. This implies that the numerical probability values attached, for example, to some specific test result should not be interpreted literally, but only as indicating a certain order of probability. Moreover, a statistically significant result is not sufficient in itself for empirical purposes. The decision whether or not to act on such a result must be based instead on empirical factors, such as the economic consequences of that result. One of the principal goals of experimental investigations is to obtain results that can be extended and refined in future investigations. To attain this goal, it will ultimately be necessary to obtain the cooperation of students in all the disciplines related to the problems being investigated. Such cooperative action is also necessary to insure that every empirical aspect of these problems is taken into account in planning the investigations.

2. Designing a Research Program on Industrial Productivity A D E S C R I P T I O N of the design and structure of the research program considered in this book serves a number of important purposes, the first of which is to explain why research work is required in this field. Its other main functions are: (1) to show why and how the program was planned so as to yield concepts and procedures that would be generally useful in all types of industry; (2) to present the underlying assumptions that were made and the reasons for making them; (3) to bring out the fundamental viewpoint and objectives adopted in making the studies and the reasons for their adoption; (4) to outline the scope and limitations of the results obtained. Such a description serves still another important purpose. It provides an instructive model for other investigators to follow in planning further research programs in specific industries. It should be emphasized, however, that the model must not be followed literally, for in order to be successful research work must be designed for the peculiar characteristics of each plant and industry. THE RESEARCH ADVANTAGES OF MAN-CONTROLLED OPERATIONS

The Degree of Variability.—Most observers familiar with this subject would probably agree that man-controlled operations have a greater degree of variability (or variation) than any other type of operation. This intuitive conclusion has been satisfactorily corroborated by experimental studies. S. Wyatt and P. M. Elton, for example, arrive at the following specific conclusion: the degree of variation in production rates reflects the relative influence of the machine and the worker on those rates.1 1

Elton, An Analysis of the Individual Difference» in the Output of Silk-Weavers; Wyatt, "Some Personal Factors in Industrial Efficiency."

12

Designing a Research Program

These students also show that the degree of variation is greatest in man-controlled operations. The explanation is simple. In such operations production rates are heavily influenced by the sociological, the physiological, and the psychological variables introduced by the worker. In machine-controlled operations, however, the situation is quite different. The influence of these human variables is sharply reduced, for the production rates are now dominated by the mechanical variables introduced by the machine. This fact does not in itself explain why the degree of variation is also sharply reduced. The explanation is that machines are always constructed so as to have as small a range of mechanical variation as possible. The net result, then, is that machine-controlled operations have a much smaller degree of variation than do mancontrolled operations. This statement can readily be extended to cover other types of operation, all falling between these two extremes. Here the degree of variation falls between the minimum degree of variation of purely machine-controlled operations and the maximum degree of variation of entirely man-controlled operations. Implications of These Facts.—In statistical terms, the facts presented above mean that man-controlled operations provide a maximum opportunity for the expression of variation in production rates. The over-all variation in this case would thus be expected to include trends, periodicities, and other kinds of nonrandom variation, some of which would not turn up when the worker does not dominate the production rate. The empirical implications of this are of the utmost importance. The key point is that all other types of operation would exhibit some, but not all, kinds of variation found in man-controlled operations. From the statistical viewpoint, then, they can be considered special cases of man-controlled operations. The result is that any statistical procedures developed for handling the variations found in man-controlled operations should also be directly useful for handling the variations found in other types of operation. On the other hand, statistical procedures developed for, let us say, operations controlled partially by

Designing a Research Program

13

machines would not necessarily be applicable to man-controlled operations. The preceding argument shows that the basic statistical concepts and procedures developed from studies of man-controlled operations should be directly applicable to all other types of operation. This does not mean, however, that the specific numerical details of these procedures can be applied to other types of operation without modification. For example, in Chapters 4 and 6 specific numerical criteria of statistical stability are developed for operations in the garment industry. These criteria are based upon the statistical characteristics of these operations and the empirical requirements of the industry. The same criteria would not apply to operations in other industries unless they had similar statistical properties and empirical requirements. However, the basic concepts and procedures from which the criteria were developed can be applied directly. If they are applied correctly, they will yield numerical criteria tailored to fit the peculiar characteristics and requirements of the immediate plant and industry. SELECTING THE INDUSTRY AND PLANTS

Selecting the Plants and Operations.—The type of operation is first to be considered in choosing an industry for study. In the present case the ladies' garment industry was selected, for its operations are clearly of the man-controlled type. In fact, the function of most of the machinery used in pressing and sewing operations—which are given particular attention here—is to provide power and mechanical advantage to the workers. Except for this, workers engaged in these operations exercise complete control over their production rates. Another advantage of this industry is that it has conveniently located plants with widely different degrees of technological development and management control. This greatly simplified the problem of investigating the effect of these factors on production rates. In one of the two plants selected the production process was organized and regulated in a haphazard and unsystematic man-

14

Designing a Research Program

ner. In addition, modern methods of performing management functions, such as cost control procedures, were generally lacking. The other plant selected represented the opposite extreme in technological development and management control. For example, it had a comprehensive program for selecting workers and training them along with a complete production-control program. It had also adopted most other modern methods of organizing and regulating the production process and of performing management functions. Another basic point of difference between the two plants was the way production-rate and wage-payment problems were handled. In the first plant a piece-rate-incentive plan was in effect on the sewing operations. The unit rates of payment were based on collective bargaining agreements rather than on time-study results. On the pressing operations, however, fixed rates of wage payment were in effect, again based on collective bargaining agreements. In the second plant standard production rates were developed for all operations on the basis of time-study results. These standards were used as a base for the wage-incentive plan, which provided that the workers attain these standards to be eligible for extra earnings. This plant also made extensive applications of motion-study procedures to develop improved work methods for the guidance of the workers. The Problem of Cooperation.—Many students have emphasized the need for cooperative action on the part of management and labor in doing fundamental research on productivity problems.1 Here, again, the garment industry proved to be an excellent choice. The International Ladies' Garment Workers' Union (ILGWU), which represents the workers in this industry, enjoys especially cordial relations with the managements of these plants. This enabled the union—through William Gomberg, its director of management engineering—to enlist their active support in the research program; the union also directly solicited the coopera* See, for example, Miles, The Problem of Incentives in Industry; Hall and Locke, Incentives and Contentment; Golden and Ruttenberg, The Dynamics of Industrial Democracy.

Designing a Research Program

15

tion of the stewards and the workers involved. As a close student of the subject, Gomberg himself became actively interested in the program and provided direct assistance in the form of union equipment and data. The plant managers instructed their foremen and technical personnel to cooperate with the program as fully as possible. In fact, they went so far as to rearrange certain work schedules to fit the requirements of the experimental studies. In the second plant time-study records and other technical information collected by its Methods Department were also made available. The result was that management and labor at all levels rendered invaluable assistance in making the experimental studies possible. This assistance was undoubtedly largely responsible for the successful completion of the program. Even under these enviable circumstances it was necessary to spend a great deal of time explaining the nature and objectives of specific studies to the workers. Careful explanations soon satisfied them that the studies had a research objective and would affect neither their production requirements nor their earnings, but created another problem of some consequence, for some workers became enthusiastic about their role as participants in a research program and consequently worked faster than usual. To counterbalance this tendency to be overzealous, it became necessary to impress upon these workers that unbiased performances were required. PRODUCTIVITY STUDIES AND WAGE-PAYMENT RATES

Many critics, notably Eric Farmer, suggest that the function of estimating production rates should be separated from the function of setting rates of wage payment.3 The essence of their argument is that otherwise it is impossible to determine the validity of the estimating procedures. Essentially the same viewpoint is also taken by certain timestudy writers. J. Keith Louden, for example, points out that determining what constitutes a fair day's work must be kept • Farmer, Time and Motion Study. See also Balderston, Group Incentives; Kennedy, Union Policy and Incentive Wage Methods.

16

Designing a Research Program

entirely separate from determining how much shall be paid for that work.4 Ralph Presgrave, another authoritative writer on the subject, argues in a similar fashion. Even if the final standard is to be a piece rate of payment, the time-study analyst is primarily interested in "arriving at a correct time standard only. The amount to be paid is settled entirely outside of time study through bargaining, tradition, or supply and demand." 5 Strong experimental evidence in support of these views is presented by S. Wyatt, F. G. L. Stock, and L. Frost.6 They show that production rates become stabilized at a level that depends on the strength of the incentive plan. In essence, this is what C. A. Mace had in mind in concluding that the most potent factor in worker performance is the wage-payment plan.7 Worker Attitudes and Behavior.—Some of the time-study writers who advocate separating the two functions do not rule out the use of production estimates as a basis for establishing wagepayment rates. They believe that the two functions can be separated simply by forbidding the time-study analyst to set wagepayment rates. Abundant evidence exists, however, that these functions must be separated much more completely. Much of this evidence is offered by Stanley Mathewson, Fritz Roethlisberger, and Burleigh Gardner.8 They point out that workers adopt protective and restrictive practices whenever they feel that wage-payment rates depend on production rates. As an example, workers protect lenient standards of production and lenient wage-payment rates by regulating their output so that it has a fixed relationship to the standards. Marvin Mundel, another leading time-study writer, also provides direct evidence on these questions. "Many standard times function well," he says, "even though they are incorrect, because the workers have learned that it is advantageous to have them function." 9 Thus, workers vary their exertion on different jobs 4

Louden, "Management's Search for Precision in Measuring a Fair Day's Work." • Presgrave, The Dynamics of Time Study, pp. 61-62. • Wyatt, Stock, and Frost, Incentives in Repetitive Work. ' Mace, Incentives: Some Experimental Studies. • Mathewson, Restriction of Output among Unorganized Workers; Roethlisberger, Management and Morale; Gardner, Human Relations in Industry. • Mundel, Systematic Time and Motion Study, p. 166.

Designing a Research Program

17

to produce at all times in some relatively fixed relationship to the standards, thus covering up time-study inconsistencies. Moreover, they "complain sufficiently on the standards which are difficult to achieve so that the standards are revised, and restrict their output on the easy ones so that the inequity there is undisclosed." Practices like these, Mundel concludes, are sufficiently widespread so that many poorly conceived time-study procedures often appear to be functioning well. The relation between wage-payment rates and production standards also provokes the workers to resist time study itself. Mathewson, for example, found that the mere intimation that a time study is to be made will often slow up an entire department.10 Roethlisberger sums up the situation succinctly. He says that "any person unknown to the workers who expresses more than a casual interest in their work or affairs is likely to be regarded with suspicion unless he takes pains to state clearly to them just what he is doing and why." 11 Additional evidence on this question was obtained in the present research program. As indicated earlier, some workers expressed strong opposition to the experimental studies until they became completely convinced that the results would not affect their earnings. This opposition disappeared completely only after enough time had elapsed to satisfy the workers that this was actually the case. Thus, although these studies were made under extremely favorable conditions, the workers still retained a residual fear and distrust of being observed. On the basis of such evidence, it would seem to be almost impossible to obtain unbiased results when time studies are made for the ultimate purpose of establishing wage-payment rate3. The Informal Bargaining Process Involved.—This behavior of workers can be classified as a form of informal bargaining, their objective being to obtain the most favorable production standards and wage rates possible. This bargaining proceeds in two stages. In the first stage the workers present a biased impression of the job requirements. In the second stage they protect lenient 10 11

Mathewson, Restriction of Output among Unorganized Workers, p. 71. Roethlisberger, Management and Morale, p. 81.

18

Designing a Research Program

standards and wage rates by restrictive practices, and they protest strict standards and wage rates by invoking grievance machinery. On the other hand, it is reasonable to assume that the timestudy analyst takes this situation into account in evaluating his results. He has an opportunity, for example, to counteract a worker's deliberately slow pace by rating him accordingly. In a similar manner, he can take advantage of the other subjective aspects of time study to counteract other biased practices of the worker being observed. The analyst also has devices at his disposal for protesting production standards and wage rates that turn out to be unfavorable to him (that is, are lenient). One such device is to introduce insignificant changes into an operation to justify making a new study and establishing stricter standards and rates. Although this practice is not nearly as widespread as it once was, timestudy analysts also sometimes use the more overt device of direct rate cutting under the same conditions. In a real sense, then, there are two parties to this bargaining process, even though they do not confront each other and state their cases explicitly. The Formal Bargaining Process Involved.—When the workers are organized formally, their attitudes and behavior on time study are explicitly incorporated into the union's official attitudes and behavior. Specific supporting evidence is supplied by the booklets published by the United Electrical Workers Union (UE) and by the United Automobile Workers Union (UAW).12 These booklets are intended to be a guide for union stewards in their negotiations with management. Each of them contains a detailed list of the controversial aspects of current time-study procedures. They also stipulate that the union is not to be a party to the time-study process. Instead, it reserves the right to bargain collective^ on all matters pertaining to individual studies, including their results. These booklets show that unions are aware of the shortcomings of time study. This puts them into a very favorable bargaining role, for they can then challenge production standards and 11 U. E. Guide to Wage Payment Plan». Time Study, and Job Evaluation; The UAW-CI O Looks at Time Study.

Designing a Research Program

19

wage rates that workers consider too strict. As nonparticipants, they do not have to take any action on the lenient standards and wage rates that might come to their attention. The over-all viewpoint of the cited booklets and other works by labor spokesmen is that time study is useful primarily as a means of narrowing the areas of disagreement between management and labor.13 Gomberg, the leading labor spokesman on this subject, sums up this viewpoint with the comment that "thus far modern industrial time study techniques can make no claims to scientific accuracy. They are at best empirical guides to setting up a range within which collective bargaining over production rates can take place." 14 This viewpoint clearly implies that labor's acceptance of time study is only a qualified acceptance. The qualification is that the primary function of time study is to supply information for collective bargaining. It is reasonable to conclude from this that labor's acceptance is based on the conviction that time study is not truly scientific, for this helps validate their argument that it can be effective only as a component of collective bargaining. The Question of Participation.—At first glance there seems to be a sharp disagreement among labor spokesmen on whether labor should participate in the time-study process. The UAW and UE booklets, for example, advocate complete nonparticipation, in contrast, let us say, to Golden and Ruttenberg, who advocate complete participation.15 Close examination shows, however, that the disagreement is of little real importance. Again the significant point is that time study is not scientific and, in particular, has numerous subjective aspects. Under these conditions labor's participation in the time-study process permits it to exercise its own judgment on these subjective aspects; it can then present its findings in competition with those of management. The result is that the differences between the two sets of findings are resolved by collective bargaining. The only essential difference, then, is that the collective bargaining process is brought to time study in the case of participation, and time 11 14 11

See, for example, Cooke and Murray, Organized Labor and Production. Gomberg, A Trade Union Analysis of Time Study, p. 170. Golden and Ruttenberg, The Dynamic» of Industrial Democracy.

20

Designing a Research Program

study is brought to the collective bargaining process in the case of nonparticipation. In any event, the problem of participation or nonparticipation would cease to be significant if time-study procedures were made more objective. Neither management nor labor would then need to concern itself with time study at all, except to see that the procedures are applied correctly. Specific Contractual Provisions.—Both the UE and UAW booklets strongly urge local unions to obtain contractual safeguards on key aspects of time study. A comprehensive report on this subject has been made by the Bureau of Labor Statistics. 16 This report points out that unions are generally allowed to appeal production standards; also, they usually reserve the right to bargain collectively on time-study procedures and results. This report also cites eighty-three separate contract provisions intended to safeguard labor's interests regarding time study. A representative group of these includes: (1) an operation must be restudied at the request of the employee or the union; (2) employees must be informed of any new production standard within 24 hours of its application. Upon request, the union may review time studies in consultation with the time-study supervisor; (3) representatives of management and labor shall cooperate in selecting the operators to be studied; (4) no time study shall take less than one-half hour in order to assure representative results under normal conditions; (5) minimum delay and fatigue allowances are to be established as follows: 5 percent for personal delays and 2 percent to 12 percent for fatigue depending on the nature of the operation. To protect its interests, management also obtains specific contractual safeguards in certain cases. The most common contract provision of this kind specifies that the union will guarantee a reasonable and just performance by its members. Another common provision is that a worker is to be transferred if he consistently produces at a rate below the production standard. The Bureau's report also indicates that the usually close relation between production standards and wage-payment rates 16 Collective Bargaining Provisions: Incentive Wage Provisions; Time Studies Standards oj Production.

and

Designing a Research Program

21

prompts unions to insist also on contractual safeguards regarding wage-payment rates. This relation also explains why almost all these provisions are concerned—at least implicitly—with time study. The most important of the 157 separate provisions of this kind enumerated in the report specify that: (1) the company agrees to consult with the union prior to changing the piece rate; (2) any employee who considers piece-work prices inequitable may bring up a grievance about them; (3) piece rates in effect are not to be changed unless a change is made in the method of manufacture, in which case the new time study and any change made in the rate of payment are to apply only to the affected elements; (4) increased skill on the part of employees shall not constitute a sufficient reason for changing established piece rates except by mutual consent; (5) piece rates shall include reasonable personal, fatigue, and machine allowances, and shall enable an average employee to earn at least 25 percent above the base rate (that is, the rate equivalent to the production standard). The Three Bargaining Stages and Their Implications.—The contract provisions described above represent the third stage of bargaining between management and labor when wage-payment rates are based on production rates. In the second stage formal bargaining also takes place, the unions exercising a general review function regarding time-study procedures and results. In the first stage informal bargaining takes place between the worker and the analyst, who represent labor and management at the local level. Considered together, these bargaining stages demonstrate conclusively that time study is not an independent scientific process in practice. It is, instead, either a direct or an indirect component of the bargaining machinery developed as a result of the relation between the time-study and the rate-setting functions. It seems clear, then, that time study must be made independent of rate setting if it is to perform its intended function of estimating production rates. Fixing the Dominating Variables.—In practice, the estimating function can be made independent of the rate-setting function only by observing production rates during a period when the

22

Designing a Research Program

wage-payment plan remains fixed. It should also be emphasized that the estimates obtained are valid only as long as the wagepayment plan remains unchanged. The reason for this is that a new set of estimates will otherwise be needed to maintain the independence between production rates and wage-payment rates. The need to separate production-rate problems from wage-rate problems has been considered in detail because of the strong relation between them in current practice. By itself, however, fixing the wage-payment plan does not guarantee that the estimating procedure will give valid results. It is also necessary to fix all the other dominating variables like the work method. The reason for this is that the estimates obtained will otherwise depend on those variables. This point can be demonstrated by a simple example. The production rates obtained by using one type of machine will differ in general from those obtained by using another type of machine. Estimates obtained for the first type of machine would therefore be of little value after the second type was introduced. In the language of experimental inference, fixing these variables satisfies the requirement that valid estimates can be obtained only under essentially constant conditions. A related requirement is that the estimates remain valid only as long as these conditions remain the same. THE MAJOR SHORTCOMINGS OF PRESENT PRACTICE

Difficulties of Scientific Verification.—Some time-study writers advance the argument that production standards must be reasonably accurate, since most workers seem to produce at a rate in the neighborhood of the standard rate. Two conclusive objections can be raised to this argument. The first is given by Clinton Golden and Harold Ruttenberg, Ralph Presgrave, and William Whyte, who find that: (1) numerous production standards are found to be inaccurate upon rechecking; (2) many such standards are as much as 50 percent too lenient or 50 percent too strict; (3) a substantial percentage of worker complaints refer to production standards.17 17 Golden and Ruttenberg, The Dynamics of Industrial Democracy; Presgrave, The Dynamic» of Time Study; Whyte, "Incentives for Productivity: The Bundy Tubing Case."

Designing a Research Program

23

The other and much more important objection has already been documented in some detail. Workers regulate their production rates during the time study so that the standard will be lenient and favorable to them; they also regulate their production rates after the study so that it has a fixed relationship to the standard. By these means the workers can easily make the standards appear accurate when it is in their interest to do so. Under present conditions, then, workers' production rates are based on their social and other vocational interests, both as individuals and as groups. As a result, the relationship of these rates to the standard rates gives no evidence whatsoever about either the validity of the standard rates or of the estimating procedures themselves. This is all the more true in those cases where the unions reserve the right to bargain formally over time-study procedures and results—which is another major obstacle to scientific verification. More Specific Shortcomings.—A wide variety of competitive and frequently contradictory time-study procedures are recommended in the literature. In the trenchant words of the UAW booklet, these procedures are "occasioned by nothing more solid than the desire to be different and, thereby, achieve exclusiveness and recognition." 18 In any event the very existence of numerous competitive procedures is in itself a tacit admission that a generally acceptable theory does not yet exist in this field. Inasmuch as numerous competitive procedures exist, it is natural to find that numerous competitive assumptions, definitions, and claims are made about time-study procedures and results. Apparently the only characteristic they have in common is that none of them has been verified by experimental means. The situation is even more serious than this. These assumptions, definitions, and claims are often so ambiguous and vague that it is almost impossible to subject them to verification studies. Louden gives a typically ambiguous definition when he says: "A fair standard is what we can expect a normal worker to do in the performance of a given job. . . . This standard is to be set on a fair, honest, and equitable basis, not requiring killing exertion, but at the same time requiring a normal day's output." 19 » The UAW-CIO Looks at Time Study, p. 29. Louden, "Management's Search for Precision in Measuring a Fair Day's Work."

19

24

Designing a Research Program

The principal weakness of nominal definitions such as this is that they fail to specify the empirical operations by which the entity defined can be given a unique interpretation. In Louden's case the definition of "a fair standard" is presented in terms which are subject to a variety of interpretations. For example, how "a normal worker" is to be identified is left completely unanswered. As the preceding chapter shows, the first requirement for effective experimental work is to define the fundamental concepts in operational terms. Excellent examples of such definitions are provided by Shewhart with regard to product-quality control. For example, he defines statistical control in the following terms: "a phenomenon will be said to be controlled when, through the use of past experience, we can predict, at least within limits, how the phenomenon may be expected to vary in the future. Here it is understood that prediction within limits means that we can state, at least approximately, the probability that the observed phenomenon will fall within the given limits." 20 The nonscientific nature of current time-study practice is underscored by the prevalence of subjective procedures. Cohen and Nagel neatly summarize the inadequacy of such procedures. According to them, the scientific method is reasonable, not because it appeals "to the idiosyncrasies of a selected few individuals, but because it can be tested repeatedly and by all men." 21 Shewhart's views on this subject are also pertinent. He says of the subjective approach that: "immediately the analysis of data would be removed from the field of logic and we would have to accept a result simply on the basis of the authority of the experimentalist. Then we would face the difficult task of determining the ultimate authority. Such a method is certainly not scientific, nor does history reveal much ground for belief that it is a method which can be relied upon to give satisfactory results." 22 The Use of Subjective Judgments.—It is frequently argued that 20

Shewhart, Economic Control of Quality of the Manufactured Product, p. 6. Cohen and Nagel, An Introduction to Logic and Scientific Method, p. 195. 11 Shewhart, Economic Control of Quality of the Manufactured Product, p. 386.

11

Designing a Research Program

25

it is impossible objectively to measure many of the variables in time-study work. This argument is then used to justify the application of subjective judgments in measuring such variables. A typical example of this kind of reasoning is supplied by Mundel. a He believes that worker performance cannot be evaluated by mathematical or statistical methods; this leads him to conclude that subjective judgments must be used instead. This conclusion can be challenged successfully on two important counts. The first is that subjective measurement methods do not necessarily yield valid results simply because objective measurement methods are not available. Furthermore, it is impossible to obtain precise and accurate results unless objective measurement methods do exist. This does not mean that subjective judgments are never used in scientific work, for a great deal depends on the investigator's judgment even in sciences as advanced as physics. However, the scientific method does demand that the estimating procedures be independent of the observer, thus enabling the results of experimental studies to be verified. It is in this sense that the subjective rating of worker performance, for example, is not a scientific procedure. Another common argument used in defense of subjective measurements is that their value is improved when the judgments of several observers are pooled. Louden, for example, concludes from performance rating results that "the group [of observers] is much less subject to error than the individual [observer]." 24 A little reflection shows that this argument is superficial and erroneous. The simple fact is that the repeated application of a nonobjective procedure cannot make that procedure more accurate or more precise than it was originally. It might be argued in rebuttal that group judgments are also used in the physical sciences. In the physical sciences, however, group judgments are employed to determine the most appropriate objective procedure. They are not used to make measure" Mundel, Systematic Time and Motion Study, pp. 155-156. Louden, "Management's Search for Precision in Measuring a Fair Day's Work," p. 29. ,4

26

Designing a Research Program

ments and estimates as they are in time-study rating. For subjective procedures the group judgment refers to results rather than to procedure. The outcome is that the results cannot be verified except by referring to the original observers. A serious consequence of the use of subjective procedures in time study is that often they lead to contradictory conclusions. An instructive example is provided by the conclusions of Phil Carroll and William Schutt regarding the so-called "snap-back" and "continuous" methods of taking stop-watch observations. These writers agree that the snap-back method involves an error of observation due principally to the mechanics of returning the watch hand to zero after each operation element. Carroll concludes, however, that this error is unimportant when compared to the error introduced by the subjective rating procedures.26 For this reason, he recommends the snap-back method except in special cases. Schutt, on the other hand, reasons that this error is large enough to warrant giving up the snap-back method in favor of the continuous method.28 Time Study as an Art.—Because of the shortcomings described above, some time-study writers assert that time study is more an art than a science. This view is often used as an additional argument for the conclusion that certain fundamental tools of the scientific method, particularly statistical procedures, cannot be applied to time-study problems. An extreme example of this type of argument is provided by William Lichtner, who even rejects the use of simple averages to summarize observed data. According to Lichtner, the required result should be neither an average reading nor the reading that occurs most frequently; it should combine both of these approaches in a manner that best represents the immediate circumstances.27 Such arguments can readily be shown to be invalid in a scientific sense, for no verifiable evidence is ever offered in their favor. It might also be commented that they are apparently not even " Carroll, Time Study for Cost Control, p. 70. * Schutt, Time Study Engineering, p. 54. " Lichtner, Time Study and Job Analysis, p. 38.

Designing a Research Program

27

accepted by the writers who advance them, for all of them claim to be able to obtain from time study numerical results which are accurate and precise. Equating Minuteness with Accuracy and Precision.—The shortcomings of current practice are brought into sharp focus by considering the problem of measurement. In general, time-study writers believe that the more minute the measurement, the more useful the result. The fact is that the scientific value of measurements does not depend on their minuteness. It depends instead upon their accuracy and precision, determined according to the specifications given in Chapter 1. The observed accuracy and precision are then compared to the order of accuracy and precision considered appropriate to the immediate problem. This comparison, not the minuteness of the measurement method, determines whether the measurements satisfy the objectives of the study. T H E SCOPE OF THE RESEARCH

PROGRAM

In addition to those considered, current time-study procedures have numerous other shortcomings when evaluated against the requirements of modern experimental inference. These will be given extensive treatment in later chapters. Enough major shortcomings have been presented, however, to establish an essential point. The shortcomings in time-study practice are not isolated matters to be corrected by doing correspondingly isolated pieces of research work. A comprehensive inquiry into the subject is required instead, leading to a completely reformulated system of objectives and estimating procedures. For the reasons given above, the first phase of the research program was devoted to key current procedures, special attention being paid to their accuracy, their precision, and their predictive value. The objective of this part of the program was to determine which of these procedures may be retained and which of them must be abandoned. The second phase of the research program was intended to develop a body of acceptable theory, based primarily on verifiable procedures and criteria. In this phase certain specific hy-

28

Designing a Research Program

potheses about work performance formulated by earlier investigators were also tested experimentally. A significant though nonexperimental phase of the program was a field survey of the views and practices of representative time-study practitioners. This survey was intended to determine: (1) to what extent and for what reasons the procedures used differ from those recommended in the literature; (2) the views of practitioners on the empirical usefulness of these procedures; (3) which aspects of time study need further fundamental research and the reasons cited; (4) whether procedures are used which are not reported in the literature. The Role of Rating Procedures.—The inadequacy of subjective procedures in applied science has already been documented in some detail. This inadequacy is particularly evident in the numerous procedures for rating work performance, where subjective judgments are made on the basis of an equally subjective standard—the hypothetical "normal" worker. Such procedures have no place in a set of objective procedures for estimating and evaluating production rates. For this reason, rating procedures are given only incidental attention here, except where they are intimately related to the topic under consideration. The subject of rating is given extensive treatment in much of the critical literature.28 A particularly thorough critique of the rating procedures in current favor is made by Gomberg. Like so many other critical students, Gomberg concludes that "it is at once apparent that nothing has been developed in industrial time study practice that can be considered an objective measure of normality or an objective method for comparing operator performance with any normal standard." 29 A Summary of the Program's Goal.—The goal of the research program was to develop estimating procedures that could be verified in terms of the modern theory of experimental inference. These procedures will not make it possible to predict how much time industrial work should require in terms of some hypothetically "normal" worker. In fact, it will be shown later that the 38 Representative critical evaluations of rating procedures are available in many of the references already cited. See, for example, Kennedy, Union Policy and Incentive Wage Methods. M Gomberg, A Trade Union Analysis of Time Study, pp. 145-146.

Designing a Research Program

29

amount of time required for any work can have scientific meaning only with respect to a set of predetermined specifications. Under these conditions current rates of production can be compared to the specifications to determine whether they are economically acceptable.

3. The Problem of Process Standardization M O S T time-study writers recommend that the process should be standardized. R. L. Morrow, for example, suggests that all the operations in a given work sequence be studied together.1 When the best equipment and sequence of operations have been decided upon, then the best arrangement of machines, and so forth, should be developed. The ultimate objective is to insure that all unnecessary movement and handling of materials is eliminated. Morrow also stipulates that other technical factors must be controlled so as to obtain maximum output along with quality assurance. Like most other writers on time study, Morrow also specifies that the "one best way" of performing the operation with the means at hand should be developed before taking time-study observations. Ralph Barnes' recommendations on this subject are similar to Morrow's. Thus, he suggests that work methods, materials, tools, and working conditions be standardized before standards of production are established.2 To accomplish this, Barnes proposes that a detailed study be made, including an analysis of the minute motions involved in the work method. In Barnes' opinion, however, a detailed study is not required to achieve standardization, for a sketchy study may be sufficient for the purpose. S. M. Lowry, H. B. Maynard, and G. J. Stegemerten have much the same viewpoint. They consider that it is necessary to determine the best method and to standardize the process before making the time study proper.® These three examples are typical of the views held by those 1 Morrow, Time Study and Motion Economy, pp. 96-97. * Barnes, Motion and Time Study, p. 260. * Lowry, Maynard, and Stegemerten, Time and Motion Study, p. 8.

Process Standardization

31

who insist on some kind of preliminary standardization. Although in the distinct minority, a few writers, such as Phil Carroll and Ralph Presgrave, feel that preliminary standardization is not absolutely necessary.4 As Presgrave puts it, it is often necessary to make time studies before standardizing the work methods. 5 CURRENT

PRACTICES

Standardization in Practice.—Despite the recommendations of most writers, the process of preliminary standardization has been found incomplete for well over a quarter of a century. In 1916, for example, Robert Hoxie reported that time studies were made with little or no previous standardization.® Recent evidence 26

8 5

6

II »I »I I _ 100 9 0 - 9 9 7 5 - 8 9 5 0 - 7 4 2 5 - 4 9 1 0 - 2 4 0 - 1 0 % FIGURE 1. THE PERCENTAGE OF OPERATIONS STANDARDIZED IN ADVANCE BY SURVEY RESPONDENTS

on this question is provided by the field survey mentioned in the preceding chapter. This survey showed that the proportion of operations standardized in advance varies widely, particularly with regard to work methods. The estimates given by the 51 respondents reporting on this 4 Carroll, Time Study for Cost Control. « Presgrave, The Dynamics of Time Study, pp. 128-129. • Hoxie, Scientific Management and Labor.

32

Process Standardization

subject are summarized in Figure 1. This chart points up the fact that only about 50 percent standardized their operations in advance. The principal explanations offered by those who did not standardize their operations in advance are that: (1) it is impossible to standardize all operations in advance; (2) standardization can be attained through time studies; (3) the cost of standardization is excessive; (4) it is necessary to standardize only key operations. Limitations of the Criteria Given.—Quite aside from their limited application, the recommended criteria of preliminary standardization are uniformly subjective and descriptive. This explains why such wide differences of opinion exist, ranging from the complete preliminary standardization insisted on by Morrow to Presgrave's willingness to take observations without any preliminary standardization. Wide differences of opinion also exist among practitioners as to what determines standardization, particularly with regard to work methods. These differences are pointed up by the criteria reported in the field survey, which were so numerous that it became necessary to put them into the following general categories: (1) all obvious improvements have been made; (2) the rules of motion economy have been fully applied to methods, equipment, and materials; (3) the operation is running smoothly with a minimum number of delays; (4) the production rates are running close to the time-study standard; (5) quality standards are being met; (6) the judgment of the foreman regarding operating performance. Three of the respondents also made the significant comment that the process of standardization is continuous—that complete standardization is an ideal that can never be realized. Six others agreed with Carroll and Presgrave; they felt that it is often impractical to effect preliminary standardization, even though it is theoretically desirable. The principal disadvantage of the recommended criteria, then, is that they lead to many different opinions with regard to the necessity for standardization. This difficulty can be overcome only by developing criteria based on procedures independent of the observer. To be realistic, these criteria must not be formu-

Process Standardization

33

lated in vague and subjective terms, as they usually are. Instead, they should be formulated in objective and specific terms. They should also take into full account the economic objectives of the individual plant and the nature of its operations. FINDINGS AND PROPOSALS REGARDING PROCESS STABILITY

Suggestive Critical and Experimental Material.—In his critical study C. Canby Balderston displays an intuitive recognition of the importance of stabilizing industrial operations. He argues, for example, that if technological standardization can be attained and "if the flow of work is uniform, the output of the worker group is not subject to fluctuation from factors beyond its control." 7 In considering the Western Electric data, Whitehead also hints at the importance of attaining process stability.8 One of his principal findings, for example, was that work patterns remained relatively unchanged, provided that the operations involved were equally difficult. Whitehead also found that the production rates of expert workers were more stable than those of novices. The second of these findings is supported by the studies of J. Loveday and S. H. Munro, who reported that highly skilled workers had more uniform production rates than did other workers.9 Numerous other investigators of the British Industrial Fatigue (later Health) Research Board have reported similar results, both in laboratory and in factory studies. Isabel Burnett, for example, found that the production-rate patterns of individual workers were strikingly similar in different experimental periods.10 Wyatt, Stock, and Frost carried this finding one step further.11 They showed that with a given wage-payment plan the over-all production rate becomes stabilized at a characteristic level. They also showed that this level remains practically unchanged as long as the conditions of work remain the same. The actual results they obtained for the three plans of wage payment considered are given in Figure 2. On this chart each ' Balderston, Group Incentive», pp. 5-6. » Whitehead, The Industrial Worker, I, 63-73. • Loveday and Munro, Preliminary Notes on the Boot and Shoe Industry. 10 Burnett, An Experimental Investigation of Repetitive Work. u Wyatt, Stock, and Frost, Incentives in Repetitive Work.

34

Process Standardization

^AfO vij

/20

^/oo / t t t t

/s

SO

W££KS

FIGURE 2. A COMPARISON OF THE RELATIVE OUTPUT UNDER THREE TYPES OF WAGE PAYMENT PLANS (FROM WYATT, STOCK, AND FROST, P. 5 )

reading represents the mean weekly output of all the workers assigned to certain packing and weighing operations. These weekly output values are presented in relative rather than in absolute terms, using a base value of 100 percent for the first week's output. Elton's Work.—In another Research Board report Elton's results crystallize the findings of this group of investigators.12 Elton studied the time required by each of twelve weavers to complete a series of warps on the same kind of cloth. Despite the small number of readings for each weaver, Elton was able to conclude that "considering the large number of causes at work which affect output the consistency of these figures, which have been taken at random and have not been specially selected, is remarkable." This conclusion led Elton to formulate an extremely suggestive hypothesis. "The rates of production of individual weavers," he says, "are fairly consistent, and when a substantial departure is made from a worker's average rate of production it ,J

Elton, An Analysis

of the Individual

Difference* vi the Output of

Silk-Weavers

Process Standardization

35

may reasonably be inferred that she has had to contend with particularly bad work (or exceptionally good), or that her normal working capacity has been stimulated or depressed." Though framed in descriptive and incomplete terms, this hypothesis is quite noteworthy. It shows that Elton had a definite notion of the meaning of statistical uniformity in production rates. It also suggests that specific procedures and criteria are needed in order to decide exactly how much variation constitutes "a substantial departure from a worker's average rate of production." Proposals regarding Shewhart Concepts.—The first suggestion, that Shewhart statistical-control concepts be applied to production problems, was made by Irving Lorge in a brief paper recommending the wider use of statistical methods in this field.13 Gomberg develops this thesis in somewhat more definite terms. He first points out that all the product-quality problems considered by Shewhart have direct analogies in the production field.14 He also emphasizes a point originally made by Shewhart : only extensive experience can show whether statistical-control concepts can be applied to a particular field. Thus, it is by no means certain that statistical-control procedures and criteria can be worked out for production problems. But unless such procedures and criteria can be worked out, Gomberg warns, it will be extremely difficult, if not impossible, to develop valid production standards. STATISTICAL STABILITY A N D PROCESS

STANDARDIZATION

The Empirical Significance of Statistical Control.—It has already been suggested that objective criteria of standardization must be based on procedures that take into account the peculiar characteristics and requirements of the individual plant. These procedures must also take into account the fact that the production rates of the same or different workers always exhibit a certain amount of variation. The most important function of these procedures, then, is to provide criteria which answer the following " Lorge and Haggard, A Physiologist and a Statistician Look at Wage-Incentive Methods. 14 Gomberg, A Trade Union Analysis of Time Study, pp. 30-49.

36

Process Standardization

questions: (1) When can the observed variations be attributed exclusively to random causes? (2) When are these variations so large that they must be attributed to nonrandom or assignable causes, such as Elton's "substantial departures"? It is important to develop such criteria because otherwise precise estimates and predictions cannot be made. The explanation for this has already been touched upon in the first chapter. In applied science experimental criteria are required to determine what constitutes a random sample (or samples) before the formal machinery of statistics can be applied to make estimates and predictions. Criteria enabling chance causes to be distinguished from assignable causes thus perform two crucial functions. When they show that all the variations can be considered due to random causes, the samples involved are called "random samples" and a state of statistical control is assumed. When the criteria show, however, that assignable causes of variation exist, the samples involved are called "nonrandom samples," indicating that statistical control does not exist.15 These concepts are of such great empirical significance that it pays to examine briefly what William Madow, a leading mathematical statistician, has to say on the subject." Tests for statistical control, he says, are applied extensively only in product-quality work. Nevertheless, statistical control is also required in order to make precise estimates and predictions in any empirical field. To point up his argument, Madow also shows how the presence of assignable causes of variation (that is, departures from a state of statistical control) can lead to erroneous conclusions regarding observed data. Applying These Concepts to Production Problems.—The variables involved in product-quality problems are much more limited in number and range of variation than the variables involved in production problems. There are several reasons for this, all depending on the fact that product quality is itself a variable with respect to production rates. For example, product 15 These questions are considered in painstaking detail in Sheivhart, Statistical Method from the Viewpoint of Quality Control. M Madow, "On a Source of Downward Bias in the Analysis of Variance and Covariance."

Process Standardization

37

quality is much less affected by the variables brought to the workplace by the worker than is production. This is partly because a worker's primary vocational interests are expressed in terms of production rather than product quality. The result is that the statistical-control procedures and criteria used in product-quality problems will have to be modified and generalized for production problems. There are three reasons for using the term "statistical stability" rather than the term "statistical control" in production problems. First, it helps distinguish between applications of Shewhart concepts to product-quality problems from applications to production problems. It also helps bring out the fact that the procedures and criteria used in the two types of application are not necessarily identical. Finally, the term "statistical stability" has a greater descriptive value than the term "statistical control," especially in connection with standardization. Local and Grand Stability.—Unlike product-quality applications, two distinct components of statistical stability can profitably be considered in production applications: local stability and grand stability. In the case of local stability the production rates involved represent a continuous series of items produced during a period of several hours or during a complete shift. In the case of grand stability the production rates represent items produced over a protracted period; these rates are examined in terms of small samples taken from successive production units made under essentially the same conditions. In the first case, then, the primary objective is to determine the key statistical properties of a continuous series of production rates, particularly the nature of the relationship among them. In the second case, however, the primary objective is to determine the key statistical properties of periodic samples representing production units of economic interest. This means that studies of local stability are intended to supply information different from that supplied by studies of grand stability. Ideally, then, the problem of examining the statistical stability of production rates should be divided into these two complementary components; a more complete knowledge of the total pattern of variation can be obtained by using them jointly.

38

Process Standardization

Under certain circumstances, however, it may be appropriate to examine production rates in local terms only or in grand terms only. If evidence is available, for example, showing that almost all the variation occurs from day to day, it would probably not be necessary to look into the variation within days (that is, local stability). The same conclusion would be reached when (as in most cases) the plant's primary economic interests are in the long-term aspects of production rates. Although they turn up much less often, cases also exist in which the circumstances make it necessary to look into the question of local stability only.

4. Case Studies of Local Stability T H E studies considered in this chapter were all made between 10 A.M. and 4:30 P.M. and never lasted more than one day. The observations proper were taken on one or more consecutive small lots of garments, varying from 48 to 60 units each. THE

OBSERVATION

PROCEDURE

Except where otherwise specified, the readings were taken with a stop-watch, using the continuous method. When the stopwatch is used in this way, the readings for successive operation elements are recorded cumulatively. Separate entries are made of all the delays observed and of the time they consumed. Ultimately, the net time for each element of an operation, as well as the total time for each operation cycle, is computed as illustrated in Table 1; these ten readings comprise a random excerpt from a series of observations made on Operation 1 considered below. In this table and throughout the book the readings are given in terms of hundredths of a minute for convenience of presentation. TABLE 1 .

STOP-WATCH READINGS FROM OPERATION

OBTAINED

1

ELEMENT

1

2

3

4

5

6

6 8 7 7 6 6 10 7 8 7

3 2 3 3 3 3 3 2 3 3

4 3 3 4 4 4 5 4 4 4

1 2 2 2 1 2 2 1 1 2

10 9 10 7 10 10 9 10 10 10

7 8 8 8 8 9 5 8 9 6

7 3 3 2 2 3 4 4 3 3 3

8

9

CYCLE

6 7 5 8 8 10 6 7 6 6

1 2 1 2 2 1 2 1 2 2

41 44 41 43 45 49 46 43 46 43

Case Studies of Local Stability

40

AN OUTLINE OF THE ANALYTICAL PROCEDURE

The Sample Size and Grouping Plan.—The first step in the analytical procedure was deciding on the size of the sample. In almost all the studies described in this chapter three units were used. This choice was made necessary by the fact that only a comparatively small number of successive readings could be obtained in most cases. Such small sample sizes, however, do have an important analytical advantage. There is less risk that significant changes will be obscured when only a few readings are involved. It will be shown later that such changes turn up fairly often in the production rates of the operations considered. Nevertheless, it should be emphasized that authorities such as Shewhart recommend samples of at least four units.1 Their argument is based on a fundamental statistical theorem: the distribution of sample means approaches the normal distribution as the sample size increases, regardless of the nature of the underlying population. The next step in the analytical procedure was to group the cycle times (and in some cases the element times) into samples of three units each. The only qualification was that samples were never constructed from production items taken from more than one lot. This explains why the number of samples almost never exactly matched the number of original observations. If a lot were made up of 59 units, for example, it would yield only 19 samples of 3 units each. The Criteria of Local Stability.—The third step was to decide on criteria of local (statistical) stability. In this case it was decided to adopt tentatively the conventional 3-sigma limits usually used in product-quality work. The actual control limits were thus computed from X ± A2R

in the case of the control charts for means and from DiR

and

DtR

in the case of the control charts for ranges. In these formulas % 1

Shewhart, Ecor^omic Control of Quality of the Manufaciurcd

Product.

Case Studies of Local Stability

41

represents the mean of the observed sample means, while R represents the mean of the corresponding ranges. The Ait Dt, and Di values for samples of three units were obtained from tables available in numerous texts on statistical quality control, some of which also give a detailed account of the analytical procedure outlined here.2 Other Empirical and Statistical Questions Involved.—The steps outlined above comprise only the framework of the studies of local stability to be described; numerous other empirical and statistical questions of some importance must also be considered in making such studies. However, these can best be understood in terms of illustrative examples. For that reason they will be discussed in connection with specific studies in which they are factors. Even so, a number of minor details of procedure and analysis have been omitted, since they are not of particular interest in developing studies of local stability. Such details are available in the literature on statistical quality control, notably Shewhart's cited works, where it is more appropriate to discuss them than it is here. In addition to the control charts, certain supplementary statistical tests were made in the studies of local stability. The conditions for making such tests and their empirical usefulness can also best be understood by means of illustrative examples. Accordingly, they will be described in connection with the studies in which they were applied. THE BACKGROUND OF THE STUDIES IN PLANT A

At the time of these studies, Plant A, where modern procedures of production and management control were in operation, employed about nine hundred production employees on a fulltime basis. Most of them were women engaged in various kinds of sewing operations. In this plant the majority of the studies were made on a group of key sewing operations performed on power-sewing machines of the same general type. Besides being key operations, these operations were also used as a basis for developing new opera* See, for example, Grant, Statistical Quality Control, which describes most of the control-chart methods used in product-quality applications.

Case Studies of Local Stability

42

tions; they thus made it possible to study operation elements in different groupings. Also, this group of operations included both old and new operations, making it feasible to compare the characteristics of operations in different stages of development. S T U D I E S ON O P E R A T I O N S

1

AND

1*

Cycle-Time Studies on Operator ID.—On Operation 1, which was made up of nine elements, 212 observations were taken on Operator ID. All of these observations were on size 32 garments except the first 13, which were on size 38 garments. Following the 50

45

J = 44.3

cn

z< Ui £

•

•

'

40

•

LCL S 38.5 35

20 l

to Ui o

UCL- 14.7 •

110

.* .

a.

•

•

• •

•

0

10

FIGURE 3 .

20

.

'

•

30 SAMPLE

,'R=5.7 'LCL=0

40 NUMBER

50

60

70

THE MEAN AND RANGE CHARTS ON LOCAL CYCLE

TIMES FOR OPERATOR

ID

analytical procedure, the cycle times were grouped into 70 samples of three units each. The means and ranges of these samples are plotted in Figure 3, the first 4 sample points on each chart representing size 38 readings. On these charts the samples corresponding to different lots are separated by vertical lines. All the plotted points on the chart for means fall inside the

Case Studies of Local Stability

43

corresponding limits. Assuming that the 3-sigma limits are appropriate criteria, the variations among the means can be considered locally stable. The variations among the ranges can also be considered locally stable, since only 2 of the 70 plotted points fall above the upper 3-sigma limit. The Ratio Test.—A close examination of the chart for means suggests that there are a number of increasing and decreasing sequences of points (that is, trends). The existence of such sequences can be verified by applying a statistical test originally developed by John von Neumann and his associates.3 The test was later modified slightly by A. H. J. Baines, who also describes its applications to problems like those considered here.4 The Baines version of the test, which will be used here, requires that the following ratio be computed: i»-i =

52

Y

(Xi+i

-

X i f / n-

1

£ (X,- - *)«/» - l i-i

In this ratio n refers to the number of observations, Xf to the ith observation, Xi+i to the i + 1th observation, and X to their mean. The numerator of this ratio represents an estimate of the underlying or population variability computed in terms of the differences between successive readings. The denominator represents another estimate of the population variability computed in terms of the deviations around the observed mean. As might be expected, significantly small ratio values indicate that the successive observations are positively serially correlated, as when a trend exists. On the other hand, significantly large ratio values indicate that the successive observations are negatively 1 See von Neumann, "Distribution of the Ratio of the Mean Square Successive Difference to the Variance," and von Neumann and others, "The Mean Square Successive Difference." * Baines, Methods of Delecting Non-randomness in a Given Series of Observations. This paper also gives a table for determining the statistical significance of an observed ratio value.

Case Studies of Local Stability

44

serially correlated, as when large readings are followed regularly by small readings. Values close to 2, however, indicate that no correlation exists. This ratio test was applied to the original 212 readings for Operator ID in two ways: (1) considering each lot separately, and (2) considering the 212 readings as a single group. The results are recorded in Table 2, which also gives the number of readings in each case. As for all other applications of this test in this book, a 10 percent significance level was used in both cases. This means that a 10 percent risk was taken of concluding erroneously that a serial correlation existed when it actually did not. TABLE 2 .

RATIO-TEST RESULTS ON THE DATA TOR OPERATOR

l

Lot

Size Readings (n) Ratio

38 13 3.51°

2

S

32 47 1.73

32 43 2.39

A

32 54 2.02

6

32 55 1.68

ID

Total

212 1.85

o This lot has a significant ratio value.

Since the significant value is large, it was concluded that the successive readings on the size 38 garments were negatively correlated. None of the other ratio values is significant, nor is the combined ratio value. These results indicate that the successive readings in this series were in general not correlated. Supplementary Studies on Operation Elements.—Control charts based on the same sample size were also constructed for the element readings obtained on Operator ID. A typical example is (zO15 < UJ 10

• •

•

'

,

7*

UCL= 13.5 ,5-10.6.

... .« .. • ... • . •• .•

LCL=7.7

g

. UCL=7.3

. . . n I i .. " . • » - aR =2.8 —* •• i •* •• • •* ••« • » . * * . 1_CLsO 10 20 30 40 50 60 70 SAMPLE NUMBER FIGURE 4 . THE MEAN AND RANGE CHARTS ON LOCAL TIMES FOR ELEMENTS 1 AND 2 FOR OPERATOR ID

Case Studies of Local Stability

45

provided by the charts for combined elements 1 and 2, reproduced here in Figure 4. In this case only two points fall above the upper control limit on each chart. It was therefore concluded that both the means and the ranges were locally stable. The same conclusion was reached for all the other elements, since not more than two readings fell outside the limits on any single chart. Studies on Operator 1H.—Another series of 258 readings on the same operation was taken on Operator 1H, who was working on size 32 garments exclusively. Control charts for the 83 means and ranges obtained from the cycle-time readings are given in Figure 5; the sample data for different lots are separated by vertical lines.

Ui

,

.

•.•

'

•

V

f 44.0.*

. .

L

37.9

uc L 1 15.5

• • •

• • • • •• •

0

•

•

10

•

• •

20

•

30

• ••

40

•

• •

• •

•

SAMPLE

FIGURE 5 .

•

•

•

•

•

••

• •

•

50

•

•

•

•

•

•

•• 60

'R'6.0

.

70

LCL'O

•

•

80

NUMBER

T H E MEAN AND RANGE CHABTS ON LOCAL CYCLE T I M E S

FOR OPERATOR

Iff

Again the conclusion was that local stability existed, a stronger indication of stability being given by the chart for means. Inspection of the latter chart also suggests that there was some trend behavior, first in a downward and then in an upward direction. These intuitive findings were confirmed by the significant

46

Case Studies of Local Stability

ratio value of 1.42 obtained for the means of the first three lots as against the nonsignificant ratio value of 1.91 for all 83 means. The element readings for Operator 1H were also treated by the control-chart method. As for Operator ID, the variations among the element times were found to be locally stable in each case, both for means and for ranges. THE IMMEDIATE IMPLICATIONS OF THESE

RESULTS

Estimating Population Values.—The results obtained thus far can be used to outline the immediate empirical implications of the procedures and criteria used. They show first that the cycle times were locally stable, with respect to both means and ranges. The ratio-test values obtained show that a certain amount of positive and negative serial correlation also existed in both cases. Since local stability was concluded to exist at both times, the samples involved can be considered random samples. This makes it possible to use the two J? values as unbiased estimates of the underlying mean local production rates of these operators. For the same reason the R values can likewise be used to obtain unbiased estimates of the underlying or population values of the variability of those production rates. Applying the Estimates Obtained.—With such estimates, precise comparisons can be made of the production rates of the two operators. The respective mean estimates of 44.3 and 44.0, which will be denoted by the X symbol, are clear 1}- of the same order of magnitude; this indicates that- these operators had approximate^ the same level of productivity. The most common and useful measure of the population variability is the standard deviation, which can be computed from

Like the multiplying factors for computing the 3-sigma control limits, the d2 values for different sample sizes are available in 4 This ratio in exact only when the underlying population has a normal distribution. However, it gives a good approximation with moat types of nonnormnl distributions encountered in practice, as Plackett shows in "Limits of the Ratio of Mean Range to Sigma."

Case Studies of Local Stability

47

numerous statistical quality control texts, such as that of Grant. 6 In the present case the ê values are 3.3 and 3.5, respectively, indicating that the cycle-time variability of these two operators was also of the same order of magnitude. Thus, control-chart studies provide specific and precise information regarding the key statistical properties of local production rates, provided that local stability has been established. With such information, these production rates can then be compared to determine whether some operators require additional training, better equipment, and so forth. Establishing that local stability exists has another essential advantage. It enables precise estimates to be made regarding future production rates as well as their limits of variation. In the case of Operator ID, for example, it is to be expected that the future mean local production rate would be approximately 44.3. In addition, it is to be expected that in the long run the means of samples of three units each would fall between the control limit values of 38.5 and 50.2 about 99 percent of the time. When Local Stability Does Not Exist.—If local stability does not exist, however, such estimates and predictions cannot be made. It would then be necessary to isolate and eliminate the assignable causes responsible. For example, several successive points outside the control limits on a chart for means might indicate that the sewing machine was defective or that the materials were of inferior quality. The elimination of such assignable causes serves two major purposes. First, this step is necessary to bring about local stability, without which precise estimates and predictions are impossible. However, assignable causes are also usually economically undesirable and generally their elimination either increases the uniformity of production rates or decreases their mean value. Using the two cited examples, it is clearly desirable to know and to eliminate the use of materials of inferior quality or to put a sewing machine into good working order. It should also be noted that the assignable causes are sometimes economically desirable. This is particularly true when a " Grant, Statistical Quality Control.

48

Case Studies of Local Stability

number of points fall below the lower control limit. In such cases the assignable cause might be a superior work method, which could profitably be introduced into the operation on a systematic basis. In man-controlled operations such as those considered, assignable causes are often temporary in studies of local stability. A typical example of this occurs when the operator's attention is partially distracted. For this reason it seems reasonable to permit two points to fall outside the limits in such operations and still to conclude that statistical stability exists, especially when at least 35 samples are involved. Comparing Performance to Specification.—Once local stability has been established (if necessary by eliminating assignable causes), the estimates obtained can also be compared to the production specifications of the plant. Suppose, for example, it had been economically desirable for Plant A to have a local mean cycle time of 25 for Operation 1 and the mean local cycle time (computed by pooling the mean cycle times of the operators) turned out to be 44. It is clear that such a situation would require a fundamental change in the process, such as revising the quality requirements or eliminating certain delay factors. Similarly, if the standard deviation of these local cycle times were too high, a fundamental change would have to be made to increase the uniformity of the local cycle times; such revisions might be effected by granting rest periods and changing the task. Since such fundamental changes must be made when local stability exists, they always involve one or more dominating variables. This point is important. It means that these changes are much more far-reaching than the relatively minor changes usually involved in eliminating assignable causes when local stability does not exist. STUDIES ON OPERATIONS 1 * AND

2

The Studies on Operation 1*.—Operation 1*, which had 13 elements, was largely constructed from elements taken from Operation 1. The relationship between these two operations is given below in terms of their respective element structures; these

Case Studies of Local Stability

49

element structures are typical of the operations considered from Plant A. Operation 1 1. Pick up left back and front, position, enter, and straighten. 2. Guide stitching of back to pocket. 3. Turn garment and straighten. 4. Guide stitching of bottom to back. 5. Turn garment, pick up eye-end and label, stitch. 6. Pick up right back and hook-end, position, stitch, turn and straighten. 7. Guide stitching of bottom to back. 8. Turn back, turn front, position under back, enter, and straighten. 9. Guide stitching of back to pocket.

Operation 1• 1. Pick up left back, enter both sides, turn (cf. element 1, operation 1). 2. Transport previous garment to lap. 3. Pick up label and eye-end, position under back, and straighten (cf. element 5, operation 1). 4. Turn and straighten short curve. 5. Turn and stitch top of back (cf. element 4, operation 1). 6. Turn, pick up front and position, lift back for positioning, enter and straighten (cf. element 8, operation 1).

7. Stitch back to front (cf. element 9, operation 1). 8. Turn, pick up right back and enter (cf. element 1, operation 1). 9. Turn and straighten short curve (cf. element 4, operation 1*). 10. Turn, pick up hook-end and position (cf. element 5, operation 1). 11. Turn, stitch both sides (cf. element 4, operation 1). 12. Turn, position garment under back, enter, and straighten (cf. element 6, operation 1*). 13. Stitch back to front (cf. element 7, operation 1*).

A total of 76 readings was obtained for Operator 1 *A, the only worker on this operation at the time of the studies. The first 31 readings were on size 32 garments, and the others were on size 36 garments. This gave 10 samples for size 32 and 14 samples for size 36, whose cycle means and cycle ranges are shown in Figure 6. Topoint up the fact that they had different means and ranges, the X values, the R values, and the control limits for the two sizes are plotted separately. The chart also gives the % value, the R value, and the control limits for all the samples considered collectively. Considered separately, each group of sample points falls inside the corresponding limits. All the points also fall inside the combined limits, even though the size 32 values are visibly smaller. This indicates that these sizes were locally stable when consid-

50

Case Studies of Local Stability 1»CL'6l3

60

8*35.5 TOTAL

•

L 2U 50
•

30 NUMBER

AND

a

LCL «0

40

RANGE

CYCI.E TIMES

FOR

FIRST STUDY

These did not prove to be significant when the ratio test was applied, for the ratio values obtained for the sample means and for the original readings were 1.54 and 1.79, respectively. The fact that the production rates of this inexperienced operator were stable is not as unexpected as it might seem. The criteria of statistical stability depend exclusively on the characteristics of the operation and the operator being observed. Thus, the control chart shows only whether an operator's performance at any given time is consistent with liis own immediate performance history. This reasoning also explains why the nature and

Case Studies of Local Stability

57

the complexity of the operation are not primary factors in determining whether statistical stability exists. The Second and Third Studies.—Two other studies were made on the same operator one month and two months, respectively, after the first one. The second study involved 58 readings on size 36 garments, while the third involved 53 readings divided between size 34 and size 38 garments. Though the number of samples was limited in both of these studies, the control charts showed that the data could be considered locally stable. These two studies provide additional evidence that an inexperienced worker can perform at a stable production rate. However, they do show that the amount of experience possessed by a worker does affect the X and 5 estimates obtained from the control charts. To point up this fact, the estimates obtained from the three studies on this operator are recorded in Table 3. TABLE 3.

T H E PRODUCTION-RATE E S T I M A T E S

ON S T U D I E S MADE ON OPERATOR

BASED

13A

EXPERIENCE

X estimate & estimate

One Month

93.3 4.8

Three Month»

76.5 4.7

Five

Month»

72.4 3.8

This table suggests that while an operator is still learning the mean cycle time decreases continuously, but at a declining rate. On the other hand, the variability of the cycle times remains essentially constant. This suggests that consistency of performance is achieved early and that subsequent improvements are confined primarily to increasing the level of productivity. These studies also indicate that the periodic checking of a beginner's local production rates has several important advantages. The first is that assignable causes of variation, such as an inferior method of work, can be isolated and removed before they become part of the worker's permanent performance pattern. In addition, once local stability has been established, the % and § estimates obtained can be used to determine whether the learning process is proceeding satisfactorily. They can also be used to determine when this process has advanced sufficiently so that the worker may be removed from the beginner's class.

58

Case Studies of Local Stability

REPRESENTATIVE STUDIES OF LOCAL STABILITY IN PLANT B

The Background of the Studies Made.—In Plant B almost all the studies of local stability were made in the hand-pressing section of the pressing department. This section was made up of 10 men in addition to the head presser. The second pressing section was made up of 16 men who worked on "buck pressers" of the type also used in dry cleaning establishments. There were also approximately 160 other production employees, mostly women, engaged primarily in hand- and power-sewing operations. As in Plant A, the production process in Plant B was organized on a "section" basis, in which individual operators worked on only one aspect of garment making, such as cutting, sewing, or pressing. Unlike Plant A, however, the production- and management-control procedures were not well developed here. This was particularly true of the materials routing and handling techniques, for workers were often interrupted in the middle of a lot of garments to work on some "rush" lot. In addition, the arrangement of the work stations was not- adapted to the swift and economical movement of materials. The situation with respect to work methods was equally unsatisfactory, for little or no time was spent on training workers. The result was that the workers had substantially different methods of performing the same operation. Many of the studies made in Plant B were too short to permit valid conclusions to be drawn regarding the existence of local stability. This situation was the result of two principal factors: (1) the extremely long time required to press most garments; (2) the pressers rarely worked on the same operation for more than a single lot, comprising 60 units or less. Studies on Operation 21.—Despite these difficulties, a number of fairly conclusive studies of local stability were made in this plant. The first was on hand-pressing Operation 21, which was made up of 6 elements. A series of 76 observations was obtained on Operator 21 A; the results were treated by the control-chart method as shown in Figure 10. On the chart for means, the two sample points above the upper limit are part of a sequence of five high points. In addition, 80

Case Studies of Local Stability

59

150 UCL.» 137.0 130 X' 115.4

110 LCL = 94.6

90 60

UCL*_533

40

z< R = 20.1

20

LCL= 0 O FIGURE 1 0 .

5

10 SAMPLS

15 NUMBER

THE MEAN AND RANGE

20

CHARTS ON LOCAL

25 CYCLE

TIMES FOR OPERATION 2 1

percent of the points on this chart fall below the central or X value. Considered together, these facts show conclusively that the means were not locally stable. On the chart for ranges, however, only one point falls above the upper limit, indicating that the ranges were locally stable. This example shows that the control chart should be evaluated as a whole in forming a judgment regarding statistical stability. In this case most of the sample points fell below the central value on the means chart; this is sufficient in itself to conclude that stability does not exist. The two points above the upper limit and the sequence of high points make that conclusion even more decisive. Studies on Operation 22.—Another set of studies was made on the same operator, working on hand-pressing Operation 22 with 7 elements. The 42 readings obtained in this case were grouped into samples of two units each, primarily to have enough samples

60

Case Studies of Local Stability

to make a meaningful control-chart judgment. The results are plotted in Figure 11. These charts show that the ranges could be considered locally stable. Some doubt exists, however, about the stability of the means, especially since only 21 samples of two units each were involved. There is also some evidence of a positive correlation on the chart for means; this hypothesis was confirmed by the significant value of 1.30 obtained in the ratio test. 120

UCLI'IO.O

no

c2n 3

X s 97.2 90

•

LCL = 84.2

.

80

(JCL=22.4 20 •

U J o 2 10 tn

i

•

R*6.8 •