The Six Sigma Practitioner’s

Guide to Data Analysis

Second Edition

Donald J. Wheeler
Fellow of the American Statistical Association
Fellow of the American Society for Quality

SPC Press
Knoxville, Tennessee
 Contents
About the Author xi
Acknowledgments xii
Preface to the Second Edition xii
Introduction xiii
PART ONE: THE FOUNDATIONS OF DATA ANALYSIS 1

One Four Statistical Problems 3

1.1 Descriptive Statistics 3
1.2 Probability Theory 5
1.3 Statistical Inference 6
1.4 The Homogeneity Question 8
1.5 Two Perspectives on These Problems 9
1.6 Axioms of Data Analysis 12
1.7 Observational Studies and Experimental Studies 16
1.8 Summary 17
Two Descriptive Statistics and Homogeneity 19
2.1 What Does the Standard Deviation Statistic Do? 19
2.2 What Descriptive Statistics Do Not Do 29
2.3 Local Measures of Dispersion 30
2.4 Are the Data Homogeneous? 33
2.5 The Difference between Global and Local 38
Three Process Behavior Charts 41
3.1 The Chart for Individual Values 41
3.2 What Do We Gain from the mR Chart? 44
3.3 What Makes the XmR Chart Work? 46
3.4 How Many Values Do I Need for Limits? 50
3.5 Rules for Detecting Nonhomogeneity 53
3.6 Average and Range Charts 58
3.7 Chunky Data 62
3.8 Caution regarding Software 65
3.9 Where Do We Go from Here? 66

v
 Four Statistics, Parameters, and Inference 67
4.1 The Concept of a Probability Model 67
4.2 Some Cautions regarding Probability Models 71
4.3 Elements of Statistical Inference 72
4.4 Interval Estimates of Location 76
4.5 Interval Estimates of Dispersion 84
4.6 Practical Statistical Inference 86
4.7 Interpreting “Degrees of Freedom” 88
4.8 Summary 91
PART TWO: THE TECHNIQUES OF DATA ANALYSIS 93
Five Data Collected under One Condition 95
5.1 What Can We Say about Our Process? 95
5.2 But What about the Significance Levels? 99
5.3 Bead Board No. 3 100
5.4 The Data for NB10 105
5.5 Summary 107
Six Data Collected under Two Conditions 111
6.1 Detecting a Difference with Histograms 111
6.2 Detecting a Difference with XmR Charts 113
6.3 Using an Average and Range Chart 116
6.4 The Analysis of Means 119
6.5 The Two-Sample Student’s t-Test 122
6.6 The Paired t-Test 125
6.7 NB10 Revisited 128
6.8 Summary of Comparing Two Conditions 130
Seven Data Collected under Three or More Conditions 133
7.1 XmR Charts for Each Treatment 133
7.2 Average and Range Charts 135
7.3 Analysis of Means 136
7.4 Analysis of Variance 137
7.5 The Tukey Post-Hoc Test 145
7.6 NB10 Again 146
7.7 Summary of Comparing Several Conditions 148

vi
 Eight Data Collected at Three or More Values for X 151
8.1 The Universe Had a Definite Beginning 151
8.2 Evaluating Terms in the Model 154
8.3 Regression: One Line or Two? 156
8.4 A Deterministic Relation with Controlled X Values 158
8.5 A Deterministic Relation with Uncertain X Values 160
8.6 X and Y Are Random Variables 162
8.7 Data Snooping 166
8.8 Trends 170
8.9 The Role of the Scatterplot 170
8.10 The Problem of Regression Models 171
Nine Count-Based Data 175
9.1 Two Types of Counts 175
9.2 Interval Estimates for Universe Proportions 177
9.3 Homogeneity for Counts 180
9.4 Should I Compare Counts or Rates? 182
9.5 A Caution regarding Counts on an XmR Chart 183
9.6 Outliers or Signals? 186
9.7 Comparing Two Proportions with Charts 187
9.8 Comparing Two Single Proportions 189
9.9 Comparing Proportions for Several Conditions 192
9.10 Summary 196
Ten Counts of Events 197
10.1 Inference for Counts of Events: One Condition 197
10.2 Using the Poisson Model 201
10.3 Comparing Two Conditions 202
10.4 Comparing Several Conditions 205
10.5 Summary 206
Eleven Counts for Three or More Categories 207
11.1 Categorical Frequencies for One Condition 207
11.2 Categorical Data for Two or More Conditions 215
11.3 Summary 218

vii
PART THREE: THE KEYS TO EFFECTIVE DATA ANALYSIS 219

Twelve The Dual Nature of Trouble 221

12.1 “We Are in Trouble” 221
12.2 A New Definition of Trouble 223
12.3 The Four Possibilities for Any Process 225
12.4 Research and Experimentation Are Not Enough 235
12.5 Operating a Process Predictably 241
12.6 Summary 242
Thirteen Capability and Performance Indexes 243
13.1 The Voice(s) of the Process 243
13.2 The Voice of the Customer 245
13.3 The Capability Ratio, Cp 248
13.4 The Centered Capability Ratio, Cpk 250
13.5 The Performance Ratio, P p 253
13.6 The Centered Performance Ratio, Ppk 256
13.7 Using the Four Ratios Together 257
13.8 The Ratios and the Four Possibilities 259
13.9 Operational Improvement 260
13.10 Short-Term and Long-Term Capability 262
13.11 Interval Estimates for Capability Indexes 262
13.12 Summary 267
Fourteen Using the Effective Cost of Production and Use 269
14.1 The Effective Cost of Production and Use 269
14.2 The Baseline Cost of Production and Use 271
14.3 The Centered Cost of Production and Use 272
14.4 The Predictable Cost of Production and Use 273
14.5 The Minimum Cost of Production and Use 275
14.6 Using the Effective Cost of Production and Use 276
14.7 Summary 280
Fifteen The Basis for the Effective Cost of Production and Use 281
15.1 The Structure of the Effective Cost of Production and Use 281
15.2 The Average Excess Costs 282
15.3 Finding the Effective Cost of Production and Use 288
15.4 Summary of ECP&U for Measurements 289
15.5 The Effective Cost of Production for Counts of Items 290
15.6 The Effective Cost of Production for Counts of Events 291

viii
 Sixteen The Six Sigma Zone 293
16.1 The Effective Cost of Production and Use Curves 293
16.2 An Illustration 299
16.3 The Six Sigma Zone 301
16.4 Summary 302
Seventeen Some Problems 305
17.1 Problems with Defects per Million 305
17.2 Problems with Defects per Million Opportunities 310
17.3 Problems with FMEA Risk Priority Numbers 311
17.4 Problems with Special Causes 316
17.5 Problems with DMAIC Models 318
17.6 Do We Need a Gauge R&R Study? 321
17.7 Problems with Narrowly Defined Projects 323
17.8 Summary 331
Eighteen Two Models for Process Improvement 333
18.1 Getting Started 333
18.2 Characterize the Status of the Process Outcomes 334
18.3 Outcomes in the State or Brink of Chaos 335
18.4 Outcomes in the Threshold State 337
18.5 Outcomes in the Ideal State 338
18.6 Focusing Improvement on Strategic Objectives 338
18.7 Summary 341
Nineteen An Honest Gauge R&R Study 343
19.1 The AIAG Gauge R&R Study 343
19.2 The “Percentages” of the Total Variation 346
19.3 The “Percentages” of the Specified Tolerance 350
19.4 The “Number of Distinct Categories” 353
19.5 The AIAG Guidelines 353
19.6 What Can You Learn from the AIAG Study? 357
19.7 An Honest Gauge R&R Study 358
19.8 Summary 363

ix
Appendix 367
Table A.1 Bias Correction Factors for Measures of Dispersion 368-369
Table A.2 XmR Charts: Charts for Individual Values 370-371
Table A.3 Average and Range Charts Based on the Average Range 372-373
Table A.4 Average and Range Charts Based on the Median Range 374-375
Table A.5 Average and Standard Deviation Charts
Based on the Average Standard Deviation Statistic 376-377
Table A.6 Average and Standard Deviation Charts
Based on the Median Standard Deviation Statistic 378-379
Table A.7 Critical Values for Student’s t Distributions 380-381
Table A.8 Percentiles of Chi-Square Distributions 382-383
Table A.9 Overall Alpha-Levels for Average Charts 384
Table A.10 Overall Alpha-Levels for Range Charts 385
Table A.11 Quick Scale Factors for Range-Based ANOM 386-388
Table A.12 Quick Scale Factors for Pooled Variance ANOM 389-391
Table A.13 Percentiles of the F-Distribution 392-397
Table A.14 Percentiles of the Studentized Range Distribution, q 398-403
Table A.15 Factors for 90% Interval Estimates of Cp and Pp 404
Table A.16 Factors for 90% Interval Estimates of Cpk and Ppk 405
Table A.17 The Effective Cost of Production and Use
When All Nonconforming Units Are Scrapped 406-408
Table A.18 The Effective Cost of Production and Use
When All Nonconforming Units Are Reworked 409-411
Table A.19 The Effective Cost of Production and Use
When the Cost of Rework Is Half of the Cost of Scrap
and the Average Is on the SCRAP Side of the Target 412-414
Table A.20 The Effective Cost of Production and Use
When the Cost of Rework Is Half of the Cost of Scrap
and the Average Is on the REWORK Side of the Target 415-417
Index 419