About the Authors Daniel Sloan is an internationally recognized Six
Russell Boyles earned his PhD in Statistics at the
...
59 downloads
890 Views
5MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
About the Authors Daniel Sloan is an internationally recognized Six
Russell Boyles earned his PhD in Statistics at the
Sigma Master Black Belt and an ASQ certified Black Belt. His 16 years of experience have been distinguished by Six Sigma seminars in Mexico, Uruguay, Brazil, Australia, and 47 of the United States. McGraw Hill and Quality Press published five of his 7 books. As a Senior Vice President of Applied Business Science for a $500 million company, he led their Six Sigma initiative. With "factory floor" Six Sigma successes ranging from non-woven fabrics, extruded products, medical equipment, aerospace engineering, automotive parts, to Internet router production and health care, Daniel has a proven track record in helping companies produce bottom line results.
University of California, Davis. He subsequently spent two years in the Applied Mathematics Group at Lawrence Livermore National Laboratory, two years as Director of Statistical Analysis for NERCO Minerals Company, and eight years as Statistical Process Control Manager at Precision Castparts Corporation. As a trainer and a consultant, Russell specializes in Six Sigma Master Black Belt and Black Belt certification courses, Design of Experiments, Gage Studies, Reliability and Statistical Process Control. A few of his recent papers have appeared in ASQ publications Technometrics and Journal of Quality Technology.
Evidence-based Decision Services and Products We are the first and best provider of evidence-based decision services in the world. We help clients rapidly use the evidence in their raw data to dramatically improve bottom line business results.
Six Sigma Services
Evidence-based Decision Support
Master Black Belt, Black Belt, Green Belt, Champion, and Senior Executive certification training for all industries including manufacturing, financial services, and health care.
Data mining, strategic Information Systems design. Bottom-line business results project coaching.
Consortium Six Sigma events for small companies who wish to pool resources.
Consulting support to private industry, government and academic institutions that are implementing evidence-based decision systems.
Custom designed training events and multi-media, evidence-based Six Sigma materials.
Custom designed training events and multi-media, evidence-based education and training materials.
For more information visit or call: http://www.evidence-based-decisions.com Portland, OR (503)-484-5979 or Seattle, WA (206)-525-7968
M. Daniel Sloan, author and owner of the copyright for this work, has licensed it under the Creative Commons Attribution Non-Commercial NonDerivative (by-nc-nd) License. http://www.danielsloan.com is the legal, file download location. To view this license visit:
http://creativecommons.org/licenses/by-nc-nd/2.5/ http://creativecommons.org/licenses/by-nc-nd/2.5/legalcode Or send a letter to: Corporate Headquarters Creative Commons 543 Howard Street 5th Floor San Francisco, CA 94105-3013 United States
Profit Signals How Evidence-based Decisions Power Six Sigma Breakthroughs
By M. Daniel Sloan and Russell A. Boyles, PhD
Evidence-based Decisions, Inc. Seattle, Washington
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Profit Signals, How Evidence-Based Decisions Power Six Sigma Breakthroughs. M. Daniel Sloan and Russell A. Boyles Library of Congress, Cataloging-in-Publication Data Sloan, M. Daniel, 1950Boyles, Russell A., 1951Profit signals how evidence-based decisions power six sigma breakthroughs / M. Daniel Sloan and Russell A. Boyles Included bibliographical references and index. 1. Six Sigma—quality control—Statistical Models. 2. Medical care—quality assurance— Statistical Models. 3 Cost Control—Statistical Models—Mathematical Models. © 2003 by Evidence-Based Decisions, Inc. http://www.evidence-based-decisions.com All rights reserved. No part of this book may be reproduced in any form or by any means, electronic, mechanical; photocopying, recording, or otherwise, without the prior written permission of the publisher. Your support of author’s rights is appreciated. For permissions the authors can be contacted directly. Sloan Consulting http://danielsloan.com/ 206-525-7968 10035 46th AVE NE, Seattle WA 98125 Westview Analytics http://westviewanalytics.com/ 3099 Rosemary Lane Lake Oswego, OR 97034 503-635-8967 Trademark Acknowledgements Profit Signals® and the phrase “Vector Analysis Applied to a Data Matrix®” and the Profit Signals tetrahedron on the book’s cover are registered trademarks of Evidence-based Decisions, Inc. Six Sigma® is a registered trademark and service mark of Motorola, Incorporated. Sculpey Clay® is a registered trademark of Polyform Products Co. Excel® is a registered trademark of Microsoft. Other copyright notices are listed in the production notes at the end of the book. Illustrations: Cover, Robin Hing. Tables and illustrations, Robin Hing, Russell A. Boyles, M. Daniel Sloan, John Pendleton, Austin Sloan, and Alan Tomko. Netter illustrations used with permission from Icon Learning Systems, a division of MediMedia USA, Inc. All rights reserved. The book’s design and layout, using Adobe InDesign 2.0.2, were completed by M. Daniel Sloan. Printed in the United States of America.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Many of the most useful designs are extremely simple. Ronald Alymer Fisher
How much variation should we leave to chance? Walter A. Shewhart
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Table of Contents
Premise ................................................................... 9 The Parable of the Paper Bags ........................... 14 The Dollar Value of Evidence............................ 16 Six Sigma ......................................................... 18 How to Read This Book .................................. 20 Endnotes .................................................................24
Chapter 1—The Five-Minute PhD ............... 27 Start Your Stopwatch Now ............................... 28 Business Art and Science ................................... 30 Profit Signals ..................................................... 37 Data Recycling.................................................. 43 The Full Circle of Data Discovery..................... 44 The New Management Equation ...................... 44 Closing Arguments ........................................... 46 Endnotes ........................................................... 47
Chapter 2—Standards of Evidence ............... 49 Poetry versus Science......................................... 50 “Scientific” Management................................... 51 Cost Accounting Variance Analysis ................... 53 Accounting versus Science ................................. 55 Delusions and Bamboozles ................................ 56 Vector Analysis 101 ........................................... 57 Degrees of Freedom........................................... 65 Bar Chart Bamboozles ..................................... 67 The Game is Afoot............................................ 74 Spreadsheet versus Data Matrix......................... 79
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
vi
Table of Contents
P-values, Profit Signals, Confidence Levels and Standards of Evidence ................................ 81 Closing Arguments ........................................... 83 Endnotes ........................................................... 84
Chapter 3—Evidence-based Six Sigma ........ 87 Six Sigma (6σ) Basics ....................................... 89 The Six Sigma Profit Strategy .......................... 90 Lucrative Project Results Map ...........................94 Define, Measure, Analyze, Improve, Control ... 96 Lucrative Project Selection ............................... 98 Financial Modeling and Simulation ............... 100 Compare and Contrast Analysis ..................... 104 Process Maps ................................................. 106 The Costs of Poor Quality ...........................110 Process Capability ..............................................113 Endnotes ..............................................................115
Chapter 4—Case Studies ...............................117 Customer Service – Governmental Agency.......119 Days in Accounts Receivable ........................... 122 Breaking the Time Barrier .............................. 128 “Beating Heart” Bypass Grafts ...................... 135 The Daily Grind ........................................... 142 “Die Tuning” for Vinyl Extrusion ....................145 Endnotes ..........................................................149
Chapter 5—Using Profit Signals .................151 A Better Way to Look At Numbers ..................152 Corrugated Copters..........................................153 Testing the Current Way of Doing Things .......156 Overcoming Obstacles ................................... 162 Comparing Two Ways of Doing Things.......... 163 Comparing Three Ways of Doing Things ........167 Comparing Eight Ways of Doing Things .........169 Comparing 256 Ways of Doing Things............172 Chapter Homework..........................................174 Closing Arguments ..........................................175 Endnotes ..........................................................175
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Table of Contents
vii
Chapter 6—Predicting Profits ..................... 177 Fingerprint Evidence ........................................178 Three Wishes ...................................................179 Prediction Practice ...........................................183 Predicting Real Flight Times........................... 188 Closing Arguments ..........................................191 Endnotes ..........................................................191
Chapter 7—Sustaining Results .....................193 Evaluating Practices and Profits....................... 194 Process Improvement Simulation..................... 199 Monitoring Practices and Profits ..................... 205 Taking Action ..................................................211 Closing Arguments ..........................................214 Endnotes ..........................................................214
Chapter 8 —The Three Rs ............................217 Six Sigma’s Hidden Factory ..............................218 Our Proposal................................................... 222 Endnotes ......................................................... 224
Appendices ....................................................... 225 I. Glossary of Terms: Data Matrix, Vector Analysis And Evidence-based Decisions ......................... 225 II. The Business Bookshelf ............................. 227 III. Evidence-Based Decisions, Inc. Six Sigma Black Belt/ Expert 16 Class Curriculum Outline ................................. 231 IV. Profit Signals Production Notes ................. 252
Index ...................................................................255
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
viii
©
Table of Contents
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
P
rofit Signals is a guide for using evidence to make better, more profitable business decisions. Face value judgments, opinions, gut feelings, suspicions, circumstance, and superstitions are not pathways to evidence. Measurements are. This book will show you how to turn measurements into evidence and evidence into profit. Measurements become evidence when they are analyzed correctly. Since 1920, the correct analysis has consisted of a vector analysis applied to a data matrix. Very few people know this. With this book, we aim to transform the arcane mysteries of vector analysis into common knowledge. Vector analysis is a must-have, fundamental job skill. Every person in every organization can use this tool to make more money. Vector analysis is a vast, audacious and empirically true Generalization. Sir Ronald Fisher first explained it at the beginning of the 1920s.1 Evidence is the foundation of Profit Signals and vector analysis is the foundation for evidence. The word ‘generalization’ usually denotes a thoughtless, broad assertion with no basis in fact. By contrast, a Generalization is a verifiable law of the universe. Thus one word has two, opposite meanings.2 Unlike a generalization, a Generalization delivers valid conclusions and accurate predictions. The laws of gravity are a Generalization. Gravity is a physical constant of our universe. The laws of motion are a Generalization. The laws of Variation, the chance fluctuations or Noise that attend every measurement, are a Generalization. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
10
Premise They are Law. Just as you can measure gravity (Newton and the apple), Generalizations like statistical variation can be tested and validated through a process of experimentation, observation, and analysis. Evidence-based decisions focus on the three vectors on the right side of the following vector analysis equation: Raw Data = Data Average + Profit Signal + Noise. A vector is a set of numbers that is treated as a single entity. A vector defines magnitude and direction. A vector is best visualized as an arrow connecting one point in space to another. The vector analysis equation is much easier to understand when it is presented as a picture. The six edges of the tetrahedron shown in Figure 1 represent the six different ways of combining the three vectors on the right side of the equation. We call this stable geometric figure “the cornerstone of evidence.”
Figure 1 A vector analysis requires a minimum of three Generalized dimensions. An evidence-based decision evaluates three key vectors: 1) a Data Average, 2) a Profit Signal, and 3) Noise.
The cornerstone of evidence is even easier to grasp in its physical form. In Profit Signals you will learn how to build one with bamboo skewers and Sculpey Clay.® The construction process is fun and informative. The keys to making better, more profitable business decisions are (1) identifying and (2) interpreting the profit signals in your raw data. Profit signals are the most important element in any data-driven business decision. Vector analysis is the only way to identify profit signals. Knowing how to find and ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
11
graph your profit signals are extraordinary money-making skills. Relatively few people are aware of vector analysis and its universal relevance. Many do not understand the fundamental difference between a data matrix and a spreadsheet. Profit Signals fills in those educational gaps. With every chapter you will understand more clearly what profit signals are. You will soon know why they are invaluable. The break-even school of thought has dominated business decisions since 1918. That was the year G. Charter Harrison, a London accountant employed by Price, Waterhouse & Company, published “Principles of a Cost System Based on Standards” in Industrial Engineering magazine. 3, 4, 5 Since then Harrison’s accounting principles and procedures have become universally accepted. They are known today as costaccounting variance analysis.6
Figure 2 The break-even school of thought was founded by G. Charter Harrison in 1918. It is inherently onedimensional.
Dollars
������� (Averaged Expenses)
������
����
��������������� ����������������� ����������������
�� �� ged � �� era e) v (A com In
Product or Service Volume For example, Figure 2 illustrates traditional break-even point analysis. It assumes that average expenses and average income are perfect linear functions of volume. The lines cross at the “break-even point.” A break-even point, “cost-accounting variance analysis” is inherently one-dimensional. It is based on the difference between the two lines—the differences between average expense and average income at various volumes. Technically speaking, these differences are called predicted values. As ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
12
Premise shown in Figure 3, these differences form the vector at the back of the tetrahedron. It is but one of the six vectors required for a complete, three-dimensional vector analysis.
Figure 3 The differences between average income and average expense form one vector in the set of six required for a complete vector analysis.
Establishing performance standards and evaluating actual results in relation to them were important steps forward. Unfortunately, there was no cross-pollination between Harrison’s work in London and Sir Ronald Fisher’s simultaneous 1918 development of vector analysis in rural England. Instead, cost-accounting variance analysis evolved as a collection of one-dimensional methods. In break-even analysis, predicted values are used in isolation. In other accounting cases, the analysis is based solely on the raw data. Because the methods of cost-accounting variance analysis are inherently one-dimensional, it is impossible for any of them to produce a correct analysis. For example, the accuracy of any predicted value depends on the length of the noise vector. Equally important, the strength of evidence supporting any conclusion depends on a ratio involving the profit signal and noise vectors. This F ratio, as it is called, compares the length of the profit signal vector to the length of the noise vector. Modern textbooks teach cost-accounting variance analysis as a way to identify causes of profits and losses.7 According to one Top-10 business-school accounting text, a cost accounting variance analysis can “…decompose the total difference between planned and actual performance into elements that can be assigned to individual responsibility centers.”8
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
13
This sounds like a vector analysis. It is not. Cost-accounting variance analysis is just arithmetic. It is a set of onedimensional analysis methods incapable of distinguishing profit signals from noise. 21st Century teaching methods and computer graphics place vector analysis in its rightful position in business decisionmaking. In the past, we had to wade through volumes of bewildering algebra to analyze data. Few of us ever saw the cornerstone of evidence. Few of us were able to take evidence to our bottom line: “How can I personally use vector analysis to solve my problems and make my business more profitable?” The one formula we will use in this book is the Pythagorean Theorem. (If you now have a frown on your face, you probably learned about this idea in your favorite high school class, geometry.) This equation defines the right triangles comprising the cornerstone of evidence: The square of the long side of a right triangle equals the sum of the squares of the other two sides. In other words, c2 = a2 + b2 . The vast majority of evidencebased decisions are based on this simple formula. We call it the New Management Equation. We trust you will too. The president of a $500 million company put it this way: “In old-school cost accounting we determine the variance and there the analysis stops. With the New Management Equation we determine the variance and there the analysis begins.” This vector analysis is represented by the forward right triangle in Figure 4 . Vector analysis is transparent. Transparency implies the full disclosure of all elements. Transparency is indispensable to evidence-based decisions. It is a desirable accounting quality. Cost accounting variance analysis lacks transparency. It hasn’t changed one whit since the day it was born during the 20th Century’s “scientific management” craze. Like the whalebone corsets of that era, it is a constricting artifact. It creates an outward appearance of propriety while it conceals covert improprieties. It suppresses five-sixths—83 percent—of the accounting and analysis information that is contained in raw data. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
14
Premise
Figure 4 Cost accounting variance analysis ends with variations from standard values. A vector analysis begins with variations around the data average. The variations vector is then broken up into two components: 1) the Profit Signal vector and 2) the Noise vector.
The differences between a vector analysis applied to a data matrix and a spreadsheet analysis applied to arbitrary clusters of numbers are irreconcilable. The more you know about the cornerstone of evidence, the more you will understand how using just one of six possible vectors can misrepresent evidence and damage profitability. Our indictment is harsh. We will help you challenge it. Then we will ask that you vote in favor of full disclosure and transparency.
The Parable of the Paper Bags We wrote Profit Signals for business leaders who are resolute competitors. Competitors, and we both belong in this class, are human. Therefore, all of us face the same challenges when we tackle the Six Sigma body of knowledge for evidence-based decisions. One of our novice students shared a personal story on a first day of training. We use her parable of the paper bags in our Six Sigma decision courses. “What you are teaching us is a new skill that is hard to grasp. This process reminds me of the way my grandfather ran his business. “I loved my grandparents very much. I was very close to ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
15
them. My grandfather was born in Sylva, North Carolina in 1893. He had to leave school in the second grade to go work on a tobacco farm. During the Great Depression, he and his wife couldn’t earn a living. So they packed up their car and traveled across country to Darrington, Washington. With his brother’s family, there were 13 of them altogether. “He originally worked for the Sauk Logging Company. But, he preferred to work for himself. When the Federal government bought his parcel of land in North Carolina for an addition to Smokey Mountain National Park, he took the money and bought property here in Arlington. “My grandparents planted 80 cherry trees. They had 10 acres of raspberries and a five-acre garden. They sold produce. There were milk cows and always some beef. Grandpa bought and sold heifers. He split cedar shakes in his spare time to supplement the family income. “There were all kinds of transactions. But, my grandfather didn’t know how to multiply. Instead, he kept all his receipts in different brown paper bags. Once a month he would arrange these bags in the living room. Then he would add up columns of numbers so he would know what to charge people. “I learned how to multiply in third grade. I got pretty good at it by the fourth grade. I have always found math to be difficult. I still do. I have to work at it and I really would rather do something that comes easy. “One day, I think it was in 1959, I came home and proudly told him I could teach him to multiply. By multiplying he wouldn’t have to spend so much time with his paper bags. He listened and he learned how to do a few simple problems correctly. After he picked it up he went right back to using his paper bags. He never trusted multiplication. He couldn’t bring himself to believe in this new-fangled way of doing things. “His system worked, but gosh, what he sacrificed.” There is no doubt about it. The break-even thinking of cost-accounting variance analysis works. But, the testimony of Arthur Andersen, Cendant, Coca Cola, Enron, Rite ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
16
Premise Aid, WorldCom, and other companies of former greatness, suggests that the way it works is costly. It would be a generalization of the non-mathematical, nonscientific variety to claim that vector analysis is the solution to problems of this magnitude. Nevertheless, we propose that vector analysis, and the vector analysis mind set, are in fact important parts of business decision solutions. We ask you to critically evaluate this proposal. Vector analysis theory and tools are to multiplication as multiplication is to addition. They are valuable time savers. Power and beauty are their strengths, but also their Achilles heel. Though these ideas do not intimidate children, they can threaten adults. As you and your colleagues work to master evidence-based decisions, do not be surprised if you observe anger, denial, bargaining, and depression. Anticipate this roller coaster. At the end, we hope you will arrive at acceptance. This cycle accompanies any and every substantive life change. Negotiate and get to “Yes” with your peers. Get to yes with your executives and those you lead so that your company is not using brown paper bags to compete against a more powerful, efficient, effective, and profitable way of doing work.9 That way of doing work is a vector analysis applied to a data matrix. The Dollar Value of Evidence The quality of a manager’s decisions and consequent actions determine profit and loss. Since large sums of money are often at risk, managers weigh their “evidence”. They watch production. They pore over monthly spreadsheet reports. Usually, the most valuable information a manager has remains buried in the spreadsheets. It is further obscured by arithmetic totals, averages, differences, bar graphs and pie charts. As senior executives, and as consultants to senior executives, we have seen this process repeated month after month, year after year, decade after decade. With the pressure to produce
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
17
profits, it is easy to understand why many managers resort to an expedient device: appearances. The privileges of position— title, clothing, automobiles, office location and furnishings, social networks, and financial reward—can and often do persuade others that appearance is evidence. Evidence-based decisions call for a higher standard. Whenever a manager looks at the numbers used to measure the performance of an organization, certain questions ought to begin to perk:10 1. Should I believe these numbers? 2. What is the evidence in these numbers, and how strong is it? 3. What actions should I take based on this evidence? 4. What evidence will confirm that management actions produced the desired results? Because we are only human, these questions are accompanied by unsettling feelings and thoughts that nurture anxiety: a) I am comfortable with the way things are. b) This new knowledge puts my previous decisions in a bad light. c) I don’t want to lose my job. You are probably reading this book because you have made business decisions. Some of those decisions were good. Some were bad. Some were based on evidence; others were not. We ask you to contrast the profit related to good decisions with the loss related to bad ones. The difference between these two numbers forecasts the initial Return on Investment (ROI) you can expect from reading this book. Because we are all human, evidence and ROI may not be enough. We must handle fear. We know this from serving individual and corporate customers in virtually every industry, in Australia, Brazil, England, Mexico, New Zealand, Singapore, Uruguay, and 44 of the United States. It may help you to know that, in every case where our students have used the information we present, ROI is at least 10:1. A more typical result is 50:1. In most cases, bottom line business results like these eventually break through the barriers of fear.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
18
Premise
Our clients welcome the opportunity to improve on their current methods of analysis and decision-making. Nevertheless, we all have a natural aversion to change. The greater the change is, the greater our aversion. Six Sigma companies have made a conscious decision to conquer their reluctance. Experience, evidence and most of all, competition are forcing all of us to improve.11 The process of making an evidence-based decision is elegantly simple. It is not new, yet it is profound. It has demonstrated its ability to improve productivity and profitability in every industry. The one, two-, three- and n-dimensional profit signals waiting to be discovered in your raw data can provide practical, profitable solutions to even the most complex, confounding and challenging business problems you face.
Six Sigma In today’s popular press, evidence-based decisions are known as Six Sigma (6σ).12 Bill Smith, an engineer at Motorola, conceived Six Sigma in 1986. This major step forward has produced trillions of dollars in profit. Each breakthrough spurs demand for further, more dramatic breakthroughs. This is natural and good. The iterative nature of the Six Sigma project cycle has taught us which parts of Six Sigma are essential. We, and other experienced professionals in the field, also have learned which parts are extraneous. The demand for additional, rapid, dramatic breakthroughs can be satisfied only if we trim fat from Six Sigma’s middle-aged spread. Six Sigma made Profit Signals possible. We return the favor by showing how to flex the evidence muscle without carrying the weight of bureaucracy. Two fundamental Six Sigma concepts, the data matrix and vector analysis, have never been explained to anyone’s satisfaction in any previous publication. Since 1986, Six Sigma has been based on two seemingly reasonable assumptions: ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
19
1. It is impossible to teach Six Sigma theory to everyone in a company. 2. Six Sigma tools are too difficult for most people to use. We have discovered these assumptions are no longer valid. Personal computers and software have changed the world. Today’s requirements for Six Sigma leadership are simply these: a) A passionate aptitude for pursuing the truth in a system. b) An understanding of the nature of a physical law or Generalization. c) The ability to operate carefully chosen statistical software. Anyone and everyone can learn this unifying theory. They can learn it quickly. Anyone can master what is called the Black Belt Body of Knowledge (BOK). Based on our experience, and with the support of senior management leadership, this process can be accomplished in 10 to 16 days. This is the path we take and the case we make in Profit Signals. You will learn fundamentals quickly. You will immediately be able to use what you learn to make evidencebased decisions. Improved decisions can lead you and your company to Six Sigma profits. If your enterprise is to succeed, its products and services must exceed the great expectations of fickle customers. Goods and services must be able to withstand the scrutiny of the free press, and even an investigative Senate sub-committee. If you expect your business to meet these objectives, your enterprise must embrace and leverage the power of evidence-based decisions. Profits are always in fashion. Six Sigma is a very classy way to earn them. The underlying principles of a data matrix and vector analysis are timeless style. For over 2,000 years, the New Management Equation ( c2 = a2 + b2) has helped people make money from measurements. The supporting evidence for this claim is overwhelming. It is well beyond any shadow of doubt. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
20
Premise Pick a profitable 21st Century product, service, or sport. Any one will do. The qualities of almonds, aviation, agriculture, beer, computers, electricity, electrocardiograms, fast food, global navigation systems, gourmet ice cream, magnetic information media, movies, music, oil, Olympic gold medal speed skating blades, pharmaceuticals, roller ball pens, surgery, skiing, scuba diving, telecommunication, textiles, windows, and X-treme competition all share a common bond. Breakthroughs in every one are driven by disciplined observation, measurement, the recording of data in an orderly data matrix fashion, and analysis. The cornerstone of evidence has stood the test of time. Its principles run deep and far beyond rote, routine, mainstream business thinking.13 We welcome you to the world of the data matrix, vector analysis, standards of evidence, improved business decisions, and the New Management Equation. Welcome to the universe of Profit Signals.
How to Read This Book You can speed read this book in about a week. To get the “big picture” quickly, skim the illustrations. Read the captions to these exhibits. “Closing arguments” at the end of each chapter summarize the key content. After this initial overview, you may want to read it again at a more leisurely pace. The ideas, analogies and activities are presented in a particular sequence for good reason. So it is best to read the book front to back. If possible, complete the suggested experiments as you go. Feel free to collaborate on these with colleagues, friends, family members, neighbors— even your old high school teachers. Chapter 1: The Five-Minute PhD – The opening chapter
lets you earn your PhD in evidence-based decisions. It takes only five minutes. Your Five-Minute PhD grants you the power of vector analysis. Call it vector power if you will. Once you get a handle on profit signals, you will be able to systematically quantify and prioritize the effects of multiple factors on any manufacturing, health care, service or financial process. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
21
Chapter 2: Standards of Evidence – We review the
distinction between story telling and evidence. You will learn the difference between vector analysis applied to a data matrix and spreadsheet calculations applied to arbitrary clusters of numbers. This chapter’s inside joke and secret handshake are that a Greek named Pythagoras invented the “New Management Equation” 2500 years ago. (The New Management Equation is easier to say and spell than Pythagorean Theorem. It is also sweeter to the ear.) In the early 1920s a genius named Ronald Fisher discovered how to apply the New Management Equation to identify profit signals in raw data and quantify strength of evidence. Fisher’s method is the international standard for quantitative analysis in all professions save two: accounting and business management. In this chapter we trace the history of the costaccounting variance analysis and show how to improve it with vector analysis. Chapter 3: Evidence-based Six Sigma – If you are new to Six Sigma, this chapter has all the basics you need to know. It reviews the traditional Six Sigma tool set. We cover organizational guidelines, project selection criteria, process maps and financial model graphs. We review the traditional Six Sigma breakthrough project cycle: Define, Measure, Analyze, Improve, Control (DMAIC). Chapter 4: Case Studies – We each have more than 20 years of consulting experience in the field of evidence-based decisions. Our results have been published in peer-reviewed textbooks. CEOs, middle managers and line workers have signed affidavits and testified to the value of our work.14 Dr. George E.P. Box, a Fellow of the Royal Society and elected member of the Academy of Arts and Sciences, endorsed one of our three-dimensional analysis books in 1997. 15 In this chapter, we tell a few of our favorite breakthrough project stories. These include:
•
Improving the quality of state government customer services with a $525,000 pay off.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
22
Premise •
Reducing the days in accounts receivable by 30 days with a 14-day project and a $425,000 bottom line impact.
•
Improving the operations of a hospital’s Emergency Department (ED) with a gross margin of $18 million for a 38.2 percent gain over the prior year’s performance.
•
Dramatically improving the patient outcomes in cardiovascular surgery while putting $1 million in additional profits on a hospital’s bottom line.
•
Tool grinding breakthroughs worth $900,000.
•
Doubling the productivity in a vinyl extrusion process while reducing the product material costs by 50% in three months time. Bottom line value for this company for each of the next three years is $1 million, equaling a grand total of $3 million.
Chapter 5: Using Profit Signals – This chapter presents
the fundamentals of vector analysis with a few pages of reading and a physical model. You will learn how to expose the profit signals in your own data and represent them with bamboo skewers and Sculpey Clay®. You will tackle the following challenges facing the Corrugated Copter Company: • • • •
Establishing baseline performance metrics. Comparing two ways of doing things. Comparing three ways of doing things. Comparing 256 different ways of doing things.
Chapter 6: Predicting Profits – Corrugated Copter
managers want to be able to accurately predict flight times, profits, costs, inventory, and other important things. If they could improve the quality of their predictions, they could confidently take better advantage of market dynamics. Fortunately, this management team is up to speed on regression analysis. Is this something new and different? No, it is just another way to use the New Management Equation. It is vector analysis applied to a data matrix. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
23
Chapter 7: Sustaining Results – At Corrugated Copter
best business results always means earning the greatest revenue with the least expense. This is more than a politically correct platitude. It is responsible stewardship. The team learns to monitor and perfect their production processes through process capability studies and process control charts. Are these new and different? No. They are just other ways of using the New Management Equation. They are vector analysis applied to a data matrix. Chapter 8: The Three Rs – In its time, the Six Sigma
business initiative created new breakthroughs in quality, productivity and profitability. Corrugated Copters now believes traditional Six Sigma organizational ideas are outdated. A team of Corrugated Copters leaders has proposed an education system that would render their Six Sigma bureaucracy obsolete. Appendices – Here you will find a glossary of Profit Signal
terms that will help you learn the language of evidence-based decisions. There is a complete bibliography of the essential evidence-based decision bookshelf. We have also included a Profit Signals Black Belt Curriculum and production information on the Six Sigma tools we used to write and produce this book. We trust this information will serve as your outline for future study.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
24
Premise Endnotes Box, Joan Fisher. R.A. Fisher, Life of a Scientist. John Wiley & Sons, New York. 1978. 1
Box, Joan Fisher. R.A. Fisher, Life of a Scientist. John Wiley & Sons, New York. 1978. 2
Harrison, G. Charter. Cost Accounting to Aid Production – I, Standards and Standard Costs, Industrial Management, The Engineering Magazine, Volume LVI, No. 5, October, 1918. 3
Harrison, G. Charter. Cost Accounting to Aid Production – II, Standards and Standard Costs, Industrial Management, The Engineering Magazine, Volume LVI, No. 5, November, 1918. 4
Harrison, G. Charter. Cost Accounting to Aid Production – II, Standards and Standard Costs, Industrial Management, The Engineering Magazine, Volume LVI, No. 5, December, 1918. 5
Johnson, H. Thomas, and Kaplan, Robert S. Relevance Lost, The Rise and Fall of Management Accounting. Boston, Harvard Business School Press, 1987.
6
Garrison, Ray H. and Noreen, Eric W. Managerial Accounting, 10th Edition. Boston, McGraw-Hill Irwin, 2003. Page 431. 7
Anthony, Robert N., and Reece, James S., Accounting: Text and Cases, Eighth Edition. Homewood, Irwin, 1989. Page 941. 8
Fisher, Roger, and Ury, William, Getting to Yes, Negotiating Agreement Without Giving In. New York, Penguin Books. 1981.
9
Royall, Richard. Statistical Evidence, A likelihood paradigm. New York, Chapman & Hall. 1997.
10
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Premise
25
Shewhart, Walter A. Nature and Origin of Standards of Quality. The Bell System Technical Journal. Volume xxxvii, number 1, January, 1958.
11
Six Sigma is a registered trademark and service mark of Motorola Incorporated. The Motorola web site is a recommended resource for researching this history of Six Sigma. For a summary overview please read: Barney, Matt, “Motorola’s Second Generation,” Six Sigma Forum Magazine, May 23, 2002, pages 13-16. http://mu.motorola.com/pdfs/Mot_
12
Six_Sigma.pdf
Fuller, F. Buckminster. Critical Path. New York, St. Martin’s Press. Page 8.
13
Sloan, M. Daniel and Torpey, Jodi B. Success Stories on Lowering Health Care Costs by Improving Health Care Quality. Milwaukee, ASQ Quality Press. 1995.
14
Sloan, M. Daniel. Using Designed Experiments to Shrink Health Care Costs. Milwaukee, ASQ Quality Press, 1997.
15
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
26
©
Premise
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 1 The FiveMinute PhD
P
hD’s, medical doctors, scientists, engineers, mathematicians, statisticians, economists, managers, and executives don’t own the lock and key to data analysis. Anyone can learn to do a vector analysis. What was once the high water mark of postgraduate study is now as simple as a Google web search. You don’t need a certificate on your wall to analyze data. Knowledge and its application are the taproots for professional stature and income. Few are eager to debunk the presumption of specialized knowledge that justifies position and paycheck. So long as information and knowledge remain shrouded by jargon, professions and authority remain secure. Knowledge and information challenge authority. They can be disrespectful of bureaucracy and hierarchy. They applaud the pointed question. They reward the cross-examination of high priests and presidents. The Five-Minute PhD is a democratic degree that exemplifies our age. It can be earned by anyone who is willing to work at it. With knowledge and information, people can and do effectively solve more of their own problems. Solving one’s own problems saves time and money. This is fun. Companies that know how to solve problems quickly make more money than those that don’t. We all acquire memories through the experience of our everyday lives. Memories make life rich and rewarding. Memories can teach, but they rarely bring innovation to our ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
28
Five-Minute PhD work places. Except for an occasional stroke of dumb luck, we get no new knowledge from casual experience. With the sole exception of pure mathematics, we obtain new knowledge only by applying the basic disciplines of experimentation, observation, and analysis. Companies that apply knowledge and intelligence make more money than those that depend on memories. You will now learn how these three basic disciplines— experimentation, observation, and analysis—work, and prove that they do work, in less than five minutes. Neither previous experience nor training nor calculations are required. Start Your Stopwatch Now Raw data contain information. In our digital world all information can be, and is, turned into numbers. Good information leads to reliable predictions. Telephones work, airplanes fly, music is played, products are manufactured, medical treatments are rendered and services are delivered, all through the use of numbers. Experimental data, measurements, come from disciplined observation by design. When time and money are valued, experiments must be sized with economy in mind. Table 1 arrays the eight observations in an economical experiment. The first column in the array labeled “Experiments Called Runs,” establishes the order in which eight observations were made. The factors1 in this three-dimensional experiment are gender (x column), backpack weight bearing (y column), and activity (z column). Heartbeats is the measured response (dependent variable) in the experiment. For convenience, each factor was set at only two levels. The high setting is coded +1. The low setting is coded -1. Table 1 contains all eight possible combinations of a three-factor experiment with two levels for each factor. This is called a 23 (two raised to the third power) experimental design. “Two raised to the third power” is a mouthful, so it is usually pronounced, “two cubed.” Think of tea with two cubes of ice, rather than an equation, and the idea will be more refreshing.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
Table 1 The cube or “design of experiments” (DOE) array is an ideal data matrix.
29
As you can see in Figure 1, this array does in fact create a cube. Come back to this illustration after you complete your PhD. The clock is ticking.
Figure 1 The ideal data matrix forms a cube.
For each of the eight experiments, or runs, the resulting number of radial artery heartbeats were measured and recorded. (You can feel your own radial artery pulse by touching the inner aspect of one of your wrists with the index and middle fingers on the opposite hand.) These ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
30
Five-Minute PhD measurements are arrayed in the far right column of this data matrix. Each of the eight response measures is the output from a unique combination of the three factors. For example, the measurement 70 for Run #1 was for a sitting man who had no weight in his backpack. This is an example of “disciplined observation.” Please turn your attention to the pattern of the response heartbeat measurements in the far right column. Consider the following questions: • • • •
Which combination of variables produced the two highest heartbeats rates? Which combination produced the two lowest heartbeats rates? Which variable appears to have the least effect on heartbeats? How would you predict future outcomes from similar experiments?
Please pause now and stop reading. Go back. Take time to look at the evidence patterns in the data matrix. When you have answered all four questions in the list, continue. Are you finished? Check your watch. We predicted that you could successfully complete a three-dimensional, doctorallevel vector analysis in less than five minutes. We bet we were right. If you concluded that aerobic exercise has the strongest effect on heartbeats, you have earned your Five-Minute PhD! If you noticed that carrying a 50-pound weight in a backpack increases the number of heartbeats for aerobic activity, but not for sitting, you are at the top of your class. If, in addition, you concluded that gender doesn’t really make much of a difference when it comes to the number of heartbeats, you graduated Cum Laude. Business Art and Science By using a special kind of row and column array called a data matrix, you were able to correctly identify one main effect (activity), one inert factor (gender) and a two-factor interactive effect (the combination of activity and backpack weight). Your eyeball vector analysis of eight numbers was accurate. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
31
Consider the economies of using this technique to solve business problems. Imagine the possibilities. Yes. You are absolutely right. With relatively small amounts of data framed in a data matrix, you can simultaneously quantify and prioritize the effects of several process variables. This applies to any manufacturing, business, service or health care process. This is the vast Generalization of the mathematical/scientific variety we mentioned. In other words, it is true. The eight numbers in the far right hand column of the 23 data matrix in Table 1 actually form a single entity. This entity is an eight-dimensional vector! You have now entered hyperspace. Science-fiction writers use the mathematical term hyperspace when they need a word to describe faster-than-light travel. Hyperspace is actually the mathematical term for a space with four or more dimensions. So, fasten your seat belts. Please keep your hands and arms inside the analysis rocket. You succeeded in analyzing the three-dimensional experiment in Table 1 because you were able to visually compare the eight-dimensional vector for heartbeats to the eightdimensional vectors for activity, gender, and backpack weight. These vectors are the basis for profit signals. We explain the details in Chapter 5. Hyperspace is not as easy to accept as a free ride on Disney’s Space Mountain. For most of us, visualizing more than three dimensions is out of the question. The hallmark of Ronald Fisher’s genius was his ability to visualize n dimensions.2 This was Imagineering at its very finest. The evidence we now have about our universe confirms that Fisher’s vision of ndimensional, hyperspace was correct. Fisher’s vector analysis is the elegant simplicity that underlies myriad, seemingly unrelated analysis techniques. Nowadays, inexpensive software effortlessly applies vector analysis to any data matrix. Software automatically calculates the profit signals. It ranks them by importance and determines the strength of evidence. Then, with the grace of a high technology thrill ride, software applications create threedimensional graphs annotated with accurate predictions.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
32
Five-Minute PhD For example, repeated experiments with the setting of (-1, -1, -1) or (Male, No Weight, Sitting) will produce an average heartbeat of about 71.25. This predicted value labels the lower, left, front corner of the cube in Figure 2.
Figure 2 Vector analysis applied to a data matrix gives analysts the power of three-dimensional graphics. The differences in appearance between this illustration and richer ones elsewhere in our book can be explained: the superior tables and illustrations were created using vectors.
������
�����
�����
���
��
������
������
������
�������
�
Activity
����� ����
����� ������
�������
������
Repeated experiments at the setting of (+1, +1, +1) or (Female, 50-pound weight, Aerobics) will produce an average heartbeat of about 188.75. This predicted value labels the upper, right, back corner of the cube in Figure 2. New vector analysis users are often amazed at the accuracy of predictions based on cubic and higher-dimensional experiments. You will discover this for yourself as you complete the exercises in this book. You intuitively used data matrix and vector analysis principles to interpret the data in the Five-Minute PhD experiment. Working at the English Rothamsted Experimental Station, Ronald Fisher conducted the first cubic and higherdimensional experiments in 1919. He applied these principles to solve difficult, important problems using small, economical sets of data. William Gosset, the student of Fisher who first conceived the theory of statistical inference for small samples in 1907, used these statistics at the Guinness Brewery. Those who enjoy a stout beer now and again have been thankful ever since.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
33
Eighty years of revolutionary advances in agriculture, biotechnology, communications, computing, finance, information technology, manufacturing, medicine, space technology, and transportation support Fisher’s mathematical/ scientific Generalization. Vector analysis applies to everything. Vector analysis is used to coordinate all commercial jet landings at Orlando International. It is used to describe Einstein’s special and general theories of relativity. It is used to graph voltage variations resulting from the depolarization and repolarization of the cardiac muscle. 3 It ought to be used to create financial statements. It is proven. It is practical.4 Frank Netter, the Norman Rockwell of medical illustration, drew a beautiful picture of a vector analysis (Figure 3). In Netter’s drawing, the x-, y-, and z-axes are labeled using medical terminology.
Figure 3 The 23 cube you used to analyze heartbeats is identical to EKG theory and Rh+/Rh- blood groups. The plus (+) and minus (–) signs for Rhesus blood groups symbolize vector analysis reference points.
The x-axis refers to the sagittal, or side planes, of the body. The lower, horizontal y-axis is illustrated while the upper y-axis plane is implied. The back “frontal” plane is the illustrated z-axis plane. This physiological phenomenon is called a spatial vectorcardiographic loop. It was used in 1908 by Willem Einthoven to create the electrocardiogram (EKG).5 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
34
Five-Minute PhD The EKG made it possible to observe, measure, and graph the heart’s electrical impulses over time. The patterns that emerge from the EKG vector analysis are critical to the prediction of a beating heart’s behavior. Knowledge produced by this Nobel Prize-winning achievement led to the creation of the most profitable niche in American medicine—cardiovascular care. Revisit this illustration when you read the Six Sigma case study on “beating heart” Coronary Artery Bypass Graft (CABG) surgeries in Chapter 4. Table 2
lists a few of the thousands of proven, profitable applications of vector analysis to a 23 data matrix. Consider the vast Generality, and the enormous profit potential of this single tool. The only limitation is imagination. So imagine. Take time to write down factors (inputs) and responses (outputs) that could help you make more money. Once you have performed the disciplined observations and recorded the measurements demanded by the cube’s data matrix, the eight-dimensional vectors—especially the allimportant profit signals—will lead you directly to the most profitable solution.
Table 2 The cube experiment works for any process in any system.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
35
Profit, loss, productivity, inventory turns, taxes, time, and sales volume—any response you can measure—depends on many factors and the interaction of those factors in your business. Some of these are under your control; some are not. Complexity is the rule, not the exception. The only way to distinguish profit signals from noise is to apply vector analysis to a data matrix. The ratio of the length of the profit signal vector to the length of the noise vector quantifies the strength of evidence in the data. A “long, strong” profit signal vector and a “short, weak” noise vector indicate large, statistically significant effects and reliable predictions (Figure 5).
Figure 5 A long, strong Profit Signal with a short, weak Noise vector means there is a statistically significant effect.
A “short, weak” profit signal vector and a “long, strong” noise vector indicate small, statistically insignificant effects and unreliable predictions (Figure 6).
ATIO
DA TA AV E
NOISE
RA
VAR I
Figure 6 A long, strong Noise vector and a short, weak Profit Signal indicate no statistically significant effect. The variation is most likely due to Chance.
N
RAW DATA
GE
IT OF AL R P GN SI
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
36
Five-Minute PhD Data in the matrix must be obtained through a process of disciplined observation. Trial and error is expensive, timeconsuming, crude, and ineffective. It is not a viable business strategy for the 21st Century. The cube is three-dimensional; it has three factors. It has served as a keystone of professional knowledge and profitability since 1630 when Rene Descartes introduced the method for three-dimensional thinking.6,7 Despite the fact that we inhabit a world of three physical dimensions, there is no reason to limit ourselves to three-
Table 3 Multi-dimensional experiments improve profits in every industry.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
37
dimensional experiments. To illustrate, Table 3 lists a few ndimensional experiments with n ranging from 2 to 6. Disciplined, multi-dimensional observations and vector analysis reduce the financial risk associated with every important business decision. Business leaders must not excuse themselves from mastering this knowledge or the skills to go with it. To do so is to gamble the future of their companies on needlessly risky decisions. Senior managers and corporate directors are knowledge workers. They, more than any other members of an organization, need to know how these tools work. Typically, they can acquire this knowledge in just four days of accelerated, hands-on training. Profit Signals Only a few years before Fisher used the cube and higherdimensional experiments to dramatically increase profitable crop yields in England, Pablo Picasso and George Braque created a new art form called Analytic Cubism. The analogies between Picasso’s and Fisher’s cubes are intriguing. Picasso and Braque aimed at presenting data as perceived by the mind rather than the eye. “Every aspect of the whole subject is seen in a single dimension.”8 In Picasso’s original Analytic Cubism, “objects were deconstructed into their components…. it was used more as a method of visually laying out the FACTS…”9 Fisher explained his model and methods using virtually identical words. He referred to vector analysis as the Analysis of Variance. In Fisher’s terminology, Variance is a statistical measure based on the squared length of the variation vector. In cost accounting, a variance is a difference between an actual value and a standard or budgeted value. Fisher’s definition pre-dates the accounting definition by forty-some years. We will discuss this further in Chapter 2. Six Sigma Black Belts, high school teachers, college professors, statisticians, spreadsheets and statistical programs often employ the hideous acronym ANOVA for Analysis of Variance. We are stuck with it, so we use it. An ANOVA ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
38
Five-Minute PhD “deconstructs” a data vector into the basic pieces essential for evidence-based decisions: Raw Data = Data Average + Profit Signal(s) + Noise We will discuss and illustrate various aspects of this vector equation in subsequent chapters. For now, we focus on the component of greatest immediate interest, the Profit Signal. Table 4
contains a coded version of the Five-Minute PhD cube experiment. Actual factor names are represented as generic X, Y and Z variables. The numbers –1 and +1 are traditionally used to designate low and high levels of each factor. Actual heartbeat data are used in the response column. As a Generalization, these could be any measurements of interest to you.
Table 4 The coded version of the Five-Minute PhD cube experiment.
As we saw in Figure 1, the three factors in a cube experiment form the edges of a three-dimensional solid, a cube. In this sense, a cube experiment is three-dimensional. However, the data matrix for a cube experiment has eight combinations. The response column contains eight measurements, one for each combination. These factor combinations and measurements correspond to the eight corners of the cube. In other words, here comes hyperspace again.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
39
The corners of the cube give a three-dimensional representation of an eight-dimensional data vector. Acquiring the knowledge and mastering hyperspace analysis skills is well within everyone’s intellectual reach. Accepting this beyond-belief reality is easier said than done. Consider what Fisher’s ingenious, genius-level model suggests. Hyperspace—the real one rather than the realm of Luke Skywalker—is a very big idea. It is beautiful art and art is the dream of a life of knowledge.10
Figure 7 The opposing planes represent the eight-dimensional profit signal vector for the overall effect of factor Z. These planes correspond to the grouping of the eight response measurements shown in Table 5. This is a case of a “long/strong” Profit Signal and “weak/short” Noise. .
We can use the cube to create three-dimensional representations of the eight-dimensional profit signal vectors in a cube experiment. For example, Figure 7 shows shaded, opposing planes corresponding to the –1 and +1 levels of the most important factor Z, Activity. These two opposing planes represent the eight-dimensional profit signal vector for the overall effect of factor Z, Activity. This main effect is defined as the difference in the average response values for Z = -1 (Sitting) and Z = +1 (Aerobics). Overall Z effect = (Average of 140, 136, 180, 190) minus (Average of 70, 68, 86, 88) = 161.5 - 78.0 = 83.5
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
40
Five-Minute PhD
Table 5 This grouping of the eight response measurements corresponds to the opposing planes in Figure 7. This is a case of a “long/strong” Profit Signal and “weak/short” Noise.
. Table 5
displays in a spreadsheet format the two groups of measurements from the original Five Minute PhD experiment corresponding to the opposing planes in Figure 7. If Z (activity) had no effect, the average response on the front plane of the cube would roughly equal the average response on the back plane. This is not the case. Z had a large effect. The average of the back plane is 83.5 heartbeats larger than the average of the front plane. In the Five-Minute PhD experiment, activity and backpack weight had noticeable effects on heartbeats. We reached these conclusions by comparing the column of response measurements in the data matrix to the columns representing the factors. In actual practice, we must also quantify the strength of evidence for each profit signal. We will discuss this, and the related concept of statistical significance, in Chapter 2.
Figure 8 Opposing planes representing the eight-dimensional profit signal vector for the main effect of factor Y.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
41
Figures 8
and 9 show the shaded, opposing planes representing the profit signal vectors for the main effects of factors Y and X. If you think about Star Wars while you mull over these images, they will be more entertaining. Plus, believe it or not, people do get up and cheer at the end of a multimillion-dollar breakthrough Six Sigma project, just like they do when good guys finally win on the big screen.
Figure 9 Opposing planes representing the eight-dimensional profit signal vector for the main effect of factor X.
When the effect of one factor depends on the level of another, they are said to have an interactive effect. For example, in the Five-Minute PhD experiment, activity and backpack weight had an interactive effect. Increasing the weight affected the number of aerobic activity heartbeats, but not the number of sitting heartbeats. Pairs of planes on the cube can also represent interactive effects, but they are perpendicular rather than parallel. For example, one of the perpendicular planes in Figure 10 contains the four corners where X and Z have opposite signs (X × Z = -1). The other plane contains the four corners where X and Z have the same sign (X × Z = +1). These planes represent the eight-dimensional profit-signal vector for the interactive effect of X and Z, defined as the difference in the average response values for X × Z = -1 and X × Z = +1. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
42
Five-Minute PhD If X and Z had no interactive effect, the average response on each plane would be the same. If X and Z had a large interactive effect, one plane would have a much larger average than the other.
Figure 10 Perpendicular planes representing the eightdimensional profit signal vector for the interactive effect of factors X and Z.
Figure 11
shows the shaded, perpendicular planes representing the profit signal vector for the interactive effect of factors X and Y. Figure 12 shows the shaded, perpendicular planes representing the profit signal vector for the interactive effect of factors Y and Z.
Figure 11 Perpendicular planes representing the eightdimensional profit signal vector for the interactive effect of factors X and Y.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
43
Figure 12 Perpendicular planes representing the eightdimensional profit signal vector for the interactive effect of factors Y and Z.
These familiar geometric models convey esthetic beauty and analytical power. This is not the 10th grade geometry taught by Mr. Greismeyer at Centerville High. This is the geometry of Michelangelo, da Vinci, Galileo, Einstein, Guglielmo Marconi, Orville and Wilbur Wright, Alexander Calder, and a Fellow of the Royal Society named Sir Ronald Fisher. Data Recycling The three-dimensional cube diagrams provide a looking glass into eight-dimensional hyperspace. They allow our threedimensional eyes to make some interesting observations. For example, have you noticed that all the corner values are used repeatedly? Every data point appears six different times, once in each of the six profit signal vectors shown above! This is a lot of work for only eight little numbers to do. This data-recycling phenomenon is a characteristic of all cubic and higher-dimensional experiments. For most companies, profit signals mean they can eliminate at least 83% of the data collection and storage costs incurred with primitive trial and error methods. Eighty-three percent is not a typographical error. It is simply a fact. The larger the number of factors, the greater the savings. This bottom-line result is enhanced by the fact that vector analysis gives you the right answers to your most pressing business problems. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
44
Five-Minute PhD
The Full Circle of Data Discovery We are back to where we started in the Premise. The data matrix and the New Management Equation, c2 = a2 + b2, form the backbone of vector analysis. Once data are entered into a data matrix, a computer automatically calculates the profit signals, determines the strength of evidence and graphs the predicted values. It even explains the results. Since 1920, vector analysis has been the path to credible, quantitative evidence. Vector analysis points to the entire family of common statistical distributions. It was the foundation of science in the 17th Century when Galileo proved that the earth revolved around the sun. It was the foundation of science at the turn of the 20th Century when Einstein created his special and general theories of relativity.11 It is the foundation of science at the turn of the 21st Century. Since 1935 the application of vector analysis to data matrices, better known to college students as ANOVA, has produced huge financial returns in agriculture, manufacturing, engineering, health care and process industries.12 ,13 ,14 ,15,16 In those days this tool set was called the Design of Experiments. In the 1980s it was repackaged as one of many Six Sigma tools. In 2003, it is simply Profit Signals that come from the New Management Equation.
The New Management Equation Use your own imagination to conduct experiments to verify the insights you gained from your Five-Minute PhD. We encourage you to use analogies to accelerate your learning. An analogy illustrates similarities between ideas or things that are often thought to be dissimilar. To quote one of our past students, “An analogy is like a comparison.” We have found that analogies, parables and old-fashioned story telling are the most effective tools for teaching people ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
Table 6 Like any true Generalizaion, the cube experiment is a Law of the Universe. Have some fun practicing with your new PhD in universes of your own.
45
the principles of evidence-based decisions. Yes. This is yet another paradox in evidence-based decisions. After you have completed your experiments with family, friends, and colleagues at work, discuss the implications of these analogies. Since you are now a PhD, feel free to throw around phrases like “Hegelian Dialectic” during your conversations. This crowd-pleaser will let other doctors of philosophy know that you know what you are talking about. Table 6
lays out two proven favorites. Specifically, how can you use what you now know about a data matrix and vector analysis to save time and/or make more money?
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
46
Five-Minute PhD Closing Arguments The following testimonies were transcribed in various historical hearings and trials about the Five-Minute PhD. Archimedes: “Eureka. I have found it. I have found it. By
comparing the weights of solids with the weights of equal quantities of water, I solved the mystery of specific gravity. By using Euclid’s magical formula, c2 = a2 + b2, one can transform the earth and the heavens into a vast design of intricate configurations. “Euclid, you made the impossible possible by the simplest of methods. But please isn’t there a shorter way of learning geometry than through your method?”17 Euclid: “Sire, in the country there are two kinds of roads— the hard road for the common people and the easy road for the royal family. But, in geometry all must go the same way. There is no royal road for learning. Now if I were to speculate a bit, and if there were a computing machine 2500 years from now that ran a vector analysis program with a data matrix, all might be able to travel an easier road.” Galileo Galilei: “Philosophy is written in this grand book the universe which stands continually open to our gaze. But this book cannot be understood unless one first learns to comprehend the language and read the letters in which it is composed. It is written in the language of mathematics, and it characters are triangles, circles, and other geometric figures without which it is humanly impossible to understand a single word of it.” Albert Einstein: “The Gaussian coordinate system of
Chance variation is a logical generalization of the X, Y, Z Cartesian coordinate system. I used the data matrix cube with a vector to suggest the passage of time in my 1916 best seller, Relativity, The Special and General Theory, A Simple Explanation that Anyone Can Understand.”18
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Five-Minute PhD
47
James Turrell is a hyperspace sculptor of international reputation. Mr. Turrell uses light in his search for mankind’s place in the Universe. James Turrell: “I want to create an atmosphere that can be
consciously plumbed with seeing like the wordless thought that comes from looking in a fire. I use the X, Y, and Z axes of light to achieve my objectives. I use the same Cartesian coordinate system to pilot my aircraft.”19 Walt Disney: “I only hope that we never lose sight of one thing—that it all started with a mouse. Born of necessity, the little fellow literally freed us of immediate worry. He provided the means for expanding our organization. He spelled production liberation for us. Disneyland will never be completed. It will continue to grow as long as there is imagination left in the world.”20
Endnotes These factors are also commonly known as independent variables. 1
Box, Joan, Fisher. R.A. Fisher, Life of a Scientist. New York, John Wiley and Sons, 1978. 2
Netter, Frank. The CIBA Collection of Medical Illustration, Volume 5, Heart. Commissioned by CIBA. 1969. 3
Netter, Frank. The CIBA Collection of Medical Illustration, Volume 5, Heart. Commissioned by CIBA. 1969. 4
Dubin, Dale. Rapid Interpretation of EKG’s, Edition V. Tampa, Cover Inc., 1996. Page 4. 5
http://www.rothamsted.bbsrc.ac.uk/pie/sadie/reprints/ perry_97b_greenwich.pdf 6
http://www-gap.dcs.st-and.ac.uk/~history/ Mathematicians/Descartes.html 7
8
http://www.ibiblio.org/wm/paint/tl/20th/cubism.html ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
48
Five-Minute PhD http://www.artchive.com/artchive/P/picasso_ analyticalcubism.html 9
Inscription on the southern ceiling of the rotunda leading to a James Turrell Skyspace installation at the Henry Art Gallery on the University of Washington campus.
10
Einstein, Albert. Relativity, The Special and General Theory, A Clear Explanation that Anyone can Understand. New York. Crown Publishers, 1952. Page 32.
11
Fisher, R.A. Statistical Methods for Research Workers, Thirteenth Edition. New York: Hafner Publishing Company Inc. 1967. 12
13
Fisher, Ronald A. The Design of Experiments. New York: Hafner Press, 1935. 14
Fisher, R. A. “Frequency Distribution of the Values of the Correlation Coefficient in Samples from an Indefinitely Large Population”, Biometrika, 10: 507-521, 1915. 15
Fisher, R.A. “On the Probable Error of a Coefficient of Correlation Deduced from a Small Sample.” Metron 1: 3-32, 1921.
16
Box, George E.P., Hunter, William G., Hunter, J. Stuart, Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building, New York, John Wiley and Sons, 1978. Thomas, Henry, and Thomas, Dana Lee. Living Biographies of Great Scientists. Garden City, Garden City Books, 1941. Pages 4-5.
17
Einstein, Albert. Relativity, The Special and General Theory, A Simple Explanation that Anyone Can Understand. New York, Crown Publishers, 1952. Page 90.
18
19
http://www.pbs.org/art21/artists/turrell/
http://goflorida.about.com/library/bls/bl_wdw_ waltdisney_quotes.htm 20
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 2 Standards of Evidence
W
hat are the objective standards of evidence your business uses to make decisions? We ask all new clients this question. Too often the answer is an uncomfortable silence or, “We’ve never asked ourselves that question before.” Evidence is the foundation for making better, more profitable business decisions. But evidence provides an operational basis for making decisions only if we have standards by which to judge the strength of evidence. Demands for improved financial performance put oldschool managers in a bind. For the first time in history, they are competing head-to-head with managers in developing nations like Brazil, China, India, Mexico, and Malaysia—take your pick. Labor is a tiny fraction of the total cost of doing business in these newly emerging competitive economies. As a result, North American businesses, large and small, must find ways to reduce production and delivery costs by at least 30 percent. Many must achieve this within the next five years or go out of business. If you doubt this possibility, visit Seattle, Washington. Since September 11, 2001 a good portion of our world-famous gridlock traffic jams vanished along with more than 30,000 jobs. For many old-school managers, staying inside their corporate cultural comfort zone, with an atmospheric vacuum of evidence standards, is more important than achieving any business goal, including profitability. This is a counter©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
50
Standards of Evidence productive and, given the painfully apparent need for jobs, a socially irresponsible attitude. Comfortable or not, spirited capitalism has put evidencebased decisions on the map. Whether they know it or not, vector analysis and standards of evidence are now on every manager’s radar screen. The only question is who will recognize and respond to the signals. Poetry versus Science Efforts to understand the world we live in began with story telling. Stories thrive in many forms, oral and written, poetry and prose. Stories convey values. They define and maintain cultures, including corporate cultures. Stories evoke fear, hope, joy, anger, sympathy, humility, respect, wonder and awe. Stories build like pearls around grains of historical fact. They tell us much, mostly about ourselves. Stories are not laws. They do not, and are not intended to, reliably describe historical facts or physical realities. Story telling does have its place, but it can be at odds with science. Story telling often involves tales of trial and error. Scientific discoveries inspire as much wonder and awe as any Paul Bunyan tale. But, the driving force behind science is disciplined observation, experimentation and analysis. The scientific method, which can be equated with Six Sigma, embraces affirmative skepticism. This skepticism is diametrically opposed to the innocence of credulity. Credulity, or as some prefer to say naïveté, is the suspension of critical thinking. Credulity allows us to experience the emotional impact of a good story. Credulity makes Disneyland, Disneyworld and the Epcot Center fun. The tension between story telling and science dates to poet John Keats’ criticism of scientist Isaac Newton’s prism. Newton discovered that “white light” contains an invisible spectrum of colored light. He made this spectrum visible by shining ordinary light through a prism. The process Newton used is called refraction. Refraction comes from a Latin word which means to break up.1 If you have ever had an eye exam for corrective lenses, your ophthalmologist or optometrist used refraction to determine your prescription.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
51
Newton used his prism to create rainbows (Figure 1). Keats was appalled. Newton ruined rainbows by explaining them.2 Glasses, contact lenses, fiber-optic cables, lasers, big-screen TV and digital cameras work because Newton stuck to his intellectual guns. We are glad he did. The process of refracting white light into a visible spectrum of colors is a form of vector analysis. We are not being overly lyrical when we say that Ronald Fisher’s vector analysis “refracts” data. Refraction makes profit signals visible. This is essentially what you did to earn your Five-Minute PhD.
Figure 1 A diamond sparkles with colorful data vectors refracted from ordinary light.
For poets, this perspective is unwelcome. They are not alone in this feeling. Again and again, we hear Keats’ critique of Newton echoed in the protests of old-school managers who reject profit signals as well as the process of disciplined observation, experimentation and analysis. “Scientific” Management Managing any business is a challenge. Complexity arises from materials, work methods, machinery, products, communication systems, customer requirements, social interactions, cultures and languages. The first step in solving ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
52
Standards of Evidence complex business problems is to frame them in terms of a manageable number of key variables. Bottom-line profitability is the ultimate objective, but other metrics must also be considered. Sales, earnings per share, cost and time to develop and market new products, operating costs, inventory turnover, capital investments and days in accounts receivable are just a few. Profit signals from one or more of these variables often demand timely, reasoned responses. Frederick W. Taylor mesmerized the business community of his day with the 1911 publication of The Principles of Scientific Management. Taylor aimed to explain how any business problem could be solved “scientifically.” As an engineer for a steel company, Taylor had conducted a 26-year sequence of “experiments” to determine the best way of performing each operation. He studied 12 factors, encompassing materials, tools and work sequence. He summarized this massive investigation with a series of multifactor predictive equations. This certainly sounds like science. Unfortunately, trying to solve complex business problems with Taylor’s methods is akin to surfing the Internet with a rotary phone. In his 1954 classic How to Lie with Statistics, Darrel Huff characterized Taylor-style science as follows: “If you can’t prove what you want to prove, demonstrate something else and pretend that they are the same thing.”3 Taylor studied his 12 factors one at a time, holding the other 11 constant in each case. 4 This invalidates his multifactor equations. One-factor-at-a-time experiments are so thoroughly discredited, that they have their own acronym, OFAT. It is physically impossible for OFAT experiments to characterize multi-factor processes. OFAT experiments are also notoriously time consuming. This is probably why it took Taylor 26 years to complete his study.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
53
Cost Accounting Variance Analysis Businessmen in the early twentieth century enjoyed comparing themselves to Einstein, Marconi, Edison, the Wright Brothers and other celebrity scientists of the day. G. Charter Harrison, an accountant with Price, Waterhouse and Company in London, chose Taylor as the celebrity “scientist” he wanted to emulate. Harrison published a series of articles in 1918 in support of his assertion that, “The present generally accepted methods of cost accounting are in as retarded a state of development as were those of manufacturing previous to the introduction by Frederick W. Taylor of the idea of scientific management.” A tidal wave of popularity was carrying Taylor’s book to best seller status. Harrison rode this wave. He advanced “scientific” principles for cost accounting. He proposed that “standard costs” be established for various tasks, and that actual costs be analyzed as deviations from the standard costs. This was an advance over previous methods. Harrison went on to describe an assortment of things that could be calculated from such differences, including “productivity ratios.” A 1964 Times Review of Industry article first used the term variance to describe Harrison’s difference between actual and standard costs.5 Perhaps old-school accountants and managers thought “variance” sounded more scientific than “difference.” They had good reason to do so. By 1964 Ronald Fisher’s vector analysis solution to a wide variety of statistical problems were widely known under his general term for them, Analysis of Variance. Analysis of Variance is the international gold standard for quantitative work in virtually every profession. Prior to the invention of Six Sigma in 1986, two notable professions were the only exceptions to this rule: accounting and business management. By 1978, business journalists were using the phrase “variance analysis” to refer to the examination of differences between planned and actual performance. The expression persists in today’s accounting textbooks: “The act of computing and interpreting variances is called variance analysis.”6 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
54
Standards of Evidence Needless to say, the cost accounting variance analysis of 1978 bore no relation to the Analysis of Variance invented by Fisher some 58 years earlier. The elements of a standard cost accounting variance analysis are shown in Table 1. �
�
�
�
�
� �
Table 1 Cost-accounting variance report formats vary. The key element is a column labeled “Variance Ratio.” It is the signed difference between an actual value and a standard, budgeted or forecast value, expressed as a percentage.7, 8, 9
�
�
�
�
� �
�
�
�
�
� �
�
�
�
�
�
�
�
� �
� �
� �
� �
� �
� �
� �
� �
� �
� �
� � �
� � �
� � �
� �
� � �
� � �
� � �
� � �
� � �
� �
� � �
� � �
� � �
� � �
� � �
� � �
� � �
� � �
� � �
�
�
�
�
�
�
�
�
�
� �
It is unfortunate that the word “variance” was redefined in 1964 to mean a difference between actual and standard values. There is nothing inherently wrong with analyzing such differences. In fact, it is a good idea. The problem comes in the type of “analysis” that is done with such differences, and the actions “variance analysis” conclusions can lead to. For example, the manager who is responsible for the $1,000 revenue “variance” in Table 1 will be asked to explain himself or herself. After all, the result is 20% under forecast! The explaining of this unacceptable negative variance occurs at the monthly meeting of the Executive Committee. This monthly ritual creates tremendous pressure to conform. It subverts critical thinking. Managers are forced to develop story-telling skills. A plausible explanation is produced. The manager vows not to let this bad thing happen again. After a month or two or three, the offending “variance” happens again. A plausible explanation is produced. The manager swears never to let this new bad thing happen again. And so on. The highest-paid employees in the company waste hours, days and even weeks every month, grilling each other over G. Charter Harrison’s 1918 productivity ratios. Objective ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
55
standards of evidence are nowhere to be found in their discussions. The Executive Committee may as well try to produce rainbows without a prism. The monthly cross-examination over variance analysis rather than Analysis of Variance (ANOVA) is an indefensible waste of time and money. It is every company’s greatest obstacle to evidence-based decisions and Six Sigma breakthroughs. Accounting versus Science Today’s Generally Accepted Accounting Principles (GAAP) are loose guidelines. Little has changed from those submitted to the United States government by a committee of Certified Public Accountants in 1932. The authors of the book Relevance Lost: The Rise and Fall of Management Accounting, winner of the 1989 American Accounting Association’s award for Notable Contribution to Management Accounting Literature, judged the accounting profession to be functioning 70 years behind the times.10 In 2003, that makes it 84 years behind the times! How could this happen? Perhaps it is a function of what the customers want. The assistant dean of a leading business graduate school recently told us that her university continued to teach a core curriculum subject—cost-accounting variance analysis—that she personally knew to be false. She reasoned, “Businesses in this region hire graduates who know how to use costaccounting variance analysis.” Another, more revealing explanation runs deeper. Empirical laws of science are forced to evolve. Over time, an inexorable process of disciplined observation, experimentation and analysis leads to improvements. Occasionally, a body of evidence becomes so compelling it becomes a new Generalization or Law. New laws force old ones to be revised or scrapped. By contrast, in the words of a Harvard Graduate School of Business Administration textbook, “Accounting principles are man-made. Unlike principles of physics, chemistry, and ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
56
Standards of Evidence the other natural sciences, accounting principles were not deduced from basic axioms, nor can they be verified by observation and experimentation.”11 In other words, cost accounting principles cannot be tested for validity. They have no objective standards of validity. There is no process of disciplined observation, experimentation and analysis to force improvements. Until one-dimensional, cost-accounting arithmetic is upgraded to vector analysis applied to a data matrix, GAAP will remain wide enough to drive a Six Sigma tractor-trailer rig through. Delusions and Bamboozles In his 1841 classic Memoirs of Extraordinarily Popular Delusions,12 Charles Mackay described massive losses related to business practices just like today’s cost-accounting variance analysis. He could well have been writing about 20th and 21st Century popular culture when he penned the chapter “The Love of the Marvelous and Disbelief of the True.” “In reading the history of nations, we find that, like individuals, they have their whims and their peculiarities; their seasons of excitement and recklessness, when they care not what they do. We find that whole communities suddenly fix their minds upon one object, and go mad in its pursuit; that millions of people become simultaneously impressed with one delusion, and run after it, till their attention is caught by some new folly more captivating than the first.” 13 This passage is eerily familiar to those of us who have watched businesses become captivated by one management fad after another: “Excellence”, “Re-Engineering”, “Zero-Based Budgeting”, “Zero Defects”, “Total Quality Management”, “Activity-based Cost Accounting”, “Management by Objective” and “Balanced Scorecards” are a few of the greatest hits. None of these well-intentioned initiatives were, or are, particularly bad in and of themselves. They simply lack the firm foundation and objective standards of evidence sound theory provides.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
57
Occasionally, a superstition, fad or fallacy—astrology, homeopathy, phrenology, cost-accounting variance analysis, take your pick—manages to survive for a few years or decades. Eventually, a kind of critical mass of delusion is established. The capacity for critical thinking erodes.14 Carl Sagan explained it this way: “One of the saddest lessons in history is this: If we have been bamboozled long enough, we tend to reject any evidence of the bamboozle. We’re no longer interested in finding out the truth. The bamboozle has captured us. It’s simply too painful to acknowledge, even to ourselves that we have been taken. Once you give a charlatan power over you, you almost never get it back. So the old bamboozles tend to persist as new ones rise.” 15 All business leaders—plant managers, doctors, billionaire CEOs—face this dilemma when they try to bring evidencebased decisions into their organizations. Generally Accepted Accounting Principles place no premium on truth or even facts. They prize only internal consistency. Once you have bamboozled the public, the stockholders, or the employees of your company, the path of least resistance is to keep on bamboozling. Our proposal? Replace cost-accounting variance analysis with a reliable, transparent, proven method of analysis based on objective, quantitative standards of evidence. That proven method is profit signal vector analysis applied to a data matrix.
Vector Analysis 101 John Keats, a poet of the Romantic Period, lived in a world of pure expression. If he were an accountant today, he would demand a certain freedom of expression. He would present his data in free verse or any other format he desired. He would analyze data however he wanted, according to his prevailing emotions. He would assign to his analysis whatever weight of evidence felt right. Amazingly, he would be granted all these freedoms today without so much as a raised eyebrow. According to a 1990s Professor Emeritus at the Harvard Graduate School of ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
58
Standards of Evidence Business Administration, “There are no prescribed criteria [for variance analysis] beyond the general rule that any technique should provide information worth more than the costs involved in developing it.”16 The lack of any prescribed criteria for financial analysis explains why spreadsheets are so popular. (See Table 2.) Just like Keats, most people like being free to arrange their data in any way that suits their fancy. What better way to spice up the workday? The situation is quite different for a Law or a Generalization like Fisher’s Analysis of Variance. Unlike Taylor and Harrison, Fisher actually was a scientist. In sharp contrast with the artificial guidelines of cost accounting, Fisher’s work was grounded in rigorous mathematics and physical reality: Cartesian coordinates, right triangles, Pythagoras’ Theorem, plane and solid geometry and trigonometry, multi-variable calculus and vector analysis. �
Table 2 A spreadsheet consists of rows and columns. The cells can contain text, numbers, graphics, symbols or formulas. There are no rules governing the interpretation of rows and columns. There are no laws for arranging or analyzing data.
� �� � � �� � � �� �
�
� �� � � ������
� ��
� ��
� ������
� ������
� ������
� ������
� ������
� ������
� ������
� ������
Ironically, Fisher developed the Analysis of Variance in a farm near London right around the same time Taylor and Harrison were promoting their “scientific” management and costaccounting principles. Managers would do well to follow his lead; with statistical software they can do so immediately at virtually no cost. Evidence-based decision companies repeatedly demonstrate why this is a profitable choice. Fisher coined the term Variance in 1918. This was 46 years before its debut in the Times Review of Industry.17 It is an understatement to say the two definitions are significantly different. They are as different as Generalization and ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
59
generalization. To quote Mark Twain, “It is the difference between lightning and a lightning bug.” Instead of a difference between an actual value and a standard value, Fisher’s Variance measures the degree of variability of a set of values around their average. It is based on the length of the variation vector. Fisher called his method “Analysis of Variance” because its purpose is to break up the variation vector into profit signal and noise components. Fisher’s work defines today’s international standard for analyzing components of variation. In 1919, at age 29, Fisher was hired to “examine data and elicit further information” from his employer’s database. According to his employer, “It took me a very short time to realize that he was more than a man of great ability, he was in fact a genius who must be retained.”18 Fisher’s job was to re-evaluate a business report identical to the ones managers use for decisions today.19 There were measurements recorded in rows and columns. Fisher’s boss subtracted average annual production numbers from each other to “determine” which years were most productive. The boss wanted to increase the annual yield in bushels of wheat. Like most people, he wanted to make more money while working shorter hours and using fewer resources. Fisher knew exactly how to help his boss achieve these objectives: apply a vector analysis to a data matrix. Like a spreadsheet, a data matrix consists of rows and columns. There the similarity ends. The rows of a data matrix represent records—the individual objects or events on which we have data. The number of rows is called the sample size. The columns represent fields—the variables measured or inspected for each object. Table 3
shows a simple data matrix. There are two variables measured on two objects. Each column in a data matrix contains the measurements which are the data vector for the variable associated with the column. Each object is represented by a particular coordinate or position in the vector.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
60
Standards of Evidence
�
Table 3 Like a spreadsheet, a data matrix consists of rows and columns. The rows of a data matrix represent records—the individual objects or events we have data on. The columns represent fields—the variables for which we have data. Each stack of numbers in a data matrix column is a vector.
� �������������� � � �������������� �
� ��������� ����������� ����� � � ��
� ��������� ����������� ����� � � ���
� ��
� ���
� For example, the data vectors in Table 3 are (3, 4) for Variable 1 and (5, 2) for Variable 2. These vectors are plotted in Figure 2 . The two coordinate axes correspond to the two objects. Fisher’s innovation was to think of the data matrix in a geometric framework. In this example the vectors are two-dimensional because there are two objects in the data matrix. In general, vectors are n-dimensional, where n is the sample size. We are back into hyperspace. Like the inside of a black hole, hyperspace will remain forever beyond our three-dimensional vision. Nevertheless, it is there. It is real. Evidence-based decision companies use hyperspace to make more money in less time while using fewer resources.
Figure 2 The two columns of Table 3, plotted as vectors.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence Figure 3
61
illustrates the first basic rule of vector analysis:
The shortest distance between a point and a line is along a path perpendicular to the line. Figure 3 The shortest distance between a point and a line is along a path perpendicular to the line.
This is no arbitrary accounting rule; it is a property of the physical universe. It is a law, a mathematical/scientific Generalization. The first step in a vector analysis is to find the constant vector closest to the data vector. Examples of two-dimensional constant vectors are (1, 1), (2, 2) and (0.5, 0.5). The dotted line in Figure 4 locates the set of all possible twodimensional constant vectors. Figure 4’s center vector masks a long segment of this dotted line, moving from the lower left point of origin to the upper right. Only a portion of the dotted line is visible at the upper right hand portion of the illustration. For our data vectors, D1 and D2, the closest point on this line is (3.5, 3.5). It is not a coincidence that 3.5 is the average of 3 and 4. It is not a coincidence that 3.5 is the average of 5 and 2. The closest constant vector is always the vector of averages. It does seem coincidental that (3, 4) and (5, 2) have the same average, but we did this on purpose. So, how do these vectors differ? Well, (3, 4) is closer to the vector of averages than (5, 2) is (Figure 4). A data vector close to its vector of averages has less variability than a data vector far from its vector of averages. This means Variable 1 has less variability ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
62
Standards of Evidence than Variable 2. You can also tell this just by looking at the numbers in Table 3. This “eyeball” analysis is just for illustration; it is not recommended for your real data sets.
Figure 4 The dotted line is the set of all constant vectors. The constant vector closest to any data vector is the vector of averages, shown here in bold.
Figure 5 identifies the variation vectors, V1 and V2 , for Variables 1 and 2. The length of the variation vector is directly related to the degree of variability in the data vector.
Figure 5 A is the vector of averages for both Variables 1 and 2. V1 and V2 are the corresponding variation vectors.
How do we calculate the length of a vector? For this we need the second basic rule of vector analysis: The New ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
63
Management Equation (a.k.a. the Pythagorean Theorem). The square of the length of the long side of a right triangle is equal to the sum of the squares of the lengths of the other two sides. (c2 = a2 + b2) Once again, this is no arbitrary accounting rule; it is a property of the physical universe. The New Management Equation is so well known in professional financial and investment analysis circles that a bi-monthly newspaper, Financial Engineering News was founded in 1997 to disseminate case studies. In Figure 6 we use the New Management Equation to calculate the lengths of data vectors D1 and D2.
Figure 6 The length of a vector is the square root of the sum of the squares of its coordinates.
Now we can figure out the lengths of the variation vectors in Figure 5. Only the alphabetic notation differs from the New Management Equation. Using the letters in Figure 5, the New Management Equation for Variable 1 is: (D1)2 = A2 + (V1)2 We can see in Figure 6 that (D1)2 = 25. Also, the squared length of the data average vector A is:
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
64
Standards of Evidence
A2 = 3.52 + 3.52 = 24.5 We can now plug these two numbers, 25 and 24.5, into the New Management Equation for data vector D1: 25 = 24.5 + (V1)2 25 - 24.5 = 0.5 = (V1)2 V1 = square root of 0.5 = 0.71. This final number, 0.71, the length of the variation vector for Variable 1, is called the sample standard deviation for Variable 1. A sample standard deviation is symbolized in technical writing by the letter s. The Greek letter sigma (σ) refers to the standard deviation of a population. This is where Six Sigma gets its name. In Six Sigma practice, the sample standard deviation, s, is often casually referred to as “sigma” or σ. This substitution is a grievous breach of statistical theory, but everyone who uses statistics does it. The New Management Equation for Variable 2 works the same way: (D2)2 = A2 + (V2)2 We know from Figure 6 that D2 = 5.39. Please do keep your eyes on the right triangles in the illustrations. We already know that A is the square root of 24.5, which equals 4.95. We can now plug these into the New Management Equation: 5.392 = 4.952 + (V2)2 29.05 = 24.5 + (V2)2 4.55 = (V2)2 V2 = square root of 4.55 = 2.13. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
65
The sample standard deviation for Variable 2 is 3 times larger than that for Variable 1! Variable 2 is 3 times more variable than Variable 1. Six Sigma values smaller variation because outcomes are more predictable. Predictions are more accurate. There is less waste and rework. Everything just works better when the profit signals are large/strong and the noise is small/weak.
Degrees of Freedom Don’t panic. Think of what follows as a mandatory Federal Communications Commission announcement on your National Public Radio station. It has to be here to ensure we are not breaking any Laws of the Universe. You can skip this section if you want, or you can stay tuned. In either case, software takes care of all this stuff. This is just background information. Sometimes an analyst might want or need to know the actual coordinates of a variation vector. (Whenever our airplane takes off or lands, we certainly hope our pilot and co-pilot have this information at their fingertips.) We get the coordinates of the variation vector by subtracting the data average vector from the data vector. The clearest way to explain the subtraction of vectors is to give the vectors a vertical orientation, the way they appear in a data matrix. The coordinates of the variation vector for Variable 1 are given by:
For Variable 2, they are given by
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
66
Standards of Evidence So far, so good. Now, here is an important Law of the Universe: The coordinates of a variation vector always add up to zero. Because of this, the second coordinate in a two-dimensional variation vector is always equal to the negative of the first coordinate. This means that a two-dimensional variation vector is completely determined by its first coordinate. We express this by saying that a two-dimensional variation vector has one degree of freedom. Suppose now we have a three-dimension data vector, for example (3, 4, 5) or (5, 2, 5). The vector of averages for both of these is (4, 4, 4). Once again using the vertical data-matrix orientation, the first variation vector is:
and the second is:
In a three-dimensional variation vector, the third coordinate is always equal to minus the sum of the first two coordinates. This means that a three-dimensional variation is completely determined by its first two coordinates. We express this by saying that a three-dimensional variation vector has two degrees of freedom. Now let n stand for the number of objects in your data set. This is the same as the number of rows in your data matrix. It is your sample size. All the vectors are now n-dimensional. Yes. We are back into genuine hyperspace again. The more often you go there, the less scary it becomes. Visits become more profitable. They become fun. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
67
The last coordinate of an n-dimensional variation vector is always equal to minus the sum of the first n – 1 coordinates. This means that an n-dimensional variation vector is completely determined by its first n-1 coordinates. We express this by saying that an n-dimensional variation vector has n -1 degrees of freedom. The upshot of all this is this: the standard deviation is exactly equal to the length of the variation vector only when n = 2. When n is greater than 2, as it usually is, we have to divide the length of the variation vector by the square root of its degrees of freedom. Don’t blame us—it’s a Law of the Universe. We will come back to this later in the chapter, and also in Chapters 5 and 6. We now return to our regularly scheduled program of writing with an improved degree of simplicity.
Bar Chart Bamboozles Bar charts and pie charts symbolize old-school management thinking as no other icon can. They are the “Gee Whiz” graphs in Huff ’s Lying with Statistics. They present data in superficial ways. They are easy to use. Consequently, they frequently are used to misrepresent data. The typical bar chart presents totals or averages with no consideration of variability. There are no deviations from the average. There is no Chance variation. In other words, there is no Noise. This violates a Law of the Universe. There is always Noise, which is statistical variation. Variation is a physical property of objects and measurements. A vector analysis forces us to consider both average and standard deviation. At best, we might say bar charts have a 50/50 chance of giving correct information because they consider only one of two aspects. At worst, they encourage managers to use Frederick Taylor’s thinking, “This bar is bigger than that bar is and I know the reason why because I am a scientific manager and I say so.” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
68
Standards of Evidence
As an example, consider the monthly revenue data in Table 4. This is a snapshot of data entered into a spreadsheet.
Table 4 Monthly revenue for four years.
The annual totals are plotted as a bar chart in Figure 7. The upward trend looks very encouraging. The Marketing Manager would certainly want to take credit for this.
Figure 7 Excel’s popular bar chart/ trend line combination is like Romantic poetry. This poetic license gives everyone the freedom to take credit for good results, whether or not they are true.
All Figure 7 really does is graphically frame the differences between the annual totals. Because Laws of the Universe are ignored, there is no way to tell whether the “trend” is a profit signal or noise. This is a bit like trying to ignore gravity.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
69
Corporate cultures that use cost-accounting variance analysis as the standard decision-making tool often use bar charts and trend lines to present “results” like Figure 7 based on data like that in Table 4. The credibility of the results portrayed by the chart, and the explanation for them, comes from the status of the person telling the story rather than the evidence in the data. There is no cross-examination of the reported results because it is considered poor form, not to mention career limiting, to question the President, Managing Director, Chief Financial Officer, or a company founder who created spreadsheet software. In corporate cultures that base decisions on objective standards of evidence, the analysis method itself is held to high standards. Quite simply, it must follow the Laws of the Universe. It must follow the rules of vector analysis. Evidence is admissible if and only if the analysis method takes all aspects of the data into account. The analysis must have transparency. All elements must be available for review, including the raw data. Anyone can ask any question because all the data are in view. The vector analyses illustrated below represent the international standard. There are several things wrong with the “analysis” in Figure 7. For one thing, it uses only the annual totals instead of the original monthly data. For purposes of illustration, we will present two vector analyses that use only the four annual totals. The first of these is given in Table 5. Table 5
lays out the basic vector calculations for the sample standard deviation, s. In this case s = 0.14. This is a vector analysis in four-dimensional hyperspace, because there are four data points.
The data average vector has one degree of freedom because one number, the average of the four data points, determines it. This leaves three degrees of freedom for the variation vector. The lengths of the vectors are related by the New Management Equation. [C2 does equal A2 plus B2, 140.36 = 140.30 + 0.06.]
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
70
Standards of Evidence We used Microsoft Excel to create the visual presentation in Table 5. The squared lengths of the vectors were calculated by using the cell function SUMSQ. This function name is short for “sum of squares”. This is appropriate because the squared length of a vector is the sum of the squares of the coordinates. The syntax for the Excel calculation is: = SUMSQ(cell range)
Table 5 Basic vector analysis of the four annual totals (millions of dollars).
�������� ������
������������ ������
��������� ������
����
����
�����
�
����
�
����
����
����
�����
����
����
����
������������������ ���������������
����
�
�
����
�
����
������
�
������
�
����
���������
����������������� � ������������������� ��
����
�������������������
����������������������������
����
Figure 8
shows the Normal distribution curve corresponding to a mean of $5.92 million and a sample standard deviation of $0.14 million. The dots just above the horizontal axis represent the four annual totals. Each vertical dotted line represents one standard deviation.
Figure 8 The four annual totals from Table 4 and the corresponding Normal distribution curve.
0.14
-3s
5.50
-2s
-1s
0
5.92
+1s
+2s
+3s 6.34
All four data points lie within two standard deviations of the mean. We must conclude that the deviations from the mean ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
71
value are a result of natural, or Chance, variation. There is certainly no evidence of significant differences among these totals. Our second vector analysis addresses directly the validity of the bar graph trend line in Figure 7. The null hypothesis for this analysis is the following statement: There is no significant trend in the annual totals. This is not a foregone conclusion. It is a special kind of hypothesis. It is used in applied research all over the world. The idea is to see whether or not the evidence in the data is strong enough to discredit the null hypothesis. Then, and only then, can we say there is a significant trend in the annual totals. The visual presentation of the analysis is shown in Table 6. The variation vector is broken up into the sum of profit signal and noise vectors. These three vectors are related by the New Management Equation (a.k.a. Pythagorean Theorem). The squared lengths of the vectors are also called “sums of squares.” The profit signal vector is equal to the best-fit line in Figure 7 minus the data average, 5.923 in this case. The coordinates of the profit signal always add up to zero, so it is completely determined by the slope of the best-fit line. As a result, the profit signal vector has one degree of freedom. That leaves two degrees of freedom for the noise vector. To get the profit signal and noise variances, we divide the sums of squares by the degrees of freedom. This is a Law of Universe. Without this adjustment, the variances would be biased. When we divide the profit signal variance by the noise variance we get a signal-to-noise ratio that measures the strength of evidence against the null hypothesis. It is called the F ratio, or F statistic, because Ronald Fisher invented it. Larger values of F imply stronger evidence against the null hypothesis. In this case F = 2.843.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
72
Standards of Evidence
Table 6 Ilustration of the vector analysis for a linear trend in the four annual totals. The squared lengths of the vectors are also called “sum of squares.” This is a reference to the New Management Equation, which involves a sum of squared numbers.
This number doesn’t seem very large. But there is no standard scale of comparison for the F ratio. Instead, we interpret it relative to a statistical distribution representing chance variation. This distribution depends on the degrees of freedom for the profit signal and noise vectors. The p-value of 0.234 in Table 6 is the probability of getting an F ratio as large as 2.843 by chance alone. If the p-value is small enough, we reject the null hypothesis. By established international standards, the evidence against the null hypothesis is ‘clear and convincing’ if the p-value is less than 0.05. If the p-value is greater than 0.05 but less than 0.15, there is a ‘preponderance of evidence’ against the null hypothesis. The p-value in Table 6 does not meet even this lowest standard of evidence. There is no significant trend. Table 7
shows the monthly revenue numbers in data matrix format. This data set is too large to use as a tutorial. We present some smaller examples in Chapter 5.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
73
Table 7 The monthly revenue numbers in data matrix format (thousands of dollars).
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
74
Standards of Evidence Meanwhile, a great deal can be learned simply by plotting the data in time sequence. This is done in Figure 9. It doesn’t take a Statistician to see that there is no trend here, just random variation. The only features of note are the three low points at the beginning of the series. In turns out these were the last three months before a change in the accounting procedures. They should have been omitted from the analysis.
��������������������� �������
Figure 9 The monthly revenue numbers plotted in time sequence. �������
������� �������
������� ������� �������
��
��
��
��
��
��
��
��
��
�
�
�
������� ������
The Game is Afoot Another example of a full vector analysis is the shoe-sole wear rate workshop in the classic 1978 text Statistics for Experimenters: An Introduction to Design, Data Analysis and Model Building by George Box, William Hunter and J. Stuart Hunter. This example uses the small data set presented in their book, with an invented story line based on our consulting experiences.20 It achieves the following objectives: 1. You can quickly see the differences between a typical spreadsheet analysis and vector analysis applied to a data matrix. 2. The manufacturing design, cost and margin analogies are appropriate. A design team is arguing over the wear rates of shoe-sole materials A and B. Material A, the current specification, is more costly than Material B. The manager wants to go with Material B because it is cheaper, and his spreadsheet analysis shows there will be no significant loss of durability. Engineers ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
75
are concerned that Material B is not sufficiently durable. Data has been collected and arrayed in a spreadsheet. (See Figure 10.)
���
Figure 10 Wear rate data as arrayed in Excel.
����������� ���������
����
��� ���� ����� ���� ��� ����� ������ ����� ������� ����� ���� ���� ��� ���� ����� ����� ������ ����� ���� ���� ��������������������� ����������������������� ����������
����� ���� ����� ����� ����� ���� ���� ����� ���� ����� ������ �����
���� ����� ����� ���� ����� ���� ����� ����� ����� ���� �����
������������������ ��������������������������� ����������
����������� ��������� ����� ���� ����� ����� ����� ���� ���� ����� ���� ����� ������ ����� ����
�����
Ten boys were enlisted for the test. Each boy wore one shoe made from Material A and one from Material B. Coin tosses were used to randomly assign Material A to the left or right foot for each boy. The average wear rate for Material B comes out 0.41 units higher than for Material A, an increase of 3.86%. Given the price difference between the two materials, the manager concludes that the difference in durability is irrelevant. Furthermore, as shown by the bar chart in Figure 11, there were a number of cases where Material A actually wore out faster than Material B! The manager is elated. By using
Figure 11 Wear rate data as analyzed by a spreadsheet bar graph.
��������������������
����� ����� ���� ����������
���� �
�
�
�
�
����������
����
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
76
Standards of Evidence material B instead of A, the shoe manufacturer can increase profit margins and maintain product durability. This change will be worth millions to the bottom line. After a long and difficult team meeting, consensus is reached. The company will replace material A with the less costly, equally durable material B. As the meeting is wrapping up, a Six Sigma Black Belt in training asks if she can analyze the data herself using a vector analysis applied to a data matrix. It is getting late. People have places to go, things to do. Nevertheless, to maintain good relationships, they give her five minutes. She imports their Excel spreadsheet into her statistical package. For present purposes, we recreate her vector analysis data in Excel. This is shown in Table 8. (We timed both methods. The Excel reconstruction literally took 10 times longer than doing a correct vector analysis in the statistical package.)
Table 8 Vector analysis of the wearrate data.
�
The Black Belt trainee starts her extemporaneous presentation by stating the null hypothesis for the analysis: “There is no difference between the average wear rates of the two materials.” The trainee explains that this is a hypothesis, a straw man to be pulled apart by evidence, rather than a foregone conclusion. The idea is to see whether or not the ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
77
evidence in the data is strong enough to discredit the null hypothesis. She goes on to explain that we should be looking at the differences between A and B for each boy—that was the whole point of having each boy wear one shoe of each kind. If the null hypothesis were true, the differences should be symmetrically distributed around zero. Also, the average difference should be close to zero. With three clicks of her mouse, the trainee produces a frequency histogram of the differences (see Figure 12). Pointing at the graph, she says, “As you can see, all but two of the differences are positive. This casts doubt on the null hypothesis—the wear rates for Material B are consistently higher than those for Material A. “But let’s not jump to conclusions. We need to complete the
������������
�������������
Figure 12 Frequency histogram of differences in wear rate (B minus A).
�����
�
���
��
���
�
����
vector analysis to establish the strength of this evidence. As you can see, the vector analysis (Table 8) breaks the vector of differences into the sum of the data average vector and the noise vector. For analyzing matched pairs like we have here, the data average and the profit signal vector are one and the same. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
78
Standards of Evidence “As you can see, the lengths of the vector of differences, the profit signal vector and noise vector are related by The New Management Equation. The vectors are 10dimensional because there are 10 differences. We are way into hyperspace. The profit signal vector is determined by one number, the average difference of 0.41, so it has one degree of freedom. That leaves nine degrees of freedom for the noise vector.” Her presentation was interrupted by one of her friends. “Let’s take a pause for just a moment here to do a little yoga stretching while our minds are bending.” After some uncomfortable laughter, the Black Belt’s Six Sigma analysis continued. “OK. We are back on task. We have to adjust the New Management Equation (a.k.a. sums of squares a.k.a. squared lengths of vectors) by dividing by the degrees of freedom. “This gives us Variances that measure the strength of the profit signal and noise vectors. When we divide the profit signal variance by the noise variance we get a signal-tonoise ratio that measures the strength of evidence against the null hypothesis. It is called the F ratio because a guy named Fisher a long time ago invented it. As you can see, the F ratio in this case is 11.215. “The F ratio can’t be interpreted on its own. We have to compare it to a distribution to see how likely it is that a value as large as 11.215 could have occurred by chance alone. This probability is called the p-value. If the p-value is small enough, we have to reject the null hypothesis. “By established international standards, the evidence against the null hypothesis is ‘clear and convincing’ if the p-value is less than 0.05, and it is ‘beyond a reasonable doubt’ if the p-value is less that 0.01 (Table 9). As you can see,” she said, pointing at her computer screen, “the p-value in this case is 0.0085. This means there is a significant difference between A and B, beyond a reasonable doubt.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
79
Table 9 The Black Belt showed the table of evidence to the team.
One engineer says, “That makes a lot of sense. Even though the difference was less than 4%, we felt that a difference of 0.41 units could cause problems. We were afraid yield losses would exceed the savings on material costs.” A potential disaster is narrowly averted by using an evidencebased decision in the nick of time. Critical-to-quality characteristics and financial margins are protected. The company’s reputation for quality is preserved. Just another day in the life of a Six Sigma company. The next Black Belt, Green Belt, Yellow Belt and Champion courses are filled to capacity. The waiting lists for the following sessions are long. The company takes the next step forward by implementing Six Sigma across all projects and functional responsibilities in the corporate matrix. Their firstwave Black Belts are now in Master Black Belt training using their own case studies. Spreadsheet versus Data Matrix Spreadsheet arithmetic is today’s cost-accounting variance analysis computing engine. While teaching the real Analysis of Variance we often hear the comment, “So what’s the big deal with a data matrix? You can do all that in a spreadsheet.” This is true. We know because we have done it. Some of that work has been presented in this chapter. There is more to come in Chapters 5 and 6. It is also true that you could eventually compute the orbital trajectories of all the planets in our solar system with an abacus.21 Unless you and your loved ones have nothing better to do with the rest of your lives, our question is this, “Why would anyone want to?” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
80
Standards of Evidence The spreadsheet is a marvelous invention. It automates arithmetic. You can write formulas. Like Keats, you can put your data wherever you want it and analyze it however you want. If you add in enough add-ins, you can actually do some statistics, even Analysis of Variance. Adding in the add-ins is a clumsy way of trying to reinvent the machinery of a vector analysis that already exists in modern statistical software. These programs give you access to this machinery with a mouse click. The greater liability in trying to do everything with a spreadsheet stems from the very freedom that makes spreadsheets so popular. Spreadsheet applications are unruly and Lawless. The Laws of the Universe do not apply to them. Statistical packages, on the other hand, follow the Law. They require the correct data matrix structure—each row an object of interest, each column a vector of data on the objects. Data vectors are the principle components of vector analysis. (Hence the name.) Vector analysis provides the transparency required to satisfy international accounting standards and scientific standards of evidence. For example, statistical packages automatically create the variation, profit signal and noise vectors shown in Tables 5, 6 and 8. One can create this table in a spreadsheet, although it is tedious. We did in fact make Tables 5, 6 and 8 in a spreadsheet, but nothing forces other users to do so. The undemanding nature of spreadsheets lures unsuspecting users into sins of omission. There is no requirement for vector analysis, no requirement for transparency. Other spreadsheet characteristics are simply inconvenient or annoying. For example, many spreadsheet functions treat blank cells as zeroes. This works fine for adding and subtracting. In reality, a blank cell indicates a missing value in a data vector. A missing value changes the degrees of freedom and dimension of the vector. The vector analysis can handle this, although it does affect the results. By contrast, the cavalier insertion of zeroes for missing values wreaks havoc on vector analysis, giving incorrect results. These and related comments are summarized in Table 10.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
81
Table 10 Comparing and contrasting a spreadsheet and a data matrix. Inductive and deductive reasoning are built into data matrix software. No such discipline exists in a spreadsheet.
P-values, Profit Signals, Confidence Levels and Standards of Evidence A null hypothesis always consists of a negative assertion. The phrasing of a null hypothesis is not a law of the universe, but it is an odd standard. Here are some examples: • • • • •
There is no difference between these two ways of doing things. There are no differences among these three or more ways of doing things. There is no relationship between these two variables. There are no relationships among these three or more variables. There are no relationships between these two groups of variables.
The null hypothesis often plays the role of a “straw man” in inductive reasoning. According to the on-line folklore database Wikipedia, the straw man concept began as a rodeo safety tactic.22 A straw man would distract bulls. It could be torn apart with no harm done. We can tear apart the straw man, the null hypothesis, if it is something we would like to disprove based on the data. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
82
Standards of Evidence The F ratio, or F statistic, is a signal-to-noise ratio that measures the strength of evidence in the data against the null hypothesis. As the F ratio increases, the strength of evidence against the null hypothesis increases. We evaluate an F ratio by comparing it to a statistical distribution to see how likely it is that a value that large could have occurred by chance alone. The distribution to which the F ratio is compared depends on the degrees of freedom for the profit-signal and noise vectors. As a result, there is no standard scale of comparison for the F ratio. We get around this by working with a probability computed from the F value. This probability, called the p-value, is the probability of getting an F ratio as large as the value we got by chance alone. If the p-value is small enough, we reject the null hypothesis. In Microsoft Excel, the cell formula syntax for calculating the p-value is this: = FDIST(value of F ratio, degrees of freedom for the profit signal vector, degrees of freedom for the noise vector) For example, the formula to produce the p-value 0.0085 in Table 8 is as follows: = FDIST(11.215, 1, 9) 11.215 is the value of the F ratio, 1 is the number of degrees of freedom for the profit signal vector, and 9 is the number of degrees of freedom for the noise vector. Enter this formula into your Excel spreadsheet and you will get the correct answer: 0.0085. The formula to produce the p-value 0.234 in Table 6 is as follows: = FDIST(2.843, 1, 2) 2.843 is the value of the F ratio, 1 is the number of degrees of freedom for the profit signal vector, and 2 is the number of degrees of freedom for the noise vector. Enter this formula into your Excel spreadsheet and you will get the correct answer: 0.234.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
83
We do not like writing spreadsheet formulas. We do like the fact that statistical software does it for us automatically. As the F ratio increases, the p-value decreases. As the p-value decreases, the strength of evidence against the null hypothesis increases. This tends to confuse people. It is easier to think in terms of confidence levels (Table 11). The confidence level is one minus the p-value, usually expressed as a percentage. As the confidence level increases, the strength of evidence against the null hypothesis increases.
Table 11 Standards of evidence in a nutshell. A p-value less than 0.05 yields a confidence level greater than 95%. A p-value less than 0.01 yields a confidence level greater than 99%.
Closing Arguments Themis is the Blind Lady of Justice in Greek mythology. Themis: “As an oracle, I used to advise Zeus when he made
decisions. I did my job so well I became the goddess of divine justice. You can see from some of my portraits that I used to carry a sword in one hand and a set of scales in the other. The blindfold I wore was more than a fashion statement. It meant I would be fair and equitable in my judgments. My whole existence hinges on objective standards of evidence.”23
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
84
Standards of Evidence Endnotes American Heritage Dictionary of the English Language, Third Edition. Boston. Houghton Mifflin Company. 1992. 1
Dawkins, Richard. Unweaving the Rainbow, Science Delusion and the Appetite for Wonder. Boston, Houghton Mifflin Company, 1998. 2
Huff, Darrell and Geis, Irving. How to Lie with Statistics. New York, W.W. Norton and Company. 1954. 3
Taylor, Frederick Winslow. Scientific Management, Mineola: Dover Press, 1998. pages 55-59. The original 1911 version was published by Harper and Brothers, New York and London.. 4
5
Oxford English Dictionary, 1989.
Garrison, Ray H. and Noreen, Eric W. Managerial Accounting, 10th Edition. Boston, McGraw-Hill Irwin, 2003. Page 431. 6
Harrison, G. Charter. Cost Accounting to Aid Production – I. Application of Scientific Management Principles. Industrial Management, The Engineering Magazine, Volume LVI, No. 4, October 1918. 7
Harrison, G. Charter. Cost Accounting to Aid Production – I, Standards and Standard Costs, Industrial Management, The Engineering Magazine, Volume LVI, No. 5, November, 1918. 8
Harrison, G. Charter. Cost Accounting to Aid Production – I, The Universal Law System. Industrial Management, The Engineering Magazine, Volume LVI, No. 6, December, 1918. 9
Johnson, H. Thomas, and Kaplan, Robert S. Relevance Lost, The Rise and Fall of Management Accounting. Boston: Harvard Business School Press 1991. Pages 10-12. 10
Anthony, Robert N., and Reece, James S., Accounting: Text and Cases, Eighth Edition. Homewood, Irwin, 1989. Page 15.
11
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Standards of Evidence
85
MacKay, Charles, Memoirs of Extraordinarily Popular Delusions, Copyright 2002 eBookMall version available for $1.75. http://www.ebookmall.com/alpha-authors/m-authors/ Charles-MacKay.htm 12
MacKay, Charles, Memoirs of Extraordinarily Popular Delusions, Copyright 2002 eBookMall version available for $1.75. http://www.ebookmall.com/alpha-authors/m-authors/ Charles-MacKay.htm Page 8. 13
Gardner, Martin. Fads and Fallacies in the Name of Science. New York, Dover Press, 1957. Page 106.
14
Sagan, Carl. The Demon Haunted World, Science as a Candle in the Dark. New York, Ballantine Books, 1996. Page 241.
15
Anthony, Robert N., and Reece, James S., Accounting: Text and Cases, Eighth Edition. Homewood, Irwin, 1989. Page 941.
16
17
Oxford English Dictionary, 1989.
Box, Joan Fisher. R.A. Fisher: The Life of a Scientist. New York: John Wiley and Sons, 1978. Page 97. 18
Box, Joan Fisher. R.A. Fisher: The Life of a Scientist. New York: John Wiley and Sons, 1978. Page 100-102. 19
Box, George E.P., Hunter, William G., and Hunter, J. Stuart. Statistics for Experimenters, An Introduction to Design, Data Analysis, and Model Building. John Wiley & Sons. New York. 1978.
20
Dilson, Jesse. The Abacus, The World’s First Computing System: Where it Comes From, How it Works, and How to Use it to Perform Mathematical Feats, Large and Small. New York, St. Marten’s Press. 1968.
21
22
http://www.wikipedia.org/wiki/Straw_man
23
http://www.commonlaw.com/Justice.html
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
86
©
Standards of Evidence
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 3 Evidence-based Six Sigma
S
ix Sigma (6σ) is a proven, pursuit-of-perfection business initiative that creates breakthroughs in profitability, productivity, and quality. It is a highly structured, project-by-project way to generate bottom line results. It produces significant dollar value through a never-ending series of breakthrough projects. Evidence-based decisions characterize the 18-year, 6σ record of accomplishment. The essential elements of Six Sigma breakthrough projects are vector analyses applied to data matrices. Hundreds of millions of dollars have been placed directly onto the bottom line of companies around the world using this improvement model and its tool set. Though large multi-national corporate results have attracted the most media attention, we have personally seen a 26-employee plastic pressure and vacuum forming company achieve proportionally identical results.
Six Sigma knowledge and know-how have evolved since the notion of perfect 6σ quality was first conceived by Motorola engineer Bill Smith. Motorola’s Chief Executive Officer at the time, Robert Galvin, was the first Six Sigma Champion. He enthusiastically led the entire program. He personally removed bureaucratic obstacles to breakthrough improvements. Six Sigma became an education and training commodity during in the late 1990’s. It gains momentum as it matures. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
88
Evidence-based Six Sigma The catchy three syllable “Six Sigma” moniker is value-added packaging for vector analysis and objective evidence. Six Sigma also conveys substance. Wall Street likes 6σ because it ties customer satisfaction directly to corporate profitability. Customer satisfaction, quality information, speed, and lean organizational structures are Six Sigma cultural values. What is valued gets measured, analyzed and is rewarded. Six Sigma measurements are recorded in data matrices. Since data matrix applications are essential to vector analysis, every true 6σ company has its own corporate software standards. Every Six Sigma champion executive and, if she or he expects to be promoted, every manager in a Six Sigma company has data matrix software loaded on their personal computers. Though many products are available, two currently dominate the market: Minitab and JMP. A Six Sigma analysis is a vector analysis applied to a data matrix. As we graphically detailed in Chapter 2, Six Sigma gets its name from the vector analysis results. This analytic process is sometimes called an Analysis of Variance, or ANOVA. Since the acronym and its equations are traditionally presented in ways that are guaranteed to bore even motivated academics, calling them Six Sigma Tools has worked wonders. Corporate executives embrace them even though only a few know what the phrase and acronym mean. That is a remarkable accomplishment in anyone’s marketing book of records. An ANOVA breaks raw data into six vectors (Figure 1). Two are priceless business intelligence commodities: 1) Profit Signals and 2) Noise. (Historically, Profit Signals have been called “treatment deviations.” That appealed to engineers and statisticians. The mass market of Six Sigma calls for better branding. We answered that call.1) Computing power transforms what was once an almost impossibly difficult series of matrix algebra calculations into a single computer command “Run Model.” Anyone who wants to correctly analyze measurement data can now do so in seconds. When a company combines computing power, the principles of accelerated adult learning and handson improvement projects, breakthroughs routinely lead to quantum leaps in profitability.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
89
Figure 1 A complete analysis is composed of six vectors. Profit Signals quantify what matters most.
Six Sigma (6σ) Basics Here is the bullet list of Six Sigma basics. The jargon side of this business initiative is as real as it is regrettable. Acronyms and algebraic symbols are Six Sigma grammar. We identify these hieroglyphics as a courtesy orientation to newcomers. 1. Top-level executives personally lead the Six Sigma initiative in highly visible ways. Authentic 6σ executives eschew the use of spreadsheet bar graphs and pie charts. Correct, rule driven analyses of financial and productivity data are evident in Six Sigma executive presentations. Executive compensation and promotion are tied to the use of data-driven, evidence-based decisions. The litmus test of leadership is the replication of high dollar value breakthrough projects. If an executive champion does not meet the challenge of these responsibilities, the Six Sigma initiative will fail to produce promised results. 2. Education and skill training in the recognized body of knowledge (BOK) permeate Six Sigma organizations.2 Computing literacy, which means decision makers know how to use a vector analysis applied to a data matrix, is an expected competency for every leader. 3. Exponential rates of improvement are an expected outcome. New ways of getting work done, with fewer ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
90
Evidence-based Six Sigma resources, and in a fraction of the time required by previous methods, take precedence over incremental process improvements. 4. Measurements and Six Sigma metrics are tied to shortterm and long-term financial performance. Executive Six Sigma leaders allocate significant personal time and resources for 6σ projects. In addition to their own investments, they assign the company’s most capable people full-time to lead Six Sigma breakthrough projects. The Executive’s job is to remove bureaucratic roadblocks to improvement so that managers who have an aptitude for implementing productive changes can succeed. The corporate Six Sigma job description hierarchy resembles titles earned in a martial arts dojo. Full-time Six Sigma professionals, called Black Belts, are expected to be able to “kick the heck out of ” any variation that leads to waste or rework.3 In addition to a Karate/Tai Kwan Do/Kung Fu/ Judo level of intellectual aggressiveness, Black Belts must demonstrate leadership and good interpersonal skills. They must be masters of evidence-based decision principles. Ideally, sensei executive champions coach and mentor 9th degree Master Black Belts, who in turn coach, mentor and lead Black Belts. Black Belts then coach and supervise Green Belts and Yellow Belts. Education and training permeate the organization. Eventually every employee actively contributes to the production of breakthrough project results: cold cash to the bottom line.
The Six Sigma Profit Strategy Six Sigma improves profits by aiming at perfect products, services, and processes. In a 6σ culture, everyone is expected to enthusiastically argue in favor of perfection. A passionate work ethic attitude carries weight in a Six Sigma culture. Protests over the possibility of a “diminishing rate of return” indicate an individual does not understand 6σ fundamentals. The lower case Greek letter, σ, is pronounced ‘sigma.’ In the professional world, σ is the symbol for the population ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
91
standard deviation. The sample standard deviation, along with the five other elements in a complete vector analysis, comes from raw data. It quantifies the amount of random or chance variation that occurs around the average in any, and every, given set of data. To understand and embrace the universal Generalization of Chance Variation is to enter the world of Six Sigma. Try the following experiment to demonstrate this physical law for yourself. First, find a friend you admire. Choose someone with whom you can discuss controversial information. Now, each of you needs to print the letter “a” 10 times on a piece of paper in the exact same way with no variation.4 Go on. Try it. This exercise is a trick. The task is completely impossible. Differences in writing tools, variations in ink, paper texture, handedness, fatigue, font, attention span, concentration, your interpretation of our instructions, and an infinite number of other variables all contribute to natural variation. Natural variation is present everywhere and always. It is ubiquitous. It is a law of our universe, as powerful as gravity. Every good product and every service suffers from the inconsistencies caused by variation. J. Bernard Cohen, the eminent historian, considers knowledge of Chance and/or statistical variation to be the distinguishing characteristic of our generation’s Scientific Revolution. “If I had to choose a single intellectual characteristic that would apply to the contribution of Maxwell [though not directly to his revolutionary field theory], Einstein [but not the revolution of relativity], quantum mechanics and also genetics, that feature would be probability.”5 We agree. This Six Sigma Revolution in business and science is defined by evidence that is based on Probability rather than determinism.6 Like it or not, probability overthrows old doctrine. There is no polite way to summarize the impact variation has on an individual’s world view. Probability, dressed up in the Six Sigma costume, is replacing old ways of knowing—revelation, intuition, and reason—with the disciplined analysis of experimental observations. Six Sigma unifies the scientific method and business. Evidence-based decisions and the power in a vector analysis are the router connections between the two disciplines. In ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
92
Evidence-based Six Sigma answer to the meta-questions, “Does this Six Sigma stuff really work?” and, “Can you prove it by replicating your results?” The answer is unequivocally, “You bet.” With any and every set of raw data we can construct a tetrahedron, the cornerstone of statistical evidence. When a standard deviation is combined with an average, we can make valuable predictions based on a family of probability curves and surfaces (Figure 2). When one knows the average and standard deviation (σ) of a process, one can improve that process to near perfect, 6σ, performance. Perfect quality first time every time is valuable. This value can be measured with money.
Figure 2 Data matrix software automatically transforms the cornerstone of evidence into probability distributions.
Figure 3 illustrates old school 1980s corporate Quality Improvement (QI) aims. Way back then, ‘three-sigma’ quality was the target.7, 8 This means that the 6σ total process spread just fits between the lower and upper specification limits (LSL and USL). At best, this means that 99.7% of process outcomes satisfy customer requirements. This near 100% quality sounds better than it is. Recall the unacceptably wide variation in the prior chapter’s bar chart bamboozling comparison. At its best, a three-sigma 99.7% distribution promises ‘only’ 2,700 defective outcomes per million produced.
A three sigma process may actually produce as many as 67,000 mistakes or defects per million (DPM). This is because processes typically drift by about 1.5 standard deviations around their long term average.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
93
Figure 3 Three-sigma quality means that the 6σ total process spread just fits between the lower and upper specification limits (LSL and USL). At best this means that 99.7% of process outcomes satisfy customer requirements.
To put these numbers into perspective, ‘three-sigma’ aviation safety would mean several airline crashes each week. In health care, it would mean 15,000 dropped newborn babies per year. Banks would lose thousands of checks daily. As it is, three sigma (3σ) quality costs businesses between 25-40% of their annual operating income in waste and rework. Six Sigma breakthrough projects aim to reduce the standard deviation. High-leverage processes that affect business, manufacturing, or health care delivery are the prime targets. The Six Sigma bell curve in Figure 4 covers only one half of specification range. This illustrates the effect of a smaller standard deviation, σ. The Six Sigma one-part-per-billion (PPB) Six Simga bell curve in Figure 4 covers only one-half of the specification
Figure 4 A Six Sigma capable distribution covers only one half of the specification range.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
94
Evidence-based Six Sigma range. This illustrates the dramatic financial benefit of reducing the standard deviation. Even when the process drifts, only 3-4 defective outcomes per million (DPM), can occur. In a σ =$1.00 example, a Six Sigma breakthrough would result in a standard deviation that equaled $0.50 or less. When this goal of perfection is achieved, costs related to waste, rework, inelegant designs, and needless complexity disappear. The proven rewards for achieving 6σ are, 1) excited customers and 2) improved profits. Historically, each Six Sigma project generates a 100-250K benefit. Full-time corporate 6σ Experts, Black Belts who currently earn about 120K in salary and benefits, lead three to four projects per year that generate $1 million in hard dollar, bottom line business benefit. This 10:1 rate of return is so dependable it has become a tradition. Prior to the development of Six Sigma in the late 1980s, the only people earning their livings full time using these tools for breakthrough projects were consultants. We were the only ones willing to study out-of-date textbooks, use handheld calculators, rulers, graph paper, and DOS programs. Thank heavens those days are behind all of us now. Anyone and everyone can enjoy the benefits of vector analysis applied to a data matrix. Six Sigma style profits are now a matter of personal choice. The Lucrative Project Results Map Flow diagrams and process maps simplify work. They make hidden process dynamics visible. Seeing waste and complexity helps people eliminate both. Flow diagrams like Figure 5 can also be used to create processes that produce perfect results. To read the diagram, begin with the hard copy documentation symbol at the upper left hand corner. Follow the arrows through each of the four levels to the right hand page bottom. The acronym used to describe the classic 6σ process is DMAIC. DMAIC stands for the iterative 6σ project cycle of Define, Measure, Analyze, Improve, and Control. Once a project is completed, the process described by this map begins again. This cycle never ends.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
95
Figure 5 This flow chart has guided projects toward bottom line business results for years.
�������������������������������������������
��������� ��������������� �������������� �������
�������
��������� ���������� ������������� ���������
����� ��������
��������� ������������������ ����������� ������������� �����������������
���������� �������
�������������� ��������������� �������������� ����������������
� ������������ �����������
������������� ��������� ������
������ �����������
��������� ������ ������������ ��������� ������
���������������������� ��������
��������� ������������ ��������
������������ �������� ���������� ���������� ����������� ���� �������
���������� ������������ ���������� ���� ������������ ���������
��������� ������������
����� ����������� ������ �������� ����� ������������
�������������� �����������
������������� �������� �����������
�������� ��������� ����� ����������� ��������
�������� �������� ������
���������� ��������� �����������
��������
��������� �������������� �������������
�������� ������������� ������������
������������������������� ������������������ ���������������
©
��������� ������� ������������� �������
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
96
Evidence-based Six Sigma Define, Measure, Analyze, Improve, Control The voice of the customer (VOC), customer satisfaction and profit goals come first and last in the Six Sigma DMAIC cycle of improvement. The map marks the boundary of each phase. As 6σ breakthroughs help companies surpass quarterly and annual financial targets, long-term objectives are continuously upgraded to sustain momentum. The series of five steps in the top row and the two final steps in the bottom row are top-level management and leadership responsibilities. The middle three levels are Black Belt project tasks. Each of these steps takes time, so every 6σ project result needs to be substantial and financial. Results interpretation, improvement and control require close collaboration between top-level leaders and Black Belts. The process of interpreting statistical results, making an evidencebased decision, optimizing a system, and implementing improvements can and does flatten bureaucracy. Occasionally organizations that value bureaucracy manage to “do Six Sigma” while they find ways to sustain paperwork, committees, and supervisory redundancy. Six Sigma window dressing is immediately apparent to any knowledgeable observer. We advise potential clients who are fond of their bureaucracies to stick with Old School Management methods. Evidencebased decisions and Six Sigma will bring them nothing but trouble. Employees will openly question executives. Costaccounting reports and risky capital investment Proformas will be challenged with physical models. Six Sigma programs are seen as disruptive when a business values group think. Don’t laugh. Many do. The ones we have worked with and for are populated with delightful, friendly people. These folks just happen to draw an interesting set of Six Sigma project boundaries. Senior management processes and decisions are off limits. “Don’t go there.” In companies with a full commitment to evidence-based decisions, there is broad-based organizational involvement. This commitment is the key to perpetual breakthrough project success at the highest levels of the company.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
97
Six Sigma employs just about every effective management tool that has ever been developed. Any project management tool you can think of that has proven to be useful is now called a Six Sigma Tool. For example, the project management chart developed by Henry L. Gantt in 1917, called a Gantt chart, is still useful and very much in vogue.9 A PERT (Evaluation and Review Technique) chart, which provides an alternate Gantt chart view of a project, is also popular. In actual practice Six Sigma focuses relentlessly on completing projects within 90-120 days. Experience shows that if a Six Sigma project improvement team fails to deliver bottom line business dollar value within this time frame, organizational commitment to 6σ instantly wanes. We saw a most eloquent occurrence of this phenomenon in a CEO’s behavior. After a few months of Six Sigma hoopla, people noticed that he wasn’t using evidence unless it supported the foregone corporate agenda. Projects were not being completed on time. Resistance to evidence-based decisions grew. One day, he casually observed to the vice president in charge of implementing Six Sigma, “Six Sigma is ephemeral.” The VP looked the word up and discovered, to his dismay, ephemeral means “dead in a day.” As a side note, it was interesting to see this Six Sigma initiative generate about $6 million in bottom line benefits by the year’s end. The dollar per dollar return on investment, ROI, was only 5:1. Nevertheless it was informative to watch a Black Belt compete, particularly an experienced Master Black Belt. They do what it takes to bring home the bacon. Though it is management’s responsibility to keep the improvement fires burning, project delays and passive criticism are favored benign neglect techniques. Old school managers can and do successfully use neglect to sabotage Six Sigma. Make no mistake. The Six Sigma field is littered with the corpses of failed Black Belt Projects. The successful disruption of projects generally returns the culture to less demanding performance standards. The Institution of Old School Management thinking does not surrender until it is surrounded and expelled. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
98
Evidence-based Six Sigma Therefore, based on experience, we strongly recommend that once a company commits to evidence-based decisions, it stay focused on the money and deadlines. �
�����������������
�������� ��������
���������� ��������
������������ ����� ����������
���������� ����������
�������������
����� ������� ��������
��������������������������
Table 1 You can program a spreadsheet to help you choose best projects.10
�������������������� ��������� ���������� �������������� ����������� ��������������������������� �������������
�
�
�
�
�
�
�
�
�
� �����
�
�
�
�
�
�
�
�
��
� �����
�
�
�
�
�
�
�
�
��
� �����
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
� �����
�
�
�
�
�
�
�
�
�
�
����
�
�
�
�
�
�
�
�
�
�
����
����
Lucrative Project Selection Selecting and prioritizing the most rewarding projects is a most important first step. Since time is money and money is time, the selection process must be efficient and fast. Table 1 is a simple, virtually universal project evaluation spreadsheet that has emerged as a favorite around the United States. Six Sigma team leaders put breakthrough improvement project ideas into this hopper. The hopper is always open, but depending on the culture, new projects are usually given serious review at quarterly and annual intervals. During project review meetings, each idea is ranked from low to high, 1-10, in five or more categories. These values are multiplied to create a priority rating. This project prioritization process promotes a consensus style agreement that has some quantitative structure to it. We have seen it improve interpersonal working relationships as it generates lists of breakthrough project targets. The suggested project with the highest total project priority number is first and so on. Clear operational definitions are a fundamental part of project selection. The project and its related issues must be defined operationally. Operational definitions must be practical. If senior management, the project champion, or the Black Belts who are assigned to the projects do not share a common understanding of these definitions, problems arise. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
99
For example, write down your definition of the word ‘pan.’ Good work. Your definition is correct. So are at least 20 others. In Spanish, pan means bread. Pan is a cooking container, a depression in the earth, a cavity in the lock of a flintlock, and the Greek god of the woods. You can pan a camera or pan for gold. Believe it or not, the misinterpretation of a three-letter word has been known to derail projects. For this very good reason, experienced Six Sigma Master Black Belts and Black Belts are very specific when they define what it is that they intend to count or measure. Do more of what works. Here is another classic example that illustrates why clear operational definitions are crucially important to even a simple process like counting. Count the number of f’s in the following paragraph. FOR CENTURIES IMPORTANT PROJECTS HAVE BEEN DEFERRED BY WEEKS OF INDECISION AND MONTHS OF STUDY AND YEARS OF FORMAL DEBATE. How many did you count? Pause to write your answer here before moving on. _______ Depending on how you decided to define the letter “f ” there are seven possible correct answers. There are no lower case, italicized f ’s. So, zero is one correct answer. If you decided to count any F, there are 6. If you proof read phonetically, in other words you defined “f ” by the sound of the letter, the F in each OF sounds like a “v.” So, if you defined an F by the way it sounds you could have counted 1, 2, 3, 4, 5, or 6. Any one, or all of these answers, taken in the context of its definition, could be considered to be correct. During the project selection phase, perfect Six Sigma performance expectations called Critical to Quality (CTQ) or Key Quality Characteristics (KQC) are defined in statistical terms. Without a statistical definition, there can be no objective evidence. The definition must include an average and a standard deviation. Both come from a vector analysis applied to a data matrix. Since the Profit Signal is also automatically produced by this analysis process, it is considered to be part of a comprehensive operational ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
100 Evidence-based Six Sigma definition. These high analytic standards are used at all levels in the organization. Once operational definitions are agreed to, an average and a desirable standard deviation for project outcomes are targeted. These figures are accompanied by the expected dollar value of benefits the company can look forward to harvesting. When projects have been identified and Key Quality Characteristics are defined, financial models are used to create credible bottom line profit signal estimates. Financial Modeling and Simulation Six Sigma budget models are dramatically different from, and superior to, spreadsheet arithmetic Proformas. Every manager who has actually participated in the old-school ritual called “spreadsheet scenarios” must candidly admit to making the numbers up. The old school cost-accounting variance analysis encourages confabulation by eliminating 5 vectors, or 83 percent, of all the information contained in raw data. This is a covert impropriety if there ever was one. By using only one vector, and masking the other five vectors, almost any story holds water. A reliable forecast is as transparent as an authentic analysis. All data and all elements are revealed. When correctly employed, Six Sigma tools raise the standards of what does and does not constitute a credible Proforma, scenario, or forecast. Increasingly accurate predictions put the world of continuous spreadsheet revisions to shame. An abacus cannot beat a super-computer no matter how fast one’s fingers are. Legitimate financial forecast models are created using vector analysis rules. Financial Engineering News is one of many trade publications that helps professionals get up to speed on the use of these tools.11 Dr. John M. Charnes is a frequent contributor. As the Area Director for Finance Economics, and Decision Science at the University of Kansas, School of Business, Dr. Charnes exemplifies leadership in the field. He used the Decisioneering product called Crystal Ball, to create a 16 module self-guided study course that we think is excellent. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
101
Figure 6 One popular Six Sigma software program uses flow diagrams to graphically detail the iterative cycle used to create and improve financial forecasts.
His open system flow diagram, with clouds representing thought processes, is shown in Figure 6. In a data matrix driven budget forecast, the historical data underlying each budget assumption are graphed in 2, 3 and more dimensions prior to including that assumption in the forecast model. Once assumptions are validated, multivariate models incorporating factor interactions, correlations, and entrepreneurial assumptions are created. The model can then be simulated thousands of times in seconds. The output is presented graphically. Simulation is proving to be as beneficial to financial managers as it is to engineers, jet pilots, doctors, and student drivers. Working under old school constraints, engineers had to build expensive physical models to test their ideas. Pilots had to practice first solo flights at 600 miles per hour. Surgeons had to test new techniques on live patients. Parents had to take their teenager into traffic and hope for the best. With multi-dimensional computer simulations, new designs, flying skills, surgeries, and even freeway entrances can be tested “off-line” first to minimize risk. The benefits to simulation are objective and overwhelming. This is why computerized simulation is a “Six Sigma Tool.” If you look closely under the hood of reputable simulation applications, each has a data matrix and vector analysis for sparkplugs.12 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
102 Evidence-based Six Sigma Beginning in the late 1980s, inventive software manufacturers began to develop programs that forced spreadsheets to behave like a data matrix. The geometry that guided their design creates graphic results that look terrific. These macros are now mature modeling programs. They are a joy to use. Every day we thank the General Electric Senior Vice President who took time out of her day in 1997 for a cold call telephone interview. She explained how these programs push Six Sigma forward. Her counsel was, and remains, rock solid. With the finance simulation tool add-in, budget forecasts inherit the power of a vector analysis. Vector models do an impressive job of helping decision makers visualize probable outcomes. They are affordable tests. They let leaders meet and beat breakthrough project goals. Analysis rules, quantification, continuous feedback and discipline improve model forecasts over time. Beyond the covert elimination of 5 analysis vectors, spreadsheet arithmetic budget models and “what-if ” scenarios fall short of evidence-based decision standards in significant ways. 1. With a spreadsheet, an individual number in a cell is accepted on face value. This number frequently misleads because it is not framed in a meaningful context. Without analysis context—an average, a standard deviation, probability information, and an analytic graph—people must guess at the number’s meaning in relation to the other variables in a system. 2. Spreadsheets encourage analysts to believe in the great bamboozle. Many now are convinced simple addition, subtraction, multiplication and division are appropriate tools for analyzing complex, multivariate systems. Because the columns and rows in a spreadsheet look just like the columns and rows in data matrix, many conclude equivalent answers are automatically produced with each one. This error is serious. a. Spreadsheets are wildly popular because they let anybody do absolutely anything with any number. It all looks legitimate! Though spreadsheet arithmetic sets the standards of evidence at a comfortably low level, they produce illusion rather than insight. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
103
3. Spreadsheet scenarios are usually created using OneFactor-at-a-Time (OFAT) methods. OFAT analysis and experimentation methods are no more reliable now than they were when Frederick Taylor used them in the 26 years leading up to 1911. Not only do they not yield an accurate answer, they waste time. Conclusions reached using this method are at odds with physical Laws of the Universe. By over-simplifying problems, answers are notoriously unreliable. 4. Spreadsheet scenarios create a false impression of precision. Spreadsheet numbers are dressed up in impressive looking data arrays. These images shout out, “Hokum!” Nevertheless they are routinely presented, accepted, and framed as “certain forward thinking statements” in a social gesture of courtesy that smacks of hubris. In corporate hierarchies these courtesies force otherwise intelligent, well-meaning people to forget what they know about mathematics. People are persistent when it comes to juggling numbers. Human nature is tireless in its allegiance to irrational beliefs. Martin Gardner’s Fads and Fallacies observation rings as true today as it did when he wrote it in 1952. “How easy it is to work over an undigested mass of data and emerge with a pattern, which at first glance is so intricately put together that it is difficult to believe it is nothing more than the product of a man’s brain… Consciously or unconsciously, their perceived dogmas twist and mold the objective facts into forms which support the dogmas, but have no basis in the exterior world.” 13 Simulation programs give spreadsheets like Excel a new lease on life. Macros that follow the rules of evidence are bringing high standards to the world of accounting and finance.14 This is a very good thing. Once a 3D cube model is embedded in a spreadsheet, analysts have a much better grasp on the range of possible budget outcomes. Likelihoods and probabilities are presented automatically in attractive visual graphs. With simulation, managers can perform tens of thousands of multivariate scenarios in minutes. This is less time than it takes a skilled controller to ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
104 Evidence-based Six Sigma complete a single, One-Factor-At-a-Time (OFAT) budget forecast scenario. In addition, and for no extra charge, the simulation automatically produces a sensitivity chart that resembles a spreadsheet bar graph. Vector analysis sensitivity charts rank Profit Signals according to the strength of the evidence for each factor. Sensitivity charts expose counter-intuitive patterns that are masked by spreadsheet arithmetic. A sensitivity analysis ensures that management focuses on the key variables that have the most impact, rather than being distracted by variables they think may be most important. Figure 5 is a sensitivity chart illustration. Simulation can increase one’s level of confidence as business decisions are made in the face of uncertainty.
Compare and Contrast Analysis The classic budget forecast (Table 2) is usually created by estimating three outcomes: 1) best case, 2) worst case, and 3) most likely case.15 Since there are no rules, personal opinion and a consensus are the only evidence required for making a decision on the decision to pursue the NanoTech Widget. NanoTech Widgets solve problems. They are multi-purpose. With a projected profit of $9.2 million, they are a sure fire new product. Note how the forecast in the bottom right hand cell catches the eye.16 When this spreadsheet was analyzed 1,000 times in under six seconds using the data matrix introduced in the Five-Minute PhD, a much different picture emerged. Figure 7 does not tell the manager what to do. However, it does point out that the most probable outcome is a $14.4 million loss rather than a $9.2 million gain. Once the simulation is complete, the analyst or manager can evaluate which of the variables has the greatest impact on the bottom line. Predictably, factors are ranked in importance according to the relative strength of statistical evidence (Figure 8). ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
105
Table 2 The classic old school budget forecast for new product development presents assumption numbers without the benefit of either context or evidence. The average, standard deviation, p-value, and analytic graphs are ignored. Forecasting spreadsheet analysts are simply expected to correctly guess all values, evidence strength, and factor interactions.
Figure 7 Based on all the actual data at hand, there is a 77.9% chance of breaking even with the NanoTech Widget. There is only about a 50/50 chance of making the projected $9.2 MM. The most probable outcome is the $14.4 million dollar loss highlighted at the left side of the forecast.
Simulations and legitimate financial forecasts are standards in Six Sigma breakthrough projects. We have seen simulations effectively tackle budgets with up to 77 variables. The level of thoughtfulness this tool creates is well worth the time investment required.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
106 Evidence-based Six Sigma
Figure 8 Success in launching with the NanoTech Widget product depends on the company’s ability to penetrate the market.
Process Maps Maps, from Babylonian clay tablet cartography to downloadable Internet driving directions, are universal communication tools. Since maps have proven their value they too are a “Six Sigma Tool.” It is no accident that a process maps are the first-choice tool taken down from the shelf after a lucrative new project has been selected. A good process map is as multidimensional as a set of nested Chinese Boxes or Russian dolls. Though nested boxes and Russian dolls are not official Six Sigma tools, these analogies encourage people to look more deeply than a surface appearance. The outermost Russian doll is called a Matreshka or grandmother. Succeeding generations are contained within her.17 Miniaturized generations are refined replicas. Each three-dimensional replica must be produced using fewer resources and a higher degree of precision. So it is with Six Sigma process maps.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
107
First blush drawings can span pages. Over time, these are simplified and distilled into diagrams that illustrate and endorse only essential elements that pull the system forward efficiently. In the same way that the Space Shuttle Radar Topology Mission used vectors, geometry, and computing power to map 80 percent of the earth’s landmass in only 10 day’s time, Black Belts are expected to map the nested dimensions of a work process in about a week. Practice makes perfect. This is one of the skills that is worth the practice time investment. These maps are impressive. The ‘Six Sigma Matreshka’ in Figure 9 is a Suppliers, Inputs, Processes, Outputs, and Customers map or SIPOC for short. It is assumed that this system has feedback loops throughout. These loops are not illustrated here in order to present a clean, simple picture. Figure 9 Black Belts use personal interviews, first hand observations, and measurements to complete this map. Drawing these maps from the end to the beginning is the best way to produce a meaningful SIPOC map showing all the relationships.
���������
����������
������
���������
�������
���������
���������
������� �����������
��������
�������������������� ���������������
����������� �������
���� ����
������ ��
����������
������������
�������
����� ���� �������
������������
���������� �������������
����������� ������
��������
��������
��������
©
��������
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
108 Evidence-based Six Sigma Invariably, every Six Sigma map uncovers Figure 10. This Six Sigma drawing describes the “hidden-factory.” Its appearance has a Charlie Brown and Lucy Van Pelt quality to it. Just as the start of every football season is marked by Lucy tricking Charlie into trusting her for the inevitable betrayal, the hidden factory always makes an appearance.
���������
��������� � ����������
Figure 10 The hidden factory of rework in this map includes Processes 4-6 and the related delay.
���������
�������� ������� �������
���
������� ����� ���������� ������� �����
��������� ���������
�� ��������� ���������
����������������� ������������ �������������������� ����������� ����������
�������� ��������� ���������
��������� ���������
Waste and rework plague every production and service delivery process. They always happen where: 1) a loop reverses the forward motion of the product, 2) a delay, bottleneck or constraint slows process flow, or 3) a barrier stops production altogether. Hidden factory maps are often posted in conspicuous places, from the boardroom to the individual work space on the factory floor. Mapping is a documentation discipline that is rewarding and informative. The most sophisticated ones, called lean process maps, are drawn in an old fashioned, lowtech way using paper and pencils. Lean maps have a lexicon and icon system that is worth studying.18 Lean is a separate business tool that deserves, and has, its own literature. Suffice it to say lean maps document the entire value stream. They track information and material flows from start to finish through an organization. Management responsibilities are visible and a host of snapshot measurements are recorded. Lean metrics make sense. They include uptime, working time minus breaks, cycle time (C/T), changeover time (C/ ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
109
O), value added time (VA), Takt time, and production lead time (PLT). In addition to its own acronym, each time has an exceptionally specific operational definition. Times are recorded in days, hours, minutes, and seconds. The ‘timeis-money-money-is-time’ theme dominates lean thinking. Rightfully so. With the lean Six Sigma strategy in place a second can be, and often is, literally worth thousands of dollars. For example, in one San Jose Internet router factory a 2 foot by 2 foot by2 foot pile of scraped motherboards was time-valued at more than $6 million dollars. In addition, lean maps record production batch sizes for every product interval (EPE), First-In-First-Out (FIFO), the numbers of operators, inventory turns, a plan for every part (PFEP), the number of product or service variations, and the scrap rate. Lean measurements and flow mapping earned their way into Six Sigma the old fashioned way. They work. This is why lean maps are a “Six Sigma Tool.” Their record of extremely profitable achievements began in the 1950s Toyota production system and continues to this day.19 Like a vector analysis, those who are familiar with lean tools do not argue against them. To do so would be as foolish as arguing against the speed of light, the existence of gravity, or the impact of variation on measurement. These maps help people identify process factors, known as the X’s, that may be driving the system toward profits or losses. Profits and losses, the dollar value of a process outcome, are called Y’s. A thought map, Figure 11, shows how these factors become a series of hypotheses in a data matrix. Once the matrix is filled with measurements, a vector analysis will point out strong and weak Profit Signals with objective standards of evidence. “Hidden Factory” costs, or the costs of waste and rework are called the Costs of Poor Quality (COPQ) Since the 1950s, breakthrough projects have focused on eliminating these expenses. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
110 Evidence-based Six Sigma
������������������������� ������������������������ ������������������ �������
Figure 11 Maps help improvement teams identify variables that will be subjected to a 3D vector analysis.
�������� ��������� ����������� �
����������� �
����������� �
������� �������� ���������� ���
�������� �������� ��������� ���
�������� ��������� ���������� ���
���� ����������
���� ����������
������ ��������
The Costs of Poor Quality The origins of the “Costs of Poor Quality” idea can be traced to Walter Shewhart’s invention of the quality control chart on May 16, 1924.20 The quality control chart is yet another way of graphically viewing a vector analysis.21 Shewhart was a physicist. He was also the accomplished statistician, friend, and colleague of Ronald Fisher who volunteered to care for Fisher’s six children during World War II. A German U-boat’s torpedo sank this plan in 1940.22 The dollar figures Dr. Shewhart symbolically placed on the corners of Fisher’s work 80 years ago are today’s Six Sigma costs of quality. In concrete terms, Armand V. Feigenbaum is credited with developing the first dollar based, quality reporting system while working at General Electric in the early 1950s.23 It is worth noting that this development ultimately can be linked to Jack Welch’s 1990’s Six Sigma initiative. Why are the costs of poor quality so important to Six Sigma breakthrough projects? Simple. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
111
Revenues are taxed. One dollar in newly earned revenue can produce as little as one penny in new earnings. One dollar saved through the elimination of waste and rework drops to the bottom line as one dollar. To a certain extent, once a Black Belt gets the hang of the breakthrough project system, saving major dollars is like shooting fish in a barrel. Each of four poor-quality cost categories can be leveraged. Figure 12 Prevention and appraisal investments, often referred to as costs, are relatively static. Internal failure costs are hidden factory expenses that remain invisible to customers. External failures are mistakes and errors that are highly visible to the customer.
Figure 12 Textbook example of a Cost of Poor Quality (COPQ) flow chart used by a Black Belt engineer, Scott Erickson, to persuade senior management to embrace evidencebased decisions.
Since Feigenbaum first created this classification system, prevention investments have been expected to produce, and have produced, a 10:1 return. For Six Sigma these costs include training, education, analytic software, planning, vendor certification using Six Sigma quality metric standards, and quality assurance system costs.
Appraisal investments include quality audits, inspections, testing, maintenance, and information systems. Information systems (IS) designed using principles of evidence-based decisions are far less expensive than those that are not. If ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
112 Evidence-based Six Sigma your company is looking for a place to begin Six Sigma, we encourage you to create an IS strategy that is based on sound geometric principles. The easiest way to learn if your system meets these standards is to ask your IS department to show you their data matrices and cube experiment arrays. This question invariably raises eyebrows. The vast majority of IS systems are modeled after spreadsheets. Transforming this system, or transferring the information in it to a data matrix, entails rework costs. Once the investment is negotiated, the payback is spectacular. Internal failure costs include all scrap and rework. Retests, Failure Mode Effects Analysis (FMEA), excess inventory costs, Corrective And Preventive Actions (CAPA) and productivity losses are recorded here. Finally, external failure costs are problems that land in the customer’s lap. Liability suits, warranty costs, returns, engineering changes, marketing and sales errors, complaint handling, and related equipment required to rework products must all be tallied up. In our increasingly litigious society it is almost impossible to overstate the costs of failure. Shewhart wrote humorously about the reality of 1939 quality costs, “I am reminded of the old saying: when a doctor makes a mistake he buries it; when a judge makes a mistake, it becomes law. I would add in the same vein: when a scientist makes a mistake in the use of statistical theory, it becomes part of ‘scientific law’; but when an industrial statistician makes a mistake, woe unto him for he is sure to be found out and get into trouble.”24 The best way to protect any business from external failure costs is to produce a perfect quality product every time a product is produced. Delivering perfect quality services 100 percent of the time is a powerful a business strategy. Perfect processes can and do produce virtually perfect outcomes. Only processes that are capable of producing perfection do produce this level of quality. A process capability index, known as Cpk, is the analytic measure used in graphic presentations of evidence documenting perfect quality.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Evidence-based Six Sigma
113
Process Capability We will use dice rolls for our example. (Chapter 7 will extend this experiment to include 4 die.) Feel free to roll your set of dice until you get 5,000 measurements. Or, you can accurately simulate the outcome of 5,000 rolls in a minute using software. Comparing both methods will give you a good feel for the value of Six Sigma vector analysis software. Each of us has personally rolled dice more than 5,000 times. We chose the simulation method for this example. Statistical software does not know that the outcome of throwing a pair of die is constrained to the range 2 to 12. Therefore, it calculates and estimates statistical limits for a distribution as if it were not constrained. In this way our teaching analogy is flawed. Skeptics complain, “See. Statistics lie! They can’t even handle dice rolling.” These complaints are ludicrous. Ignore them. The point in this exercise is a principle. And, by now you get the point. With the click of a mouse button, software graphs our data and tells us how capable it is. For this example, you can see we set our Lower Specification Limit (LSL) for perfection at 2 and our Upper Specification Limit (USL) of perfection at 12. Figure 13’s process capability curve tells us our process is not capable of producing perfection. The Cpk value is calculated by taking that old favorite, σ, and dividing it into the spread of the data. A Six Sigma process yields a Cpk of 2 or more. If and when the tails of our curve fall above and/or below our perfection specifications, these portions would be scrap and rework.
Let’s game this system and improve our Cpk by setting our LSL and USL perfection expectations at –10 and 30. Figure 14 shows that our process is a smoking Six Sigma process fully capable of producing perfect quality outcomes 99.99999 percent of the time! We’re in the money. Note how tightly the distribution curve is centered on the target of 7. In real life, Six Sigma companies earn “high” Cpk values not by lowering their standards, but by raising them relentlessly, geometrically, and exponentially. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
114 Evidence-based Six Sigma �������������������
����� �� ������� �������� ����� ���� �������� ��������� �������� ��������������� ����������� ����������
���� �� ������� ��� ������������� ���������������� Figure 13 This process is not capable of perfection. Its Cpk value is only 0.640.
������������
���
����
���
���
���
�
��
To achieve these levels of perfection, they use a vector analysis applied to a data matrix. The only way perfection can be pursued and achieved is by using quantitative measurements and analysis. The only set of tools that makes this rate of improvement possible is the scientific method and the geometry of a vector analysis. This is why Six Sigma is not a fad. This is why there is such a bandwagon rolling with Six Sigma Breakthrough Projects and 6σ tools.
�������������������
Figure 14 A Six Sigma process will produce perfection every time.
���� �� ������� ��� ������������� ����������������
����� �� ������� �������� ����� ���� �������� ���������
��������������� ����������� �� �����������
��� ���� ��� ���
��� �
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
��
Evidence-based Six Sigma
115
Endnotes
Box, George E.P., Hunter, William G., Hunter, J. Stuart. Statistics for Experimenters, An Introduction to Design, Data Analysis, and Model Building. New York. John Wiley & Sons, Inc. 1978. Pages 170- 201. 1
The body of knowledge that is widely regarded as the most comprehensive is posted by the American Society for Quality http://www.asq.org/cert/types/sixsigma/bok.html 2
Mikel Harry, a popular leader in the Six Sigma field, reported this history on a video tape recorded in 1995. 3
Shewhart, Walter. Economic Control of Quality of Manufactured Product. Brooklyn, D. Van Nostrand Company, Inc. 1931, page 5. 4
Cohen, J. Bernard. Revolution in Science, Cambridge, 1985, Belknap Press of Harvard University Press. Page 96. 5
Cohen, J. Bernard. Revolution in Science, Cambridge, 1985, Belknap Press of Harvard University Press. Page 96. 6
Our bell curve illustrations were inspired by a drawing originally produced by Control Engineering Online. 7
Deming, W. Edwards. Out of the Crisis. Cambridge. Massachusetts Institute of Technology, Center for Advanced Engineering Study. 1982. 8
As a sidebar note, it is interesting to know that Gantt patented a number of devices in collaboration with Frederick Taylor when they worked together at the Bethlehem Steel Mill on Taylor’s Scientific Management theory. 9
Inspiration for this particular grid came from Moresteam.com. http://moresteam.com/ Their on-line Six Sigma Black Belt course is interesting and informative. 10
11
12
http://www.fenews.com/ http://www.processmodel.com/ ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
116 Evidence-based Six Sigma Gardner, Martin. Fads and Fallacies in the Name of Science. New York, Dover, 1952. Page 184. 13
14
http://www.decisioneering.com
http://www.decisioneering.com This spreadsheet is used with permission along with the flow diagram for financial models. 15
http://www.decisioneering.com The numbers and layout of this budget come from Decisioneering’s tutorial example, ClearVision. 16
An interesting history of this symbolism can be found at http://www.nestingdolls4u.com/history/history.htm 17
18
http://lean.org
Womack, James P., Jones, Daniel T., and Roos, Daniel. The Machine that Changed the World. New York, Rawson Associates Scribner Simon and Shuster, 1990. 19
Harrington, H. James. Poor Quality Cost. New York, Marcel Dekker, Inc. 1987, page v. 20
Shewhart, Walter A. Economic Control of Quality of Manufactured Product. New York, D. Van Nostrand and Company, 1931. Page 40. 21
Box, Joan Fisher. R. A. Fisher, Life of a Scientist, New York. John Wiley and Sons, 1978. Page 377. 22
Harrington, H. James. Poor Quality Cost. New York, Marcel Dekker, Inc. 1987, page xiv. 23
Shewhart, Walter A. Statistical Method from the Viewpoint of Quality Control. New York, Dover Publications, Inc. 1986. 24
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 4
Case Studies
C
ase studies needed to meet four criteria. Though names, places, and data were altered to protect privacy, each story had to be true. It had to be entertaining. Each example also needed to graphically explain how evidence-based decisions produced crowd pleasing financial returns. Finally, the story needed to be a fair, representative sampling of what we each have repeatedly seen over the past 20 years of our professional life. Occasionally, the stories in this chapter trouble some managers. Though we tried, we were unable to completely resolve this perplexing, vexing journalism quandary. One senior executive reviewer echoed Daniel Sloan’s own 1986 Vice President of Marketing’s sentiments. “The stories in this chapter upset me. As a senior executive, maybe I just took them personally. They hit too close to home. It is difficult for me to keep reminding myself that what is past is past. I have to keep telling myself that evidence-based decisions can and will prevent me from repeating history.” Case studies are essential to understanding. We bit this bullet and chose to include them. We decided to tell them the way clients tell them. Recounting a Six Sigma project victory is like explaining a magic trick. Once a wizard’s secret is revealed, someone in the audience thinks, “Shoot! I could have done that.” But no one, including the magician, can do it without knowing how. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
118 Case Studies Magic tricks are illusion. Evidence-based decisions put real dollars in real banks. Evidence is a funny thing. Many of us are interested in evidence only when it confirms an existing belief or policy. This human tendency creates a resistance to transparent reporting systems in business and government. Confidentiality is necessary in business and government. Too often, confidentiality is used to justify secrecy. Another human tendency is to equate evidence with authority. Then both are tarred with a brush of cynicism. This position was summarized in the March 1998 issue of Discovery Magazine: “Anybody who claims to have objective knowledge about anything is trying to control and dominate the rest of us…There are no objective facts. All supposed ‘facts’ are contaminated with theories, and all theories are infested with moral and political doctrines… Therefore, when some guy in a lab coat tells you that such and such is an objective fact,…he must have a political agenda up his starched white sleeve.” 1 This “know-nothing” doctrine stems in part from inadequate science and mathematics education. It is contradicted by the documented successes of the evidence-based decisions that power Six Sigma breakthroughs. Still, it is also true that data can be suppressed, ‘massaged’ or just plain falsified. Disraeli’s comment “Lies, damn lies and Statistics” was a reference to this problem. No analysis method can deliver us from the unethical corruption of reported data. However, given good data in a data matrix, vector analysis makes it virtually impossible to misrepresent the information in that data. Vector analysis is based on immutable Laws of the Universe. It is transparent. All aspects are revealed. It uncompromisingly tells the truth. The cornerstone of evidence, a tetrahedron, symbolizes ‘solid evidence’. Transparency, full disclosure and international standards for data analysis are the reasons Six Sigma works. They are also the characteristics that some find most disturbing about Six Sigma. It is no surprise that spreadsheets have ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
119
sensational appeal. Spreadsheets snap tightly to the New Age mantra, “Tell your own truth.” New Age know-nothings can structure data any way they want, and they can analyze it any way they want.2 Transparency and secrecy, honesty and misrepresentation, are equally weighted options. Spreadsheets are the engines for the cost-accounting variance analysis. These methods are inherently one-dimensional. Each uses only one of the six vectors in the cornerstone of evidence. None of them recognize the essential Profit Signal and Noise vectors. In this sense, cost-accounting variance analysis suppresses five-sixths—83 percent—of the information needed for an evidence-based decision. Because break-even thinking and cost accounting variance analysis allow management to ignore five of the six reality vectors, it is easy to construct any story that is consistent with any one vector. Naturally, people tend to construct stories that favor their point of view. As Master Black Belt teachers, we use forthright honesty, computers, software, graphics, the cornerstone of evidence, and the New Management Equation to dispel the mystery surrounding evidence-based decisions. Once people harvest Six Sigma profits by making better decisions, objections diminish. Everyone wants to make more money in less time, with less work, and using fewer resources. Doing more of what works is a doctrine to embrace. Knowledge and reliable information start the Six Sigma DMAIC ball rolling. It leaves the trialand-error methods of old-school management in the dust.
Customer Service – Governmental Agency Political pressure was forcing a Washington State government department to improve the quality of their services or face the loss of $500,00 in funding as a penalty. The department’s Executive Director gave employees the opportunity to choose a consultant to help them in their efforts to maintain current funding levels. A Five-Minute ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
120 Case Studies PhD demonstration and evidence-based decision tools attracted their attention. Define: Jobs, including management jobs, were on the line. One-half million dollars in legislative funding was at stake. Negative regional news coverage over departmental problems made state citizens angry. Poor customer satisfaction had put this department on the legislative target hit list. To begin the project, flow diagrams and a Pareto analysis exposed breakthrough improvement opportunities. A specific criticism concerning this department’s performance had to do with the way it answered its telephones. Several full time clerical staff answered phones that literally rang off the hook. The agency’s executive director knew calls went unanswered. Armed with her own good judgment, she had instituted a department policy by edict. “All phones will be answered by the third ring. Answering machines are prohibited because they symbolize poor quality service.” The ringing phones were right outside her office, and she monitored her policy. Measure: The team of secretaries who answered the phones claimed they had a good solution to the problem. “We can’t get anyone to listen to us. We just do what we are told.” We suggested that they use a check sheet to collect data; we promised to help them present their evidence. There were a number of suspected causes, or hypotheses, for the unanswered phone flash point. These included: • Hypothesis 1 (H1): The day of the week makes a difference. Some days are busier than others. • Hypothesis 2 (H2):The time of day makes the difference. Some times are busier than others. •Hypothesis 3 (H3): The telephone line made the difference. One line was busier than the other one. Using a paper and pencil, the team of secretaries constructed a check sheet to record matches with the cube experiment data matrix (Table 1).
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
121
Table 1 The cube experiment data matrix guided the collection of data recorded by hand on a check sheet.
Analyze: The data matrix revealed a distinct profit signal. The main effect was so obvious everyone could see it immediately just by looking at the matrix. Figure 1 presents the data in a cube plot. Hypothesis 3 was the “big hitter”. The numbers of calls on line 2, the back face of the cube, were an order of magnitude larger than the number of calls on line 1, the front face of the cube. No other variable had an effect. ���������
�������
������
������
����
���������
�������
������� ����
Figure 1 All of the high numbers fell on the back plane.
������� �� � ����
������
������ ���
�����
���
������
Line two was sending a clear profit signal. It turned out that line two had been listed incorrectly in telephone directories across the entire state. This proofreading error was embarrassing. It would have been expensive to fix. No one had the courage to bring it to the attention of the agency’s executive director.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
122 Case Studies The executive director’s edict compounded the fear factor. Rather than addressing the core issue, the workforce decided it was much easier to keep their heads down. They became telephone operators and gave dialing assistance to callers. We took their evidence forward with a firm conviction that, in at least this case, the messenger would not be shot so early in a consulting engagement. Improve: One hour after the presentation of our evidence, a telephone answering machine was purchased and installed. The executive director gave this improvement her blessing with a belly laugh. The answering message announced the Yellow Pages error and then the correct number to callers. Six secretaries and other workers could now focus their attention on real work. Control: Telephone listing corrections were made the following year. A breakthrough in the proofreading process ensured 100 per cent, Six Sigma accuracy. This breakthrough played a role in persuading legislators to sustain funding at existing levels. The total time to collect data for the data matrix was five days. The analysis and presentation took one hour. Eventually one full time position was eliminated through attrition for a bottom line savings of more than $25,000. This, combined with avoiding the loss of funding, brought the total value of the project to $525K.
Days in Accounts Receivable A service company needed to reduce its number of days in accounts receivable, or AR days. The number of days in AR ranged from 35 to 110 days. A breakthrough improvement project could yield as much as 35K per month, or $420,000 per year in cash flow. Define: For more than a year debate had raged over what could be done to reduce AR days. Suspected causes for this ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
123
problem varied. These suspicions, or straw men hypotheses, included the following: • Hypothesis 1 (H1): Management is the solution. Good managers have short AR days. Bad managers have lots of days in AR. • Hypothesis 2 (H2): Sales calls are the answer. The number of visits made by a salesman to the customer is key. The more visits, the larger the number of AR days. The fewer the visits the smaller the AR days. • Hypothesis 3 (H3): The customer is the main reason for long or short AR days. Good customers pay fast. Poor customers pay slowly. • Hypothesis 4 (H4): The longevity of our customer relationship makes the biggest difference. Long-term customers pay more slowly because they know our business depends on them. • Hypothesis 5 (H5): The number of services provided determines the number of AR days. More services create complexity. Billing complexity slows payment. Measure: Significant AR data had been collected. These records were stored in file cabinets. Each customer, and there were hundreds of them, had its own manila folder. It was with a great deal of pride that the accounting team showed bills were filed in near perfect chronological order. The Chief Financial Officer of this company was committed to keeping productive hours in line and on budget. Workers in his department were required to do their jobs, as well as to work on breakthrough projects. No overtime would be paid for improvement tasks. A regular work schedule would be continued. Moreover, in order to keep operating costs low, no statistical software would be purchased. “Spreadsheets work fine.” We interviewed every employee and constructed process flow diagrams. We identified five important variables that might affect AR days.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
124 Case Studies
Figure 2 Statistical software automatically determines the best data matrix geometry for a vector analysis involving five independent variables.
The CFO had vetoed a recent budget request for a PC workstation and relational database software, so automatic queries and data mining were out of the question. Going to Plan B, we used our own statistical software to create an optimal data matrix for five independent variables at two levels each (Figure 2). This took all of five minutes. It is virtually impossible with a spreadsheet. Creating a data matrix is one thing. Collecting data that fits the profile of each run is another. A billing clerk and a billing manager volunteered to come in over the weekend and pull records. They believed they were familiar enough with customer profiles that they would be able to find bills that would match each of the 32 different “runs” in the matrix. These two front line leaders wanted to find out what combination of variables actually made a difference. They knew that if they found an answer, it would be valuable. Their daily workload was so challenging, they simply didn’t have time to array any more spreadsheet data than they already were doing for the CFO during the regular workday. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
125
Figure 2 shows the first 28 rows of the data matrix with the number of AR days visible in the far right hand response measure column. The every-other-row pattern of a short number of days in AR followed by a long number of AR days was evident immediately to the accounting department workers on a Sunday morning.
Analyze: Three strong profit signals emerged from the vector analysis we applied to their data matrix. The p-values in Figure 3 appear under the heading “Prob > F”. We could say with better than 99.999 percent level of confidence that the customer was a main effect. The computerized vector analysis also showed, with a 99% level of confidence, the length of the customer relationship was another active factor that influenced the number of days in AR. The main effects were controversial. Anxiety filled the air. ��������� ���
Figure 3 P-values less than 0.05 imply a 95 percent level of confidence or more in the results. The two factors, Customer and Relationship, and their interactive effect, were statistically significant at this confidence level.
������ �������� ������������ �������� ������������ ������� ��������������������� ����������������� ��������������������� ��������������������� ������������������������� ��������������������� ���������������� �������������������� ���������������� ��������������������
����� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ����
�� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ����
������ �������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ���������� ����������
������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� ��������
�������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� �������� ��������
The key difference between the two customers was widely known. Customer A was billed electronically. Customer B was billed manually. New customers were able to bill electronically. Old customers were not. The company’s Chief Financial Officer and Chief Executive Officer disliked computers. They still do. The CFO openly opposed the use of statistics. The CEO had excused the finance department from participation in breakthrough projects “until the data matrix and vector analysis tools proved to be useful.” A year earlier, the Chief Financial Officer had refused to approve the purchase of a $15,000 PC workstation ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
126 Case Studies in his department to keep costs down. The electronic billing and relational data base topics were verboten. Improve: The team spent a week gathering its courage and preparing evidence for a presentation to senior management. Following the presentation, the company purchased and installed a top-of-the-line workstation. A $1,000 request to purchase data matrix software for the finance and accounting department was denied. “Spreadsheets work fine.”
Figure 4 The two-level, five-factor, 25, vector analysis compares all the factor interactions using the traditional 3D cube.
Figure 4 explains part of the reason that executive resistance to evidence-based decisions continued. 3D vector analysis pictures do not look like bar graphs or pie charts. This particular finance department found 3D cube graphs to be upsetting. Figure 4 presents accurate AR day predictions for differing combinations of all five factors. Note that both of the top cubes have shorter AR days. When the AR Days come from customer A and a new relationship, AR days are lower than with any other combination of factors.
��������� ���������������������������
���������������� �����������
������
������
�����
����
������
������������
������
������
��
������������
�����
�����
������
������
��
������
�����
����
�������
�����
����
������
������
����
���
����
�����
������
�����
������
������
�����
������������
�������
���������������������������
���������������� �����������
������
������
����
����
���
����
������
����
������������
�����
����
������
������
�����
©
�������
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
���
����
�����
�����
����
������
����
��������
�������� ����
����
��������
��������
��
������
����
�������
���
����
Case Studies
127
Control: Results produced by lowering days in AR by 30 days exceeded the projected $420,000 cash flow gain in the first year. Total time required to complete project was 90 days. As the financial crisis passed, so did the use of evidence-based decisions. The heads up improvement team put their heads down and went back to work. To this day, the company has refused to invest in either the education of its finance and accounting workforce, or the purchase of data matrix software. “Spreadsheets work fine.” This experience taught us to present profit signals using a special kind of a bar graph known as a Pareto Chart rather than the more powerful cube. People just want the answer. The Pareto chart in Figure 5, which rank orders profit signals from strong to weak, gives customers what they want in a way that poses no visual threat.
Figure 5 Profit signals are easy to spot using a graph that ranks vectors from strong to weak.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
128 Case Studies Breaking the Time Barrier3 Long waits in hospital emergency departments are legendary. The list of likely causes includes over crowding, a shortage of nurses, an aging population, a shortage of inpatient and/or long term care beds, and a saturated primary care system. The Joint Commission on Accreditation of Health Care Organizations (JCAHO) has recognized the critical nature of Emergency Department (ED) overcrowding. JCAHO has instituted new Emergency Department Overcrowding Standards requiring a hospital’s serious attention. The Emergency Department is the front door of a hospital. It often accounts for a significant percentage of all admissions. Service excellence that meets or exceeds the public’s expectations is essential. The newly hired administrator of a community hospital, a certified Six Sigma Black Belt, selected the ED as the hospital’s initial Six Sigma Project on her first day of work. The ED Charge Nurse had told her, “Our ED is closed to ambulances. We are on divert status.” In response, the Black Belt RN, CEO administrator asked, “What are the standards of evidence you use when you decide to close the ED?” The Charge Nurse responded, “Well, we are simply overwhelmed. We cannot provide safe care if one more sick patient comes through those double doors.” The RN, CEO administrator, a 30-year veteran with a Masters of Public Health administration degree, glanced around. She saw vacant treatment rooms. Three staff members were cautiously watching her from behind the nurses’ station. They were all pretending to be charting. The nearby ED physician gave her an apologetic smile and said, “This happens all the time. You might as well get used to it.” The following Monday, after reviewing the hospital’s admission data and top revenue producing departments, the administrator called a meeting to discuss the “ED Divert” issue. The ED Medical Director, ED and Critical Care nurse managers, directors of the laboratory, imaging, environmental ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
129
services, and the Emergency Medical Treatment (EMT) director from the local fire department responsible for the paramedics all attended. Everyone was resigned to the status quo. No amount of effort could reduce ED divert time. It was an inevitable result of growing volumes. “Every hospital in the city is having the same problem. Why at Mid-Valley, their ED is closed twice as much as ours.” “If you want us to put our nursing licenses at risk, well . . .” “There are never any beds in the ICU.” “If the Cath Lab crew was in-house 24/7, why that would solve the problem.” “The CT tech takes call from home after midnight, . . . We’re always waiting for her to come in.” As they reviewed the actual numbers from data that had been collected and arrayed in a data matrix, they were surprised. During the past six months, the Emergency Department had been closed or on diversion (divert) more than 5300 minutes/month. This totaled eleven, 8 hour shifts, or three and two-thirds 24 hour days, or 12% of available time. Those closures penalized patients. They cost the hospital hundreds of thousands of dollars in potential revenue. The list of suspected reasons for going on diversion status were as varied as the professional team that sat around the table. Every member had her or his own favorite reason, or two or three, that they firmly believed was the primary cause of ED divert. The general theme was that Emergency Department diversions were caused, in large part, by “other” departments in the hospital, from Admitting to X-ray. The Black Belt CEO led a brainstorming process to identify Critical To Quality (CTQ) factors. This is a crucial DMAIC first step no matter what the project is. No one in the ED was familiar with Six Sigma techniques or tools. Nevertheless they began to wrestle with a complex process that involved most of the hospital’s departments. The Team began planning steps in the Six Sigma DMAIC breakthrough process. They designed an experiment. They ran it and analyzed their data.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
130 Case Studies
Break Through Results The profit signal vector analysis showed that once a decision to admit a patient was made, performance pressure was off. The admission process simply slowed to a near stop. An “acceptable” wait time for a patient being admitted was openended. Any amount of wait time could be rationalized. Once this practice was halted, everything changed. The department was astonished. In the first two months their initial project results list included: • The average hours on ED Divert dropped from 88 to 50 per month, a 48% reduction from the same period the prior year. • The number of Emergency Department visits increased by 12.6%. • The average ED Length Of Stay (LOS) shrank from 3.6 hours to 1.9 hours. • A 38.26% increase in Emergency Department gross margin was generated. • Patient satisfaction increased from 59% to 65%. • Catheterization Lab time dropped from 93 minutes to 10 minutes. • Intensive Care Unit bed availability increased by 10.6%.
DMAIC
DMAIC is the standard Six Sigma breakthrough project methodology. The application of DMAIC to this hospital’s ED overcrowding and diversion problem produced dramatic, measurable, sustainable results. Define issues systematically, statistically and practically. Issues identified included time on ambulance divert, ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
131
unacceptable ED patient length of stay (LOS), low patient and staff satisfaction levels, patients leaving without treatment (LWOT), and considerable lost revenue. The team established performance measures and targeted benchmark targets for each goal. Measure using maps, models, diagrams, and process flow diagrams. Identify and array CTQ factors. Collect data, observe the process and begin data mining. Door-todoor Length of Stay (LOS), Left Without Being Treated (LWOT) as a proportion of all patients, and patient satisfaction levels were identified as the CTQ factors the team wanted to study. They prioritized reducing (minimizing) Emergency Department Length of Stay (LOS) as the key response. Everyone felt that all other issues would improve if LOS could be reduced. Analyze data using a vector analysis applied to a data matrix. Prepare appropriate quality control charts and Design of Experiments (DOE) to determine CTQ factors. Improve the process using evidence-based decisions to power Six Sigma breakthroughs. Control the process to insure that break-through improvements were sustained. Wisdom Gained Along The Way
The knowledge experts on the ED Six Sigma team used inductive reasoning. They identified potential Critical to Quality (CTQ) variables they believed influenced the department’s Length of Stay (LOS). CTQ variables identified for evaluation were the patient’s gender, a “slow” and “fast” physician or nurse (identified by employee number), a decision to admit or not. Ready availability of an ICU bed, laboratory and imaging testing were other variables. The Black Belt Administrator, created an 8-Factor Designed Experiment on her laptop, using data matrix software, JMP 5.0. Evaluating these 8 factors required only 16 runs to complete. The data were gathered in less than 24 hours (Figure 6). ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
132 Case Studies
Figure 6 Custom design for an 8 factor, two level, Emergency Department Length of Stay (LOS) Experiment.
Before the project, ED staff and physicians ranked laboratory turnaround time as the most significant CTQ factor that influenced the length of stay in the Emergency Department. CT technician availability ranked a close second. The results of the Designed Experiment (DOE) were surprising to members of the Six Sigma Project team (Figure 7). This was one of many “ah ha’s.” Don’t trust your assumptions or your “gut,” even if you are an expert. Six Sigma techniques, including a carefully designed experiment and rigorous data analysis (computerized software makes it easy) provided evidence at the 95% level of confidence. This confidence level helped managers make critical decisions quickly.
Figure 7 Results of an 8-Factor Designed Experiment.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
133
The Project Team discovered the CTQ factor with the most impact on ED LOS was admission status. Those patients being admitted had a significantly longer Length of Stay than those who were treated and released. Running a close second, was the availability of an Intensive Care Unit bed. Both were at the 99.99% confidence level of significance. While these CTQ factors, admission status and availability of an ICU Bed, may appear obvious now, they were not at the outset of the project. Finding these two highly significant factors focused the team’s efforts. A drill-down of data revealed that less than 9 percent of the patients who entered the hospital’s ED by ambulance were ultimately admitted to the ICU. Yet, the most frequent reason given for instituting the ED diversion status and closing it to customers was “No ICU Bed.” Staff and physicians operated under the false assumption that “most” ambulance admits were very sick and “nearly all” would require an admission to ICU. There was a related assumption that the EMTs expected an ICU bed to be immediately available or they would take the patient to another hospital. The small percentage of admissions to ICU was a surprise to the EMT medical director. He voluntarily educated his staff so they would rely on the judgment of the ED staff. When the administrator began discussing ICU bed availability with the nursing staffs in the ED and ICU, she quickly uncovered an insidious attitude. It was, “Us against them.” Nurses (and, to a lesser extent, physicians) believed that “their” department worked harder and the “other” department was attempting to shift their workload to them. The lack of trust between the ED and the ICU required immediate attention. Nurse managers evaluated and resolved issues between their departments. They arranged schedules and provided time for nurses to “walk in the shoes” of nurses in the other department. Attitudes quickly changed. Nurses gained an appreciation of the unique and essential role each service provided to quality patient care. ED and ICU service medical directors developed patient admission and transfer criteria that were approved by the ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
134 Case Studies Medical Staff. The criteria, based on a patient’s need for intensive nursing care, authorized nurses to transfer patients out of ICU to open a bed for a new admission. Communication between the two departments was difficult. The ED nurse was required to provide a detailed report to the ICU nurse before she could initiate a patient transfer. Telephones might go unanswered in the ICU due to the immediacy of patient care needs. Working together, the nursing staffs developed a 1-page report that the ED nurse would fax to the ICU in the event they were unable to complete a telephone report. A transfer of the patient to the ICU occurred automatically 30 minutes after the report was faxed, unless ICU notified the ED to hold the patient. This is now an uncommon occurrence. An unintended but exciting result of the Six Sigma ED Project was a stunning reduction in “Door to Cath Lab” time. (This is a measure of time from the patient’s arrival in the ED to the initiation of treatment in the cardiac catheterization lab.) Before the ED project, average door-to-cath-lab-time was a respectable 93 minutes. While this time met national standards, it was longer than the hospital’s nearby competitor. EMTs transported their most critical patients to the competitor hospital because of their superior door-to-cathlab-time. A flow process diagram revealed the problem. Patients in the field with a potential diagnosis of myocardial infarction (MI) were evaluated by the EMTs, in consultation with the ED physician. When they arrived in the ED, they were reevaluated by the ED physician, including completion of lab work and a repeat EKG, before the cath lab team was notified. Drill down analysis of outcome data revealed that the EMTs diagnosed MI with nearly 100% accuracy. The delay in calling in the cath lab team cost precious heart muscle-saving time. With the support of the administrator for the potential cost of additional cath lab salaries in case the EMTs diagnosis was incorrect, ED staff were encouraged to rely on the EMT’s field diagnosis and initiate the call to the cath lab team as soon as the EMTs called in from the field. Door-to-cath-labtime plummeted to 10 minutes!
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
135
Effecting and sustaining significant change is hard work. The need for change creates strong emotions in people, particularly when you are the one who is expected to change. People experience roller-coaster emotions of fear, loss, and denial, before reaching acceptance. This is all normal. A critical function of the Black Belt is to manage people’s feelings and emotions so improvements occur and are sustained. The success of this project had a positive impact across the hospital. All departments and staff learned to value ‘their’ ED as the ‘front door to their hospital.’ At the end of the first year, with an ED diversion time of near zero, the Emergency Department treated over 37,000 patients and realized a gross margin of $18 million. This was a 38.26% improvement over the previous year.
“Beating Heart” Bypass Grafts Though altruism and evidence influence medical treatments, economic pressure drives improvement. Multi-million dollar savings created by “beating heart” or “off-pump” coronary artery bypass outcomes are a case in point. Historically speaking, medical “Six Sigma” style breakthroughs have astonished the world. Near zero death rates related to surgical anesthesia and the polio vaccine’s safety record are but two near perfect success examples. Sir Austin Bradford Hill’s 1951 sentiments sound as fresh as a 21st Century General Electric Six Sigma news release: “In treating patients with unproved remedies we are, whether we like it or not, experimenting on human beings, and a good experiment well reported may be more ethical and entail less shirking of duty than a poor one.” (Br. Med 2:1088-90, 1951, Hill, 1952)
The ability to consistently replicate experimental outcomes with a high degree of confidence is of paramount importance to everyone in the health care system. Again, off-pump surgical technique provides an ideal compass setting that points the way to breakthroughs. Since health care Six Sigma ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
136 Case Studies breakthroughs simultaneously improve the quality of patient outcomes and profitability, “off-pump” coronary artery bypass grafts (CABG) projects are substantive. Limited financial resources fostered the early 1980’s development of “beating heart” CABG surgeries in Argentina. Compelling statistical evidence is leading to the reluctant acceptance of this surgical technique in competitive, 2003, USA health care markets. Patient demand for this lower cost, higher quality procedure has forced, and is forcing, surgeons to master a challenging, higher standard of care. Again, the classic evidence-based decision cycle, Define, Measure, Analyze, Improve, and Control, provides a convenient way to summarize this story. Define: For over 40 years, the use of cardiopulmonary bypass (CPB) pumps defined coronary artery bypass grafting (CABG) procedures. Good outcomes and the relative ease of working on an arrested heart led most cardiac surgeons to favor the use of CPB.4 Statistically significant blood utilization and neurological side effects associated with on-pump surgeries were considered to be acceptable—necessary— collateral damage related to the bypass operation. Though statistical evidence suggested off-pump operations were safe and advantageous for select patients, the prevailing beliefs of cardiac surgery sustain physician commitment to the on-pump surgical technique. It has taken a decade for surgical practice patterns to emerge that reflect sentiments expressed by researchers in 1992. “Further research should be directed to which subgroups can be operated on to advantage off-pump and which, if any, groups of patients should be confined to on-bypass operations.”5 Patterns and pattern recognition are key elements in the identification of breakthrough improvements. Database and computing systems accelerate both when they are included in an open system feedback loop. Figure 8 illustrates the classic, standard Six Sigma closed feedback system. The closed feedback loop idea is a serious theoretical error that can be traced to the 1990’s pseudoscience of “systems thinking.”6 Closed feedback loops create entropy. Closed feedback systems are driven by opaque, spreadsheet analyses and story ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
���������� �������� ��������
Figure 8 The recommended Six Sigma closed loop feedback system is contrary to evidence-based decisions. Closed loops create entropy.
�������������������� ������������� ������������������ ������������������
������������������������� ������������������
137
������������������� ��������� �����������
���������������� ������������������������� ���������������� �������������
������������������������� �����������
telling where 83 percent of the information contained in raw data are suppressed. Evidence-based decisions must have open feedback systems. Open feedback systems depend upon the continuous entry and flow of objective evidence into judgments. Obviously doctors, nurses, allied health professionals, and administrative leaders are the Six Sigma “executive champions and Master Black Belt” experts who initiate breakthrough improvement actions. In addition to quantitative, open loop feedback measures, qualitative impressions frequently expose opportunities. In the off-pump/on-pump dialogue, one qualitative signal is the long running practice of opinionated debates between surgeons. Without a commitment to evidence-based decisions, these discussions are generally sustained without referencing or generating statistical evidence for analysis. Measure and Analyze: Though surgical practice data are often collected by hand, increasingly this data is automatically entered into databases. Integrated statistical software packages now make it possible to analyze measurement data almost as quickly as they are recorded. Figure 9 shows columns and rows of data for a single cardiac surgeon who, after a number of his patients canceled their scheduled on-pump surgeries in order to have them performed off pump by a different
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
138 Case Studies surgeon at a competing hospital, decided to master the offpump surgical technique.
Figure 9 A data matrix arrays historical data so a vector analysis can be used to identify profit signals. This array documents charges, lengths of stay (LOS) for patients and type of CABG surgery either off-pump or on.
The computerized analysis of length of stay data in Figure 10 reflects findings that are similar to the 443 peer-reviewed articles published on the on-pump/off-pump subject since 1992. The peer-reviewed literature on this topic is consistent to a remarkable degree. Patients who undergo off-pump CABG surgeries experience dramatically lower lengths of stay.
���� ������������������ ������������������ �� �� �� � ������������
Figure 10 The strong profit signal between the lengths of stay for on pump and off-pump surgeries are eye catching with a statistically accurate “flying saucer” graph.
� � � � �
�
On the hyperspace vector analysis applied to a data matrix thrill ride, the difference between data sets is significant at the 95% confidence level if the saucers can fly past each other without crashing,
� � �� �
�� ���� �
���� �������� ������ �� ������������ ������ ���� � ����� ��������
�� ������ ������ ������
������ �������� ���������� ���������� ����������
����������� �������� ��������
����� ��������� �������� ����� �� � ��
������ ����� �����
���� �������� ��������
��������� ������� �������
��� ������ ��������� ������� ������� ������� �������
������������ �������������� ��������� ���������������
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
������� ��������
�������� ��������
Case Studies
139
Figure 11 The Profit Signal in patient Lengths of Stay (LOS) were related to off-pump CABG surgeries. These improvements were dramatic.
Literature searches used to cross check statistical inferences are a value added service physicians appreciate. A quality control chart, Figure 11, provides another view of the impact off-pump surgical technique brings to the quality of patient care. As the average length of stay shrinks, so does variation around the mean. Since 1931, this pattern has symbolized the classic breakthrough pattern of an evidence-based decision. These breakthroughs now lead to near perfect performances known as Six Sigma. The surgeon’s database was stratified to facilitate a threedimensional statistical analysis to consider the effect a number of other factors might have had on length of stay outcomes. Factors we considered were diagnostic (ICD) code variations, co-morbidities, age, gender, and race. An example is shown in Figure 12’s cube plot. The Cartesian coordinate system’s cube is an ideal graphic for presenting multidimensional statistical evidence. The numbers contained in the rectangular boxes at the cube’s corners are average values. Even a novice can interpret the results at a glance. In Figure 12, all four of the shortest lengths of stay related to CABG are located on the cube’s left plane. The shortest average length of stay, 1.875, was a result of an off-pump surgery with a male patient with ICD code 36.11. All of the ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
140 Case Studies longer lengths of stay are located on the cube’s right plane. The longest average length of stay, 6.875, was the effect of on pump surgeries for men with ICD code 36.12. Though three factors are presented simultaneously, the only statistically significant factor related to a lower length of stay was a surgery performed off-pump. This case study did not include a Pareto chart analysis summary for two reasons. First, the data matrix software used to produce the evidence in this case did not have that feature. In addition, the organization had progressed beyond the need to present data in a simplistic way. Decision makers wanted to look at advanced, Six Sigma style, evidence charts. Figure 12 Profit signals compare the surgeon against herself. We can say with a 95 percent level of confidence that when off-pump surgeries are used on appropriate patients, they produce medically superior outcomes and lower lengths of stay.
Improve: Sixteen years of experience in promoting breakthrough improvements in health care quality and productivity teach an important lesson. Before changes occur in physician or hospital practice, benefits must be translated into a compelling financial story. Though this reality can be disheartening for caregivers who put patient safety first, leaders must prioritize cost accounting if they expect to see system wide improvements take place. Simulation modeling using spreadsheets is a relatively easy data matrix tool to master. The psychological impact of seeing 1,000 or more iterations of multivariate spreadsheet practice scenarios is significant. More often than not, spreadsheet simulations are persuasive. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
141
Figure 13 shows the profit signal’s probable financial impact for one surgeon. The low end of the forecast’s distribution suggests that by mastering the off-pump procedure for the majority of her patients, an additional 448K in revenue would be generated. On the high end of the distribution, this change could produce as much as $1.45 million.
Actual results fell near the center of the prediction parameters. Savings were achieved through lower nursing care costs and overhead. Off-pump patients avoided adverse side effects while the hospital enjoyed improved profitability. These results are classic hallmarks of a Six Sigma style breakthrough.
Figure 13 Spreadsheet add-ins for modeling and simulation are a compelling, persuasive use of the data matrix and profit signal analysis. Revenue gains for offpump surgeries are predicted to range from a net gain of 448K to $1.4 million.
Control: The final step in the Six Sigma DMAIC (Define, Measure, Analyze, Improve and Control) process is to standardize breakthroughs and hold the gains. Discipline is as important to success here as it is with each of the other steps. Leadership and culture determine the rate of adoption for breakthroughs in productivity and quality. When the medical staff and other senior leaders are disciplined, and when they role model the use of science, statistical analysis, and systematic experimentation, breakthrough improvements occur. Six Sigma culture evolves along with the breakthroughs. The degree of success in every Six Sigma breakthrough is directly related to the level of commitment that is demonstrated by senior leadership.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
142 Case Studies
The Daily Grind Don worked in the belt grinding department. Day after day, he and his co-workers removed “gate stubs” from metal castings to prepare them for final processing and shipping. The grinders were paid a handsome hourly rate. The other major expense for the area was the cost of belts. They went through a lot of belts on a typical shift. Define: If you try to use a belt beyond a certain point, your efficiency in removing metal goes way down. The supplier representative had given the area manager a rule to use for deciding when the grinders should throw a belt away and put on a new one. The rule was called “50% used up”. There were examples of belts that had been “50% used up” hanging on the walls in the grinding area. The purpose of the rule was to minimize the total expense of the operation. Don thought the rule was wrong. He thought it caused them to discard the belts too soon. He had a hypothesis that using the belts a little longer would reduce the belt expense with no loss of grinding efficiency. He also suspected that the supplier wanted to sell more belts. We had no way to evaluate this, so we let it go. Don had come up with a new rule called “75% used up”. He proposed doing a designed experiment to determine whether or not the new rule was more cost effective than the old rule. We met with Don, the area manager and the supplier rep to discuss the project. To our surprise, the supplier rep was vehemently opposed to the project. He said the “50%” rule was based on extensive experimentation and testing at his company’s R&D laboratory. He said we were wasting time trying to “reinvent the wheel”.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
143
Don argued that laboratory tests may not be good predictors of shop-floor performance. We thought he had a point. We were also starting to see why he was suspicious of the supplier. The area manager also thought Don had a good point. He gave the go-ahead for the project. He allowed Don one full day to complete the experiment. Measure: Don figured he could get 16 castings done in one day. When the other grinders heard about the experiment, they suggested other things that could be tested at the same time. The contact wheels currently used on the grinding tools had a low land-to-groove ratio (LGR). One of the grinders wanted to try a wheel with a higher LGR. Another wanted to try a contact wheel made out of hard rubber instead of metal. A third reminded Don that belts of at least two different grit sizes were routinely used. He felt that both grits should be represented in the experiment to get realistic results.
Table 2 The data matrix for Don’s grinding experiment. There were four factors at two levels each. The response variable was the total cost for each casting divided by the amount of metal removed. The total cost was calculated as labor cost plus belt cost.
Table 2
contains the data matrix for the grinding experiment as it was eventually run.
Analyze: An eyeball analysis applied to Table 2 suggested that Don was on to something with his “75% used up”. It also suggested that high land-to-groove (LGR) is better than low, and rubber wheels are worse than metal ones. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
144 Case Studies
Figure 14 Pareto Plot ranking the factors and interactions in the belt grinding experiment by the strength of their profit signals.
But let us not be hasty. Figure 14 shows the Pareto Plot ranking the factors and their interactions by the strength of their profit signals. The strongest signal was the comparison of steel to rubber contact wheels (MATL). This signal told us that rubber was not a good idea. The next-largest signal was the comparison of the 50% rule to the 75% rule (USAGE). It predicted significant savings in line with Don’s idea. The third-largest signal was the comparison of a low to high land-to-groove ratio for the contact wheel (LGR). The next two signals involved interactive effects. The message here was that the actual cost reductions from implementing the USAGE and LGR results would different for the two grit sizes. Improve: Don’s experiment produced two recommendations: 1. Use his 75% rule instead of the supplier’s 50% rule. 2. Use contact wheels with the higher land-to-groove ratio. The combined impact of these two changes was a predicted cost reduction of $2.75 per unit of metal removed. This multiplied out to about $900,000 in annual savings. Don’s recommendations were quickly implemented throughout the grinding department. The actual savings came in a little under the prediction, but everyone was happy. Not bad for a one-day project. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
145
Control: Some degree of cost reduction was achieved by all the grinders, but it did not apply uniformly. There was still a lot of variability in grinder performance. Attacking this variation was the obvious next step. We don’t know if our recommendation was ever implemented. “Die Tuning” for Vinyl Extrusion A vinyl extrusion operation receives a “die package” (blueprint) from a customer for a new “profile” (part). The extruder then designs and machines the “die” (tooling) for extruding the profile. The extruder bears the development cost in exchange for a life-of-contract “sole supplier” status. The process of machining, testing, and revising dies is called die tuning. Each “revision” involves re-machining the die. The average cost per revision is about $2000. The number of revisions required to get a new die ready for production varies unpredictably from 0 to as high as 30. As a result, the total cost varies unpredictably from $2000 (no revisions needed) to something like $50,000 (lots of revisions needed). An extruder can easily spend $1.5 to $5.8 million or more each year on die tuning. Reducing the dramatic variation in the number of revisions was identified as a project with potentially huge financial benefits. Define: We started with a “Kaizen-blitz”, a very fast and focused review of the die tuning process. Once the initial machining of a die is completed, a tester runs that die on one of several extrusion lines reserved for testing new die. Once the production line stabilizes, the tester does visual inspections and measures the control dimensions with a caliper. The inspection results and the dimensions are taken to a revision programmer who determines whether a revision is needed. If it is, the revision programmer sends the die back to the machine shop with a revision sheet describing the needed changes. The tester is also supposed to determine the best run conditions for the new die. Potentially these factors could include: ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
146 Case Studies • Line speed • Die-to-calibrator distance • Calibrator vacuum • Screw Revolutions Per Minute (RPM) • Screw oil temperature • Barrel zone temperatures • Die zone temperatures • Melt temperature • Melt pressure • Weight Testers are under time constraints. They adjust some of these variables by trial and error to get the dimensions closer to nominal and improve the cosmetic quality. The variables most commonly adjusted are line speed, die-to-calibrator distance and weight. The other variables tend to remain at “baseline run conditions” assigned before the die is machined. Our findings were as follows: 1. Die revisions were based on single measurements taken by a hand-held caliper on plastic parts. In all industries the repeatability of such measurements is notoriously bad. 2. The trial-and-error method has virtually no chance of finding good run conditions. 3. Letting testers choose which variables to adjust may have long-term economic consequences. Examples are lowering the line speed or increasing the weight. Item 1 looked like a possible “smoking gun” for the problem of too many revisions. We proposed that small series of designed experiments be made a routine part of die tuning. This would require more time for each revision cycle. But this process held the promise of dramatically reducing the number of revisions. The basic idea was this: before we cut metal again, let’s see if we can “process our way out” of some of the dimensional or cosmetic problems.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
147
We felt the Design of Experiments (DOE) approach could address all three. Some of the team members wondered how it could help with Item 1. The answer was that the results of a DOE are always based on weighted averages rather than individual measurements. This automatically improves the reliability of the data used to determine revisions. Measure: For the initial experiment, the team decided to observe four continuous factors: line speed, die-to-calibrator distance, calibrator vacuum and weight. We used statistical software to generate a data matrix similar to one shown in the first six columns of Table 6. The matrix in Table 6 is the as-run version with the weights and calibrator vacuums actually obtained in place of the nominal values in the original matrix. The levels of the four factors are coded to protect proprietary information. The die in this case had a dual orifice. This means that two profiles are extruded at the same time. Results for the two profiles are distinguished in the matrix as Sides 1 and 2. The responses included 13 control dimensions and a 15 distortion rating, where higher is better. The control dimension data are expressed as deviations from nominal in thousandths of an inch. Table 6 The data matrix that fills the next page is from the die tuning experiment. There were four continuous factors at three levels each. The response variables included 13 control dimensions and a 1-5 distortion rating, where higher is better. Remember, because this table is a data matrix, each column is a single entity or vector. A correct analysis breaks up the variation vector in the cornerstone of evidence into Noise and Profit Signals.
Analyze: A matrix of distribution curves was the result of jointly optimizing all 14 response variables. The statistical software performed this optimization in just a few seconds. Please accept our apologies for the fact that the complexity of this statistical graph exceeds the boundaries of this introductory book. The quick story follows. Improve: The implications were staggering. By doubling the line speed and reducing material costs by 50 percent the production line produced perfect quality product after just one revision and some very minor additional die tuning. Additional key findings were as follows: • We were able to run a four-factor die tuning experiment in one day.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
148 Case Studies
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Case Studies
149
• We generated a wealth of information on how each factor affects each response variable. Some results confirmed prior beliefs, others contradicted prior beliefs. • We showed that using weight and line speed as adjustment factors in die testing lead to unnecessarily high weights and low line speeds. This locks in unnecessary costs for the life of a contract. It may also contribute to problems with quality, which in turn lead to a larger numbers of revisions. A conservative estimate of the annual cost reduction from extending this method to all new die was $1.2 million, half of the current annual budget for die tuning. Control: The process of changing the way die tuning is done is underway. Similar experiments have been run on other new die with similar results. In one case a die was saved in the nick of time from going back for an incorrect revision that would have spawned further revisions to repair the damage. Much has been accomplished. More is expected.
Endnotes Cartmill, Matt. “Oppressed by Evolution”. Discovery Magazine, March, 1998, pages 78-83 as reported by Richard Dawkins on page 20 in his book Unweaving the Rainbow. 1
2
http://gi.grolier.com/presidents/aae/side/knownot.html
Cheryl Payseno, an RN, former hospital administrator and certified Six Sigma black belt wrote this case study for us. Cheryl led the charge for the use of Designed Experiments in health care in 1995 with Daniel Sloan. Results from those early innovations were published by the American Society for Quality’s Quality Press. 3
Pfister, Albert J., Zaki, M. Salah, et al. “Coronary Artery Bypass without Cardiopulmonary Bypass.” Ann of Thorac Surg 1992; 54:1085-92. 4
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
150 Case Studies
Pfister, Albert J., Zaki, M. Salah, et al. “Coronary Artery Bypass without Cardiopulmonary Bypass.”Ann of Thorac Surg 1992; 54:1085-92. 5
Senge, Peter M. The Fifth Discipline, The Art and Practice of The Learning Organization. New York. Doubleday Currency. 1990. 6
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 5 Using Profit Signals
P
rofit signals show you the money. Profit signal vectors literally and figuratively show you what works best in any business, financial, health care, manufacturing or service process. This chapter explains how vector analysis applied to a data matrix showcases the information contained in raw data. Once the tools have done their job, the graphic presentation of evidence paves the way to breakthroughs in quality, productivity and profitability. Profit signals are like televisions, radios, cars, telephones and the Internet. They attract attention. People want to play with them. They want to use them. This natural occurrence unsettles to old-school managers. Some react like the mythical John Henry: “Before that steam drill shall beat me down, I’ll die with a hammer in my hand.” Vector analysis applied to a data matrix is the steam engine that humbles them. The spreadsheet is the first and only computing program many business people learn to use. When all you have a sledgehammer, everything looks like a spike. Hammering out spreadsheet revisions keeps employees occupied. These people are occupied reworking Proformas, business plans, and trying to explain why actual monthly financial results do not fall exactly on the predicted straight line of a one-dimensional “variance” analysis. Though they are busy, they may not necessarily be productive.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
152 Using Profit Signals Fortunately, the cornerstone of evidence is appealing. By constructing a cornerstone-of-evidence tetrahedron using bamboo skewers as vectors and spheres of Sculpey Clay as points-in-hyperspace connectors, people can weigh evidence in their own hands. The look and feel of an Analysis of Variance tetrahedron in one hand and a single stick in the other, convert wouldbe 21st Century Luddites into evidence-based decision champions. Physical models win hearts. The following are a few of the many reasons why so many former skeptics embrace the use of profit signals to make more money. 1. With profit signals, you have only one formula to remember. 2. With profit signals, you don’t have to solve equations. 3. With profit signals, you can produce 10 times the work in a fraction of the time now spent doing arithmetic with a spreadsheet. 4. Profit signal pictures are aesthetically pleasing. 5. Profit signals help you make more money with less work. Money, number 5, is THE big reason Six Sigma projects are so popular around the world.
A Better Way to Look At Numbers Think back to your Five-Minute PhD. In a data matrix, each number is an integral part of an entity called a vector. Each column of numbers is its own vector. Each column is a field or variable with a precise operational definition. Rigorous inductive and deductive reasoning, which dates back to Aristotle, is built into statistical software designed specifically for the data matrix structure. In other words, a data matrix channels the intelligence and logic of the best minds our human species have produced. Each number is framed in the geometric context of a profit signal vector. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
153
Vectors show you the money. Measurements presented in the rows and columns of a spreadsheet convey no sense of unity. There is no sense of purpose. Each number is an orphan locked in its own cell. Logic takes a back seat to manipulation. Commonsense relationships between numbers are ignored. A ghost named Zero inhabits empty cells. Arithmetic is the two-stroke engine running Abacus Prison. There are no vectors, no arrows, pointing to the money. Vectors have physical properties. These properties can be measured and displayed in three dimensions. We strongly urge you to actually build a Sculpey-Clay/bamboo skewer model whenever the dimensions of a vector analysis are revealed to you in one of our examples. It costs about one buck for the whole kit. Corrugated Copters C. B. Rogers created the helicopter analogy while working at Digital Equipment in Marlboro, Massachusetts. Professor George E. P. Box introduced us to it at the University of Wisconsin, Madison in May 1995. Dr. Box, a Fellow of the Royal Society and the American Academy of Arts and Sciences, was the Fisher Professor of Statistics. Box was also a riveting teacher who taught us that an analysis of variance was so simple, “You could tell the answer just by looking at the numbers on a cube.” He and his colleagues used the helicopter in Figure 1 to illustrate. Please take a moment to build one now so you can follow along with our data explanation. First, tear a piece of 8.5 inch by 11 inch paper in half, long ways. If you have a pair of scissors and quality paper use them. These tools will make the construction process more satisfying. The results will be more rewarding. If you are in a hurry, tearing paper works fine. Next, cut or tear the top section to form the “blades.” Finally, follow the folds at the bottom to form the helicopter’s
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
154 Using Profit Signals
������
���
Figure 1 This inexpensive product is an analogy that works well for teaching data matrix and vector analysis principles to people in all industries.
������
����
���� ���� �����
���� ���� �����
������
fuselage, or body. You may tape the body to give it some rigidity if you like. Hold the finished product with the blades perpendicular and away from the body at shoulder height. Let it drop. Like seeds from a maple tree, the blades will catch air while the aircraft spins to the ground. This is fun to do and fun to watch. Now, time the flight using the black, blue, purple, or pink plastic digital chronometer you wear on your wrist. For this game, each helicopter costs $9 million to build. For each second of additional flight time, customers are willing to pay an additional $1 million in price. Longer flight times are worth quite a bit more money than shorter flight times. Corrugated Copters learned a big lesson when their company was founded in 1996. Eliminate all costs associated with take offs and you can really make money.1 Their original corporate slogan was, “Drive down costs!” Their current, more enlightened view is wordier: “The best way is the most profitable way.” This saying has become a ritual chant that opens all management meetings.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
155
Take a moment now to draw a Six Sigma Supply, Input, Process, Output, and Customer (SIPOC) flow diagram. You will learn Corrugated Copters is a behemoth that demands global logistical support. The paper began as a seed that was planted on a tree farm in the Pacific Northwestern United States in 1948. The quality and cost of that tree affects the quality and cost of your building materials. One company cuts down the tree, the Supply. Another ships it as Input to the pulp mill. The pulp mill Process creates the paper. The packaged Output is sold to its wholesale Customer. Corrugated Copters is the retail customer who buys it from the wholesale customer. You and your products are parts of a system. The company’s measuring device is a five-mode wristwatch with alarms. It breaks hours into hundredths of a second. It used to be silicon, some oil, and ore. Since time is money, and money is time, the calibration of this instrument is exceptionally important. Accuracy matters. One of your employees has created a Lean flow diagram to show the entire value stream for your watch. The most efficient routes for delivering these devices to your engineers are annotated with dollars, times, and inventory turns. The store that supplies this watch keeps a supply of them on hand just in case you need a new one in a hurry. Just-In-Time has eliminated almost all of Corrugated Copter’s inventory costs. The pen or pencil you used to record your measurements also has an informative SIPOC diagram archived for reference in the event another new Six Sigma breakthrough is needed. Last and certainly not least, the brains behind Corrugated Copter’s success have been, to varying degrees, educated. Not everyone is cut out to be a helicopter pilot. Not everyone could hope to be a timer. It almost goes without saying that collecting data is a big job. The analysis of that data is yet another specialized task that has its own job classification. Complexity surrounds Corrugated Copters. The market is filled with uncertainty and risk.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
156 Using Profit Signals
Testing the Current Way of Doing Things Avona Sextant, a Corrugated Copter senior executive, has a Five-Minute PhD. Avona is often called upon to facilitate meetings. When she is not in the room, Copter teams seem to argue amongst themselves. When Avona joins their dialogue, teams just naturally converge on answers that lead to a consensus and a “path forward.” More than anyone else in the company, Avona is committed to evidence-based decisions. Some suspect her peculiar predisposition is a genetic disorder. In any case, Avona will listen only to stories that have evidence in their punch lines. Bets are routinely placed, money is won and lost, over when and how many times she will say the word “evidence” in a meeting. Some think Avona is goofy. Others think she is crazy like a fox. Though there is resistance to her methods, no one argues with her fundamental point of view, “The best way is the most profitable way.” On this they are in full agreement. The problem is how to determine which way is best. On this there is a considerable amount of debate. Some employees have heard quite enough of her New Management Equation speech. They suspect that Avona’s little formula for calculating Chance variation only works with simple numbers like 3, 4, and 5. They also know this Six Sigma stuff is a passing fad. They are going to wait it out and hope for the best. During a recent productivity breakthrough, the midmanagement team of Tom, Dick and Mary produced a double-digit flight time! Just yesterday they booked a recordbreaking 10 seconds. They are proud of themselves and bragging when Avona walks in. “Ten seconds. What an awesome and terrific flight time!” Avona cheers. “That’s worth ten million dollars in gross revenue. That would be a profit of one million dollars.” She adds, “I can’t wait to see the rest of your evidence.” “Evidence?” asks the team. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
157
“This is so exciting. You must have flown this machine more than once. I just want to see your other measurements. If we average more than 9 seconds when we launch the product line I will be euphoric. Otherwise we won’t make any money in the long run.” The team showed her all their data: 9 seconds, 8.9 seconds and 10 seconds. “Oh, I see you’re using a spreadsheet,” said Avona. “I used to use one of those. Plus I also had an abacus for my backup system.” Avona’s aunt in Hong Kong taught her how to use an abacus when she was a little girl. The abacus was the world’s first computing system.2 Because her abacus was a Chinese rather than a Japanese machine, she learned long ago to translate binary numbers into regular old numbers and back again with the flick of her right index finger. When Avona first saw vector analysis applied to a data matrix, she knew the time had finally come to retire her abacus and her spreadsheet too. Avona was still waiting for her statistical software purchase order to be approved. In the meantime, she had programmed a worksheet with vector-analysis formulas built into the cells. With her templates, people didn’t have to type in any formulas. She input the three data points. Her Excel spreadsheet immediately produced the vector analysis displayed in Table 1. “Our objective of 9 seconds is a fixed number rather than a measurement. So the first step is to subtract it from the raw data,” she explained. “This gives us the Differences vector.” She drew the picture in Figure 2 to illustrate the vector analysis of the difference data in Table 1. Mary observed, “Is that what accounting calls a variance?” “Good call Mary. Yes it is.” “I don’t think that department ever thought of a column of numbers as a vector,” said Dick. “They do now. The new Six Sigma Black Belts in Accounting are changing history!” said Avona. “I have no idea how ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
158 Using Profit Signals
�������� ������
������ ������
�
�
���
�
�
������������� �������� ����������� ������������ ������ ������ � �
�
�� ������������������
��������������� �������������������������� �������� ����������������� ���������������������
����
��� �
���
�
���� �
���
���� ���
�
�
�
�
�
����
�
����
�
����
����
����
� ������ ������������������������� ����������������� � ������ ��������������������������������������������������������������� ������������������ ��������������������������������
����� ������
���� ���� ����� ����
Table 1 Vector analysis for testing the current helicopter design against the performance objective, 9 seconds. The raw data are flight times in seconds. The profit signal coincides with the average difference from the 9 second objective. It has one degree of freedom because it is determined by a single number—its average—0.3. Noise is calculated by subtracting the Profit Signal value from the respective value in the difference vector. This arithmetic is a Law of the Universe. See Figure 2.
they are going to solve the 1,000 year old waste and rework problems related to the 14th Century’s double entry bookkeeping system. Our Black Belt CPA Peruzzi told me the tip off for her was the word “double”. Get it? ‘Double entry? Rework entry?’ Well, Peruzzi is convinced the entire double entry ‘bookkeeping system’ is nothing more than a massive hidden factory loop. With a properly designed data matrix, the second entry is needless rework. “Our Black Belt CPAs are arraying entries into a data matrix. Some have already doubled their personal productivity. It is a marvel what Six Sigma education and training can do, even for a Merchant of Venice.” “See how the squared length of the difference vector, 1.01, is equal to the sum of the squared lengths 0.27 and 0.74?” she shined (Table 1). “That’s how the New Management Equation works. Everything always adds up. It’s wonderful.” “Oh come on Avona,” chided Mary. “Having those numbers add up is no big deal.” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
159
Figure 2 This is the picture of the key vectors in Table 1.
“Oh it is. It is!” said Avona. “Did you notice that when you multiply a minus times a minus, the sign becomes a plus? And look how confusing that –0.4 in the Noise vector column is. I always have a hard time remembering that a negative number like –0.1, minus a positive number like 0.3 turns out to be a bigger negative number!” “The last time I saw this stuff was when I had to learn to use a slide rule in Mrs. Beamer’s algebra period.” complained Tom. Though the team was tired of Avona’s boundless enthusiasm, with a push of the square root button on their calculators they could see that the sample standard deviation—the square root of the 0.37 Variance—was about 0.6 seconds. Their average flight time was about 9.3. Even if these numbers perfectly described what would happen in long-run production, future times would vary. Laws of the Universe strike again. The best performance the team could expect would be that future flight times were unlikely to fall below a lower “threesigma limit” of about 7.5 seconds [7.5 = 9.3 – (3 × 0.6)]. To make matters worse, Avona starting talking about evidence. “The null hypothesis here is that our future average flight time will be 9 seconds. We want to disprove this ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
160 Using Profit Signals hypothesis. We want the average flight time to be higher. We want to make money. “The data will show this is true if and only if the p-value is small enough. By international standards, a p-value less than 0.05 gives ‘clear and convincing’ evidence against the null hypothesis (Table 2). A p-value less than 0.15 gives a ‘preponderance of evidence’ against the null hypothesis. Our p-value is 0.428, not even close to the lowest standard. This means there is no evidence at all that the average flight time is significantly different from 9 seconds. These differences are Table 2 Standards of evidence probably due to Chance. It is a Law of the Universe.” table.
To illustrate the implications of her conclusion, Avona drew the picture in Figure 3. “There is no signal here, just a lot of noise in our system. Depending on variations in the weather, wind, pilot, paper, timing device, and construction, the time could vary up to 10.8 and all the way down to 7.2 seconds. “Also, if the mean is exactly 9 seconds, our average gross revenue will be exactly equal to our cost, $9 million. We will be making money on half, and losing money on the other half. This is not good. This means our long-run profit will be zero, 0. The best way is the most profitable way.” The room was quiet.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
Figure 3 The Normal distribution of flight times if the mean is 9 seconds and the standard deviation is 0.6 seconds.
161
0.6
7.2
7.8
8.4
9
9.6
10.2
10.8
Everyone had taken a liking to Avona’s signal/noise analogy months ago. They all agreed with her interpretation of their data. As usual, the team ended their argument with an agreement. Ten was an exciting, encouraging number. But, they needed to know more before they could launch the new product. There were two possibilities: (1) The problem might just be the small sample size of 3. They could do more tests of the current design to strengthen the signal and reduce the noise. This would let them determine the average flight time with greater accuracy. (2) They might need to go back to the drawing board and find a way to further increase the flight time. For 2500 years the right triangle has shown us the route to profitability. Ancient Greek mariners used the sextant to navigate the Mediterranean Sea’s lucrative markets. In his little book, Posterior Analytics, Aristotle equated the right triangle with truth.3 Applying vector analysis to a data matrix on a regular basis is a good way for today’s seekers of truth to learn about Aristotle’s principles. Six Sigma experts know that the New Management Equation discovered by an old Greek named Pythagoras 2500 years ago is worth billions of dollars today. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
162 Using Profit Signals In analysis, as in telecommunications, customers want a strong signal. Just like Avona, communications engineers from Marconi in 1901 to Nokia in 2003 have appreciated the value of a high signal-to-noise ratio. The data matrices and vector analyses employed by engineers differ only superficially from the matrix and vectors you used to earn your FiveMinute PhD. Overcoming Obstacles “Science phobia is contagious,” wrote Carl Sagan, an astronomer and television celebrity, just prior to his death in 1996. 4 “Some people consider science arrogant—especially when it purports to contradict beliefs of long standing or when it introduces bizarre concepts that seem contradictory to common sense. Like an earthquake that rattles our faith in the very ground we’re standing on, challenging our accustomed beliefs, shaking the doctrines we have grown to rely upon can be profoundly disturbing.”5 The transparent analysis principles in the cornerstone of evidence shake the foundations of business decisions. Executives may find the data matrix and vector analysis distressing. Until they get the hang of using these tools, both concepts tend to terrify cost-accounting analysts. Once on board, these executives and analysts become vital assets for breakthrough project teams. Math phobia is another, even more daunting obstacle on the high road to evidence-based decisions. In our consulting practices over the past 20 years, we have found that some of the people who fear math most are Accountants, Controllers, Financial Analysts, Chief Financial Officers, Chief Operating Officers, and Chief Executive Officers. Many corporate officers “did not do well in high school algebra.” Take Daniel Sloan for instance: “Numerical dyslexia, reversing numbers instead of letters, has plagued me since I memorized my times tables in Mrs. Peiffer’s fourth grade classroom. I can no more do math in my head than I can read the letters at the bottom of an eye chart without my glasses. I must wear glasses to see. I must use a computer to do math. Confronting math phobia was the most painful, anxiety©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
163
provoking, downright embarrassing, and humiliating career step I ever took. “Overcoming my math phobia was a more strenuous challenge than all of my five years as a Vice President of Marketing, publishing five peer-reviewed statistical textbooks, my stint as a Senior Vice President in a publicly traded, $500 million corporation, and founding and running my own business for 14 years. It has been as rewarding as it has been difficult. One of the best things success has given me is the opportunity to help other business leaders like me take that frightening first step forward.” Larger than science and math phobias combined, is the fear of losing one’s job. Experience shows, money motivates. It can and does persuade executives and line workers alike to face and overcome both these phobias. Six Sigma is a cultural business force that compels people to step up to a difficult task. Privacy is exceptionally important to adult learning. Computerized, personal, learning programs deliver privacy. They are available for adults who suffer from science and math phobias. Math Blaster, Alge-Blaster, Pro-One’s CD-ROM multi-media course Mathematics, and many other programs are great ways to re-learn the principles of addition, subtraction, multiplication, division and the order of operations. They are fun. Private tutors and educational consultants are other options that work well. The best news for executives and workers alike is that cheap, reliable, and very user-friendly software makes vector analysis as easy to learn as sending an E-mail. Comparing Two Ways of Doing Things “Hey Avona!” shouted Tom. “We think we have some evidence you are going to like. Just look at this stack of numbers.” Avona’s eyes opened wide. “Way to go. It looks like there might be a genuine difference between the two different helicopter designs.” Still not having her statistical program, she entered the data into one of her spreadsheet templates and showed them the vector analysis in Table 3. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
164 Using Profit Signals “I can’t believe it took me an hour to program this worksheet template so it will act like a data matrix,” Avona complained. “What a waste of time. I sure hope the purchase order for my statistical software gets approved soon.” Avona loved evidence, but patience was not her long suit. Table 3 Vector analysis for comparing two helicopter designs. The raw data are flight times minus the objective of 9 seconds. The profit signal consists of the average variation for each design. The average variation for white helicopters is –0.2 seconds of flight time. The average variation for pink helicopters is 0.2 seconds of flight time. The Profit Signal Vector has one degree of freedom because a single number, 0.2, determines it. When the numbers in this column are squared, the minus sign disappears. The squared lengths of all the vectors are connected by their part in the New Management Equation (NME).
“So Table 3 is where your ‘cornerstone of evidence’ comes from?” asked Mary. “Right. Just look at my models (Figure 4). The labeled edges correspond to the vectors in Table 2. We can make a model of your data and our new Analysis of Variance using some bamboo skewers and Sculpey Clay. I just happen to have a supply in my desk drawer.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
165
Figure 4 The cornerstone of evidence represents any vector analysis. A Polydron regular tetrahedron model is next to a cornerstone of evidence. Differences in raw data change the dimensions. It is a Law of the Universe.
Avona played with all sorts of modeling toys. Her office was filled with them. She told people they were symbolic. She would go on and on to anyone who would listen about some artist named Alexander Calder. “Let’s use my $1 handheld calculator to help us cut the bamboo skewers to length. We will use inches as the units. The length of the raw data vector is the square root of 8.38, which equals 2.89 inches. The length of the data average vector is the square root of 8.00, which equals 2.83. The length of the variation vector is the square root of 0.38, which equals 0.62 inches. The length of the profit signal vector is the square root of 0.32, which equals 0.57 inches. The length of the noise vector is the square root of 0.06, which equals 0.24 inches. The noise is so short it will be buried completely in the Sculpey Clay. The profit signal and noise vectors are the fine print in a vector analysis. “I sure wish we had our data matrix software. It is silly for us to use a hand calculator. “We haven’t talked about that last vector in the back of the tetrahedron. This is the vector of hypothetical predicted values. I didn’t include it in my spreadsheet templates because it isn’t important in the type of experiments we’ve been doing. It’s tremendously important in response surface experiments. That’s where we are optimizing over several continuous variables.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
166 Using Profit Signals
“Anyway, we get the prediction vector by adding together the profit signal and data average vectors.” (Table 4) It is always just a tad shorter than the raw data vector. In this case, the length of the prediction vector is the square root of 8.32, which equals 2.88 inches.”
Table 4 The vector of hypothetical predicted values is the sum of the profit signal and data average vectors. It gives the predicted average flight times for the two designs. It has two degrees freedom because it is determined by two numbers.
“It sure is colorful,” noted Dick, looking at their model. “Could we have hot pink Sculpey Clay points in space instead of green ones?” “We sure can. I think Sculpey Clay is a Six Sigma product,” Mary hypothesized. “Say. I just realized if you set one of those up on its end, it even looks like a radio tower sending out profit signals. I am going to need to bake mine in the break room toaster oven for a few minutes so the clay firms up and holds onto the vector skewers.” “Wow, look at that p-value in the table,” said Tom tearing his gaze away from Mary’s profit signals radio tower, “There really is a difference between the two designs.” “Absolutely right,” said Avona. “The null hypothesis is that there is no difference between the designs. A p-value less than 0.01 gives evidence beyond a reasonable doubt against ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
167
the null hypothesis. We are shredding that straw man like a mogul field at Mount Baker. “The spreadsheet actually has a formula called FDIST that calculates the p-value. It was named after Ronald Fisher. See, you just plug in the F ratio value, one degree of freedom for the profit signal and three degrees of freedom for the noise vector and voilà. There is hardly any noise in this data at all. It is almost all profit signal! “By subtracting the p-value 0.001 from the number 1, we get the number 0.999. This means we can be 99.9% confident that there is a difference between the pink and white designs. Even though the average difference is only 0.4 seconds, the vector analysis is sensitive enough to detect it. Plus, that’s another $400,000 in profit per helicopter. Phenomenal work team! “So, which design works best?” “What is most profitable is best!” Tom, Dick and Mary sang out. “Pink helicopters are best.” “It certainly looks that way,” said Avona. “But before we release the pink design to production, let’s do a confirmation experiment. And while we’re at it, let’s include the green design in the comparison. We don’t have much data on that. See you guys later.” “Gee whiz Mary,” said Tom after Avona had left. “Everyone can see pink helicopters are best. Why is Avona such a stickin-the-mud? And why does she keep saying ‘we’ when she really means us?” “Just be grateful she didn’t talk about evidence again.”
Comparing Three Ways of Doing Things “Wow! I think we are onto something with these pink helicopters,” said Dick. “We even checked them against green helicopters. They still came out best.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
168 Using Profit Signals “And what is best is most profitable,” said Avona. “Let’s plug your numbers into my spreadsheet template. I want to show it to Rotcev Sisylana, our new CEO from Uzbekistan. He’s gonna love this. Maybe he will get me two copies of my data matrix software. Shoot, they cost less than a thousand dollars. I wasted more than that last week dinking around with my spreadsheet templates.” Avona’s analysis is presented in Table 5. The null hypothesis is that all three designs will have the same average flight time. The p-value of 0.004 says there is evidence beyond a reasonable doubt that this is false. In other words, at least one of the designs is significantly different from another. Which one is best? From the profit signal, we can see that the pink design flies 0.325 seconds longer than the average flight time of 0.9 seconds. The white and green design flight times are 0.125 and 0.200 shorter than average. Once again, pink is best. Table 5 Vector analysis for comparing three helicopter designs. The raw data are flight times minus the objective of 9 seconds. The profit signal consists of the average variation for each design. It has two degrees of freedom because it is determined by two numbers, -0.125 and - 0.200 in this case. The third number, 0.325, is minus the sum of these two.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
169
After reviewing the results in Table 5 Dick observed, “This table looks just like all the others except it’s taller.” “Thank you Dick,” Avona responded. “Can I see those helicopters first hand? I would love to watch them fly.” After carefully observing a few flights she noticed something the others had missed. “Have you noticed that the pink helicopters have longer blades than the white and green ones?” “What?” blurted Tom and Mary. “We never noticed that before! Maybe it’s actually the longer blades that cause the longer flight times.” “Of course,” added Dick. “It’s obvious that flight time should depend on blade length, not on color.” Tom and Mary said nothing, but they each wondered why Dick had not mentioned this “obvious” thing earlier. “Do we have to start over, Avona?” asked Tom, Dick, and Mary. “Not completely. But we are wasting time and money by analyzing only one factor at a time. We’ve spent $216 million and we still don’t know anything about our other product features. When I got my PhD, I learned that the way to maximize the evidence in an experiment is to study several factors at the same time. Let’s do a cube experiment!” “Oh no,” whispered Mary to Tom. “It’s bad enough when she talks about evidence. Now it’s cubes.” Comparing Eight Ways of Doing Things But Avona was right. The cube experiment they decided to run had three factors: color, paper-clip ballast, and blade length. As shown in Table 6, each factor had two levels (settings or choices).
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
170 Using Profit Signals
Table 6 The data matrix for the cube experiment run by Avona, Tom, Dick and Mary.
Avona had lost patience with her senior management peers. She had finally purchased her own copy of the statistical software and installed it on her laptop. As a point of comparison, she first showed everyone the vector analysis in her spreadsheet template (Table 7). “I notice this table is just the same as the others, except it’s wider.” “Thank you, Richard. Anyway, it looks like we have two statistically significant profit signals. The p-values for paper clip ballast and blade length are 0.047 and 0.028, respectively. By looking at the profit signal vector for paper clip (Y), we can see that not adding the weight to the helicopter adds 0.12 second to the overall average flight time. Also, we can see that adding weigh subtracts 0.12 seconds from the overall average flight time. Overall, this means that not adding the weight to the helicopter increases the average flight time by 0.24 seconds compared to adding the weight. That’s $240,000 additional profit per helicopter sold. “By looking at the profit signal vector for blade length (Z), we can see that using the short blade subtracts 0.20 second from the overall average flight time. Also, we can see that using the long blade adds 0.20 seconds to the overall average flight time. Overall, this means that using the long blade instead
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
Table 7 Vector analysis for the cube experiment run by Avona, Tom, Dick and Mary.The raw data are flight times minus the objective of 9 seconds.
171
of the short blade increases the average flight time by 0.40 seconds. That’s $400,000 additional profit per helicopter sold. “The combined effect of these two changes is an increase of 0.64 seconds. This means a total of $640,000 additional profit per helicopter sold. We’ll make millions.” Tom asked, “I know that X, Y and Z are code names for the three factors. But what do XY, XZ and YZ mean?” Avona said, “They are code names for the interactive effects among the factors. An interactive effect exists when the effect of one factor depends on the level (choice or setting) of another factor. In this case there were no significant interactions. Usually there are.” Mary asked, “Is that why it was OK to just add together the effects of paper clip and blade length?” “Exactly!” Next, Avona opened her statistical software and clicked her mouse a few times. Up came the Pareto chart in Figure 5. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
172 Using Profit Signals
Figure 5 Modern statistical software presents analysis results as pictures. Everyone can see just by looking which factors make the biggest difference.
���� ���������������� ��� � ���� ����������� ����������� ����������������������� ������������ ������������������������ ������������������������
�������� ����� ����� ����� ����� ����� ����
Everyone was taken aback to see Avona use a bar chart. “For heaven’s sakes Avona,” cried Mary. “Have you become a bar chart bamboozler?” “Not really. It’s just that modern software manufacturers are smarter than they used to be. They found out customers wanted a quick visual analysis of which factors have the largest effects.” “Rock on!” shouted Tom, Dick, and Mary.
Comparing 256 Ways of Doing Things “Rotcev wants us to test eight different variables,” complained Mary. “That is 28 or 256 combinations. That would cost us $9 million times 256, or $2.3 billion!” “Good thinking,” said Avona. “But with our data matrix software we can screen all eight factors with only 16 helicopters. That would cut our R&D costs by 94 percent.” The team built 16 helicopters with different configurations using two different levels of Rotcev’s 8 factors: paper type, body width, body length, blade length, paper clip, aerodynamic folding, wing tape, and body tape. Figure 6 shows the data matrix for the experiment, including the flight times that were obtained.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
173
Figure 6 Statistical software automatically determines the hyperspace geometry for testing eight different variables simultaneously using only 16 experiments.
The software calculated the vector analysis in less time than it took to click the mouse. The Pareto chart ranking the eight factors by strength of signal is shown in Figure 7.
Figure 7 Statistical software automatically rank orders each factor according to the size of its Profit Signal strength.
“If I read this right,” observed Dick, “It looks like we could be over-engineering our product. Very few of the other factors, including the expensive paper, makes a difference.” “Very astute thinking Dick,” complimented Mary. “I think you just figured out a few good ways for us to make more money.” “Roctev needs to meet this team and hear about these results soon,” said Avona.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
174 Using Profit Signals Chapter Homework Think of these two elements—profit signals and noise—by using your cell phone as an analogy. Strong signals are easy to understand. Noise or static are impossible to decipher. The strong signals in our exercise data matrix came from the two factors that influenced the outcome. Noise has its own vector. Noise, Chance or random variation is a phenomenon of nature. Variation surrounds every measurement and measurement system. Variation is everywhere, always. For example, weigh yourself on a bathroom scale and record this measurement. Wait a few moments and weigh yourself again. Weigh yourself every hour and keep a running record throughout the day. You will discover that your weight may vary by as much as six to 10 pounds per day. Why? Everything varies including your weight and the system used to measure it. A data matrix and the rules of a vector analysis sort profit signals from noise. Statistical evidence is a ratio. Evidence is the length of the profit vector divided by the length of the average noise vector. Evidence, when used to make business decisions, leads to consistently reliable predictions and Six Sigma style profits. The right triangle vector illustrations in this book show how all measurements, all data sets, can be decomposed into these two parts. In this way, facts can be seen by anyone at a glance. There is no need to work equations. So long as you stick with the inherent discipline of a data matrix, you and your colleagues will simply be able to see the answers to problems, be they complex or simple as pie. The deceptive simplicity of 23 cube arrays makes visual, intuitive, correct, and statistically significant inferences possible. It makes analysis fast. Deliberate analytic speed saves enormous amounts of time.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Using Profit Signals
175
Closing Arguments Orville Wright, one-half of the team that used The New Management Equation to create the airplane, comments on the use of data: “I have myself sometimes found it difficult to let the lines run where they will, instead of running them where I think they ought to go. My conclusion is that it is safest to follow the observations exactly.”6
Endnotes Sloan, M. Daniel. Using Designed Experiments to Shrink Health Care Costs. Milwaukee. ASQ Quality Press, 1997. 1
Dilson, Jesse. The Abacus, The World’s First Computing System: Where It Comes From, How It Works, and How to Perform Mathematical Feats Great and Small. New York, St. Martin’s Press,1968. 2
3 A New Aristotle Reader, Edited by J.L. Ackrill. Princeton, Princeton University Press. 1987. Page 39. Sagan, Carl. Science as a Candle in the Dark, The Demon Haunted World. New York, Ballantine Books, 1996. Page 328. 4
Sagan, Carl. Science as a Candle in the Dark, The Demon Haunted World. New York, Ballantine Books, 1996. Page 32. 5
Jakab, Peter L. Visions of a Flying Machine, The Wright Brothers and the Process of Invention. Washington, Smithsonian Institution Press. 1990. Page 140. 6
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
176 Using Profit Signals
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 6
Predicting Profits
M
aking accurate predictions is an important, difficult task. By now, you may not be surprised to learn vector analysis is the international standard for making predictions as well as for making comparisons. This is good news for Corrugated Copters and your company too. The vector analysis methods for solving prediction problems are known as regression modeling and analysis.
Before returning to the exploits of our Six Sigma breakthrough project heroes Mary, Dick, Tom, Avona, and Rotcev an orientation to basic correlation and regression concepts is in order. “Correlation assesses the tendency of one measure to vary in concert with another,” wrote Stephen Jay Gould in The Mismeasure of Man. “The invalid assumption that correlation implies causation is probably among the two or three most serious and common errors of human reasoning.”1 Experienced managers candidly acknowledge that costaccounting variance analysis is based on this faulty premise. No wonder old school spreadsheet forecasts bear so little relationship to actual business sales, revenue, inventory, earning, and other performance metrics. What a wonder. Month after month, trillions of dollars in corporate and governmental resources are squandered trying to explain prediction errors that are inevitable. Never is it recognized that the one-dimensional “prediction” methods mechanized in spreadsheets and institutionalized by business ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
178 Predicting Profits school curriculums were made obsolete in 1890 by Charles Darwin’s half-cousin Francis Galton. In his 1890 essay Kinship and Correlation, Galton wrote, “Few intellectual pleasures are more keen than those enjoyed by a person who, while he is occupied in some special inquiry, suddenly perceives that it admits of a wide Generalization, and that his results hold good in previously-unsuspected directions. The Generalization of which I am about to speak arose in this way.”2 Though Galton did not capitalize the word Generalization in this instance, we did so readers could see he was speaking about a true Law of the Universe.
Fingerprint Evidence It is an entertaining and obscure footnote in the history of evidence-based decisions that by 1893, the same year the grandfather mentioned in our Premise—the man who used paper bags and arithmetic to cipher out his farm’s business transactions because he didn’t trust the new fangled way of doing things called multiplication—the genius Galton was pioneering the use of fingerprints as forensic evidence.3 This breakthrough soon found its way into courtrooms of judgment and justice around the world. The graphic statistical results of vector analysis, applied to a data matrix, are fingerprints. They are the fingerprints every process leaves behind. Each fingerprint data set exhibits unique swirls, bifurcations, endings and statistically significant patterns. This technology, this Generalization, is the grandchild of Francis Galton’s imagination.4 Quoting from Internet sales literature, “Biometrics is a technology that analyzes human characteristics for security purposes. The voice, iris, hand, and face can be used in addition to fingerprints, but among these fingerprints are the most cost-effective.” 5 Think movies. Think Disney. Think Spielberg and Lucas. Think prime time, reality TV. Most all forensic evidence presented by the entertainment industry in whodunits, ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
179
murder mysteries, science fiction, and gumshoe adventures is based on the New Management Equation. Serious biometric predictions—be they concerned with acute lymphoblastic leukemia, therapeutic vaccine studies, or criminal investigations—all use the Pythagorean Theorem.6 This evidence adds up. Profits are too important to be left to the Chance coincidence that a paranormal guess will sometimes be right. Forecasts conjured without the cornerstone of evidence and the New Management Equation, belong in a dust bin with auras, divining rods, Tarot cards, palmistry, past lives, soothsaying, Rune readings, and good old-fashioned wishful thinking. This chapter is weighty. Given the stakes of international commerce, the weight of evidence we present in this chapter is appropriate. If the reading gets a bit heavy for you, peek at the illustrations. Look for the right triangles. Those pictures are our wink at you. You know the secret handshake and inside joke. A vector analysis applied to a data matrix is a correct analysis. Remember, c2 = a2 + b2. You will never have to do any of these calculations. Ever. Data matrix software takes your data and lays it all out for you.
Three Wishes Cost-accounting variance analysis has been around almost as long as Aladdin’s Lamp. Surely, it must have some merit. If we rub it, and rub it, and rub it again with our erasers, the Genie will appear. Will we not be granted our three wishes? Unfortunately, no. We will not. G. Charter Harrison’s standardized cost model was a step forward in 1918. In 2003 it is too simplistic to satisfy international standards for quantitative analysis. Even worse, it is the mother of whitecollar waste and rework. For example, consider the traditional break-even analysis pictured in Figure 1. Here are the three wishes: ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
180 Predicting Profits
Figure 1 The traditional break-even analysis is a good example of wishful thinking in the white-collar work place.
Dollars
������� (Averaged Expenses)
���� �� �� ged � �� era e) v (A com In
������
��������������� ����������������� ����������������
Product or Service Volume
1. I wish my revenue were exactly a straight-line function of volume. 2. I wish my expenses were exactly a straight-line function of volume. 3. I wish the relationship between these lines never changed. Granting these three wishes would be equivalent to suspending the physical laws of our universe. Even a widescreen Disney genie would decline this opportunity. Noise, Chance or random variation, attends every measurement. The mythological Greek Sisyphus had a better chance of rolling his rock to the top of his hill than a manager has of making his monthly results fall exactly on a hypothetical straight line. Table 1
compares and contrasts wishful thinking with reality. As numbers in the first column increase, numbers in the second column increase by an exactly proportional amount. This perfect linear relationship produces perfect predictions. These are plotted in Figure 2. They get rave reviews in management meetings. This sort of line is a sure sign that shenanigans, rather than standards of evidence, are in use. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
181
Table 1 Wishful thinking versus reality. The ‘wish profits’ are hypothetical straight-line predictions. The ‘real profits’ are actual results.
Figure 3
shows the actual performance numbers on which the linear relationship was based. There is, you guessed it, a huge amount of variation. A single-number prediction is useless without a statement of prediction error based on the degree of variation in the process being predicted. We need to see the profit signal and noise vectors. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
182 Predicting Profits
Figure 2 Wishful thinking results falling exactly on the straight-line prediction earn rave reviews in management meetings.
Figure 3 The straight-line predictions were based on real profit data containing a huge amount of variation. A single-number prediction is useless without a statement of prediction error based on the degree of variation in the process being predicted.
A persistent leadership “homily” suggests that idealized targets “inspire” superior performance. We have observed the opposite. Even the best of intentions cannot redeem a patently false premise. Arbitrary goals are products of wishful thinking. They are exercises in futility. They demoralize and debilitate the people assigned to achieve them. They waste time and money that might otherwise find its way to the bottom line. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
183
Prediction Practice “Say Avona,” said Mary, “I have an idea. You would probably call it a hypothesis. First I noticed there is quite a bit of variation in our flight times. Then I noticed there is quite a bit of variation around our target blade length. We already know blade length is a key control variable. Tom told me it would be difficult and expensive to put tighter controls around our tolerance specifications. Finally, I noticed there is also quite a bit of variation in our drop height. “My hypothesis is that drop height could be used as another control variable. This would be relatively easy to do. I have a hunch we might even be able to predict the flight time from the drop height.” “You know physics, so your hunch has my attention,” said Avona. Then, with a knowing smile, “But why would you want to predict flight time from drop height?” “You know perfectly well why! If we can accurately predict performance we can anticipate the future. It would be like knowing what the stock market is going to do tomorrow. We could use that knowledge to make more profits. What is most profitable is best.” “Yes, but can you be a little more specific?” asked Avona. “Well, if we could predict flight time from drop height, we could compensate for a variation in blade length by making an opposite, off-setting change in the drop height. This way, we could hit our advertised flight times with much less variation.” “That’s a great idea,” said Avona. “Make it so.” “Wait a minute,” said Mary. “This isn’t the same as the other things you showed us. We aren’t trying to find the best way of doing something. We’re trying to determine a relationship. Can you give me a preview of how we’re going to do this?” “With pleasure,” said Avona. “Let’s say your data looks like this (Table 2). In problems that involve a relationship between two variables, we have to know which is the ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
184 Predicting Profits
Table 2 The data matrix array for three flight times paired with three different drop heights.
independent variable and which is the dependent variable. The dependent variable is the one we want to predict. The letter Y is used symbolize the dependent variable, which for us is the flight time. “The independent variable is the one we will use to make the prediction. The letter X is used to symbolize the independent variable, which for us is the drop height.” “OK, but what’s this ‘coded drop height’ about?” asked Mary. “The values -1, 0 and 1 are codes for low, medium and high drop heights. At the minus setting I was sitting in my chair. At zero I was standing. At 1 I got up on my desktop. “I know it would look better if we put in the actual drop heights, but it’s easier to explain if we use the codes. It’s also easier to draw the picture and set up the spreadsheet template. Of course, we don’t have to bother with the coding when we use our statistical software. It takes care of everything. “Anyway, here is the vector analysis from my spreadsheet template for this practice problem (Table 3). We’re modeling flight time as a linear function of drop height. If we had more data, we might try something more elaborate. “The only difference between this vector analysis and the ones we did before is the way the profit signal vector gets calculated. The best-fitting straight line is shown in Figure 4. Look closely and you can roughly see that the slope of the predicted line in this case is 1.25. For our actual flight times 8.5, 9.5 and 11.0 the straight line predictions are 8.42, 9.67 and 10.92. Notice that 8.42 plus 1.25 equals 9.67. Also notice that 9.67 plus 1.25 equals 10.92.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
185
Table 3 Vector analysis for fitting Y (flight time) as a linear function of X (drop height).
“This means the forecast goes up 1.25 Y units for every coded X unit. We get the profit signal vector by multiplying this slope times the coded X data vector. “Can you tell me whether or not the slope is statistically significant?”
Figure 4 Best-fitting straight line for Y as a function of coded X.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
186 Predicting Profits Mary thought for a moment, then said, “It doesn’t achieve the ‘clear and convincing’ standard of evidence, because the pvalue is greater than 0.05. It does achieve the ‘preponderance of evidence standard’ because the p-value is smaller than 0.15.” Figure 5 Picture of the vector analysis for fitting Y as a linear function of X.
“Exactly,” Avona exclaimed. “Well done. Here is the drawing for this vector analysis (Figure 5).”
8.5 9.5 11.0
Re gr
es sio
n
Lin
e
Raw Data
Coded
“Notice that the profit signal vector is parallel to the coded X data vector at the lower left. This is always true because the profit signal vector is always equal to the coded X data vector multiplied by the slope of the best-fitting line, 1.25 in this case. “If the slope is larger, the relationship between X and Y is stronger, and the profit signal vector is longer. If the slope is smaller, the relationship between X and Y is weaker, and the profit signal vector is shorter.” Now Mary had a question. “OK, but how do I use all this to make a prediction?”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
187
Avona responded, “Graphically, what we do is start at a coded X value in Figure 4, go up to the fitted line, then over to the raw Y data axis. The number on that axis is the prediction. For example, if we had an X value of 0.5, the predicted Y value would be somewhere between 10.0 and 10.5. “We can be more exact if we are willing to deal with the actual equation of the line: Predicted flight time = 9.67 + 1.25 × (Coded drop height) “9.67 is the average flight time in our practice data set, and 1.25 is the slope. If the coded drop height is 0.5, then we get: Predicted flight time = 9.67 + 1.25 × (0.5) = 9.67 + 0.67 = 10.34 “Applying this equation to the coded X values -1, 0 and 1 is the same as adding together the Y data average vector and the profit signal vector. The result of this is called the predicted Y vector (Table 4). “If you plotted the predicted Y values versus the coded X values, they would fall exactly on the straight line in Figure 4.” “Is the predicted Y vector visible in Figure 5?” asked Mary. “Yes,” answered Avona. “It’s the vector in the shaded plane labled “Regression Line”. It goes from the point (0, 0, 0) up to the point where the noise and profit signal vectors intersect.” Mary had one final question. “If we make a prediction, don’t we have to state the prediction error based on the variation in our data?” “Right again,” said Avona. “But let’s save that for when you get your real data. The statistical software will automatically show you the prediction error.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
188 Predicting Profits
Table 4 Predicted Y vector from fitting Y as a linear function of X.
Predicting Real Flight Times “OK, Avona, I have my data now,” announced Mary as she barged into Avona’s office. “I see you bought your own copy of the statistical software, too,” observed Avona. “Good job.” “Yeah. I get four times as much work done in a quarter of the time it would take with a spreadsheet. I’m giving Corrugated Copters much better information and spending more time with my family. “Anyway, about my study. We did five tests at each of three drop heights. I entered my data into the spreadsheet template you gave me (Table 5). I used coded values -1, 0 and 1 for the low, medium and high drop heights. The p-value is 0.000, so there is a real relationship here, beyond a reasonable doubt. “The overall standard deviation of the flight times is 0.288 seconds. The noise standard deviation is 0.056 seconds. This is astonishing. We can eliminate 81% of our flight time variation by controlling the drop height.” “How in the world did you come up with 81 percent?”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
189
“That was hard. I took a lucky guess. I used the last line in Table 5 to puzzle it out. See the standard deviation of the variation vector is 0.288. The standard deviation of the noise vector is 0.056. Well, 0.056 is 19.4 percent of 0.288. It all added up so I figured that is why you put the last line in your spreadsheet template. “Impressive,” admitted Avona. “Great job, Mary.” “Thank you. Now check me on this next thing. I think the predictive equation is this: Predicted flight time = 1.60 + 0.34 × (Coded drop height). Is that right?” Table 5 Vector analysis for fitting flight time (Y) as a linear function of coded drop height (X). The raw data are Mary’s actual flight times minus the objective of 9 seconds.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
190 Predicting Profits “Yup, said Avona. “You’re batting 1000 today, Mary. 1.60 is the overall average of your Y data, and 0.34 is the slope of the best-fitting line using coded X data. Your equation is right on the money. And, given the reduction in variation you’ve demonstrated, I mean ‘money’ in the literal sense. “Yesterday, I said I would show you how to use the statistical software to get a picture with predictions and prediction limits. Let’s do that now.” Avona clicked her mouse two or three times to produce the graph in Figure 6. “This shows the data, the best-fitting line, and the 95% prediction limits. When these limits are narrower, your predictions are more accurate. When they are wider, your predictions are less accurate.” “Is there an easy way to calculate the limits?” asked Mary. “Yes,” answered Avona. “The upper limit is approximately two noise standard deviations above the predicted value, and the lower limit is approximately two noise standard deviations below the predicted value. Your noise standard deviation is 0.056 seconds. Two times this is 0.112. So, with 95% confidence, predictions from your equation will be accurate almost to plus or minus one tenth of a second. Not bad.”
Figure 6 Picture of the best-fitting straight line for predicting flight time (Y) as a linear function of coded drop weight (X). The 95% prediction limits are approximately two noise standard deviations above and below the line.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Predicting Profits
191
Closing Arguments Peter L. Bernstein, noted economist, economic advisor to nations and multinational companies, and author of the Business Week, New York Times, and USA Today best seller, Against the Gods, The Remarkable Story of Risk, comments on the value of prediction and regression.7 “Forecasting—long denigrated as a waste of time at best and at worst a sin—became an absolute necessity in the course of the seventeenth century for adventuresome entrepreneurs who were willing to take the risk of shaping the future to their own designs. “Commonplace as it seems today, the development of business forecasting in the late seventeenth century was a major innovation. “If nature sometimes fails to regress to the mean, human activities, like sweet peas, will surely experience discontinuities, and no risk management system will work very well. Galton recognized that possibility and warned, ‘An Average is but a solitary fact, whereas, if a single other fact be added to it, an entire Normal Scheme, which nearly corresponds to the observed one, starts potentially into existence.’” 8 Endnotes Gould, Jay Stephen. The Mismeasure of Man. New York: W.W. Norton & Company, 1996. page 272. 1
Galton’s complete works are available from a variety of library and Internet sources. This quote comes from http: //www.mugu.com/galton/essays/1890-1899/galton-1890nareview-kinship-and-correlation.html The origin of this web page is the comprehensive http://www.mugu.com/galton/ start.html 2
3
http://www.mugu.com/galton/ ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
192 Predicting Profits http://www.fme.fujitsu.com/products/biometric/pdf/Find_ FPS.pdf 4
http://www.fme.fujitsu.com/products/biometric/pdf/Find_ FPS.pdf 5
6
http://www.wvu.edu/~bknc/BiometricResearchAgenda.pdf
Bernstein, Peter L. Against the Gods, The Remarkable Story of Risk. New York, John Wiley & Sons. 1966. Page 95. 7
Bernstein, Peter L. Against the Gods, The Remarkable Story of Risk. New York, John Wiley & Sons. 1966. Page 182. 8
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 7 Sustaining Results
S
tewardship entails honorable conduct in the management of other people’s property. But, it is broader than that. It also includes respect for the people’s moral responsibilities. These responsibilities require a constant demonstration of trustworthiness in economic and personal conduct.”1 observed William G. Scott and David K. Hart in Organizational Values in America. The privileges of executive corporate leadership are paired with responsibilities. Every corporate director and senior level executive lives in the world of uncertainty. More than profit is at stake when a manager begins the workday. Jobs and the welfare of one’s community are on the line with every significant decision. Six Sigma products and services are powered by evidencebased decisions. Evidence-based decisions reduce uncertainty. They quantify financial and human risk in ways that can be validated and replicated. They produce near perfect, Six Sigma performance. Near perfect, Six Sigma performance engenders trust. Customers confidently bet their lives, and the lives of their loved ones, on Six Sigma performance. Poor quality management decisions can and do injure our world. Three Mile Island and Chernobyl are monumental governmental management blunders. The Copper7 Intrauterine Device (IUD) and Thalidomide are quintessential medical management mistakes.2 The after effects of the Exxon Valdez wreck in Alaska, the Union Carbide plant explosion at Bhopal, India, Hooker Chemical’s ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
194 Sustaining Results Love Canal disaster at Niagara Falls, the sinking of Brazil’s PetroBras platform and its resulting million-gallon oil spill, all bear witness to the wide ranging effects of poor quality decisions. Evidence-based decisions provide a safety net that can help protect us from poor quality judgments. Seventeen years ago, Nobel Laureate Richard Feynman used an “Avona” model to demonstrate in no uncertain terms that NASA scientists did not understand the concept of correlation.3,4 It is now widely acknowledged that Challenger might still be flying if NASA managers had applied vector analysis to a data matrix in 1986. These powerful tools were widely available at the time. Unfortunately, in 2003 it is clear that NASA managers are still basing important decisions on spreadsheet calculations. Six Sigma has shrunk, and is shrinking, our globe. As one executive told us recently, “Our home office is Earth. The staff meetings I go to look like the United Nations. We emphasize diplomacy both inside and outside the company. Our international relationships build the teamwork we need to compete.” This new level of thoughtfulness is a good thing. Managers around the world now understand, in dollars and cents, that how they treat a worker sewing a soccer ball in Pakistan affects not only the outcome of the World Cup, but also the political stability of the nations where they do business. These are things to think about as we rejoin Corrugated Copters. Tom, Dick, Mary and Avona have stabilized their flight times. They know which way is best. Their challenge now is to hold and gain market share. They need to sustain their best practices and profits.
Evaluating Practices and Profits “Avona,” said Mary, “Our customers are insisting that we manufacture helicopters with virtually no variation in flight time. They told me we have to have a Cpk of 1.5 or better, or we are out as their main supplier. What on earth is a Cpk? Is it an abbreviation for something?” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
195
“No, it’s just another goofy Six Sigma symbol,” said Avona. “What?” “Just kidding. Seriously, Cpk is a numerical index that quantifies how capable a process is of producing virtually perfect quality output. A Cpk of 1.5 implies no more than 3 or 4 defective products or services per million delivered.” “That level of quality is impossible,” said Dick, who just wandered into the break room with a fresh steamed latte. “Not at all,” chided Avona. “A Cpk of 1.5 is an accepted international standard, at least for components and subprocesses. In fact, companies that cannot or will not produce that level of quality are getting edged out or even kicked out of the marketplace.” “You’re kidding.” “No, I’m not. This is old news. Six Sigma is not just a passing fad. It is an extremely disciplined way of competing for market share. Let me show you how to calculate a Cpk value from your data.” “What? Cpk isn’t just the same old New Management Equation?” “Like most other things in Six Sigma, Cpk is based on the New Management Equation. It is a function of the average and the standard deviation. It puts an additional spin on these by combining them with the Upper and Lower Specification Limits.” Avona pulled out a piece of paper and wrote the following expressions.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
196 Sustaining Results “Cpk is defined to be the smaller of two other numbers called Cpl and Cpu. Cpl is the number of process standard deviations between the process average and the Lower Specification Limit, divided by 3. Get it? Process? Lower? Capability? Cpl ? Don’t you just hate the way acronyms aren’t even arranged in order? Cpu is the number of standard deviations between the average and the Upper Specification Limit, divided by 3.” “Why are they divided by 3?” asked Mary. “It’s an arbitrary convention. Everyone uses it. We’re stuck with it. Anyway, we want Cpk to be as large as possible. This means we want both Cpl and Cpu to be as large as possible.” “Maybe you could draw us a picture,” suggested Dick. “You could even put a right triangle in it.”
Figure 1 A process with Cpk = 0.67. The average is 2 standard deviations above the Lower Specification Limit (LSL) and 4 standard deviations below the Upper Specification Limit (USL). The Cpl is 2/3 = 0.67 and the Cpu is 4/3 = 1.33. Roughly 2.5% of outcomes from this process will fall below the Lower Specification Limit.
“Good idea, Dick. The right triangle actually gives us a convenient place to put the standard deviation. Let’s say we have a Lower Specification Limit (LSL) of 7.2 and an Upper Specification Limit (USL) of 10.8. I’ll use a bell-shaped curve to represent process variation. Here’s a process with an average of 8.4 and a standard deviation of 0.6 (Figure 1). The average is 2 standard deviations above LSL and 4 standard deviations below USL. Therefore, Cpl = 2/3 = 0.67
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
197
and Cpu = 4/3 = 1.33 Cpk is the smaller of these two numbers, 0.67. This implies that roughly 2.5% of outcomes from this process will fall below LSL.” “That’s not good,” observed Mary. Avona drew another picture. “Here’s a process with an average of 9.0 and a standard deviation of 0.6 (Figure 2). The average is 3 standard deviations above LSL and 3 standard deviations below USL. Therefore, Cpl = 3/3 = 1.00 and Cpu = 3/3 = 1.00 Cpk is the smaller of these two numbers, but in this case since the process is perfectly centered, these numbers are the same. So Cpk equals 1.00. This implies that roughly 0.3%
Figure 2 A process with Cpk = 1.00. The average is 3 standard deviations above LSL and 3 standard deviations below USL. The Cpl is 3/3 = 1.00 and the Cpu is 3/3 = 1.00. Roughly 0.3% of units produced from this process will fall below LSL or above USL.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
198 Sustaining Results of outcomes from this process will fall below LSL or above USL.” “That’s better,” observed Mary. “But still not good enough for Six Sigma,” said Avona. She drew a third picture. “Here’s a process with Cpk = 1.33. The average is 9.6 and the standard deviation is 0.3 (Figure 3). The average is 8 standard deviations above LSL and 4 standard deviations below USL. Therefore, Cpl = 8/3 = 2.67 and Cpu = 4/3 = 1.33 This implies that roughly 32 outcomes per million will fall above USL.”
Figure 3 A process with Cpk = 1.33. The average is 8 standard deviations above LSL and 4 standard deviations below USL. The Cpl is 8/3 = 2.67 and the Cpu is 4/3 = 1.33. Roughly 32 outcomes per million will fall abve USL.
Avona drew a fourth picture. “Here’s a process with Cpk = 1.67. The average is 9.3 and the standard deviation is 0.3 (Figure 4). The average is 7 standard deviations above LSL and 5 standard deviations below USL. Therefore, Cpl = 7/3 = 2.33 and Cpu = 5/3 = 1.67 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
199
This implies that roughly 287 outcomes per billion will fall above USL. “OK, now for your quiz.” Mary and Dick groaned. “What would Cpk be if we moved the average to 9, right in the center of the specification range?”
Figure 4 A process with Cpk = 1.67. The average is 7 standard deviations above LSL and 5 standard deviations below USL. The Cpl is 7/3 = 2.33 and the Cpu is 5/3 = 1.67. Roughly 287 outcomes per billion will fall above USL.
Mary answered first. “Then the average would be 6 standard deviations above LSL and six standard deviations below USL. Cpl and Cpu would both be equal to 6/3, which is 2, so Cpk would be 2.” “Right on,” said Avona. “And FYI, a process with that level of capability would produce no more than 2 or so defective outcomes per billion.” After a moment of silence, Dick said, “Now that’s something to aspire to.” Process Improvement Simulation “Just for grins, let’s roll some dice,” suggested Avona. “I will record them as you roll. Viva Las Vegas!” After a few rolls, Mary said, “What a surprise! I’m getting numbers between 2 and 12. Hmm, it looks like more of them are in the middle of the range than at the extremes.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
200 Sustaining Results She entered the numbers into her calculator. “The average is about 7 and the standard deviation is a little over 2. “Something must be wrong with my dice,” observed Dick. “All I get are elevens. What’s up with that? Oh, wait a minute. One of my die has a five on every side and the other one has a six on every side. Avona, where did you get these?” “Wizards of the Coast,” answered Avona. “They have lots of games based on probabilities and three dimensional reasoning. By fixing the dice so you always get the same outcome, you are playing with a Six Sigma process. Perfect elevens every time. The only way you can get a different answer is to write down the wrong number.” “Let work out how many ways there are to get each possible outcome with two regular dice,” said Dick (Figure 5).
Figure 5 Seven is the most probable outcome when rolling two dice. Six different combinations will produce a seven, while only one combination will produce either a 2 or a 12.
Tom chimed in, “I know a better way to present the outcomes (Figure 6).” “Don’t you always?” “See, this way it looks sort of like a bell-shaped curve. I bet that makes Avona happy.” “It does,” said Avona. “In fact, it’s a nice segue into what I wanted to show you. We can use the throwing of two or more ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
201
Figure 6 The frequency distribution of outcomes when throwing two dice.
dice to simulate the evolution of process capability through Six Sigma breakthrough projects. Each die will represent a cause of defects. The sum of the dice will represent the number of defects per helicopter. “The average flight time for our current design is 11 seconds. That means our average profit margin is $2 million. Let’s say these defects cost an average of $100,000 each to repair. If a helicopter has 20 defects, we break even. If it has more than 20 defects, we lose money. Therefore, our Upper Specification Limit is 20 defects.” “20 sounds like an awful lot of defects,” said Dick. “Can’t we make it a lower number?” “Work with me, Dick. It’s just a simulation. OK, let’s get started. Our initial process involves four dice.” For each simulation, one of them threw the four dice, one of them called out the result, and one of them entered the result into their data matrix statistical program. They traded jobs once in a while. After completing 1000 simulations, they decided they had had enough. Avona showed them how to do a process capability analysis with two mouse clicks. (Figure 7).
Figure 7 Process capability analysis of the initial “four dice” process.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
202 Sustaining Results
“We get only Cpu in this analysis because there is only an upper specification limit. In situations like this, Cpu and Cpk are the same thing. Using the bell-shaped curve with ‘Mean’ marked on it, my calibrated eyeball tells me the average number of defects per unit is about 14. The standard deviation is 3.4. Cpk is 0.56. As you can see, the number of defects on 4.7% of helicopters will be ‘Above USL’. In other words, 4.7% of them will have more than 20 defects, and we will lose money.” “We might also lose market share,” said Mary. “There’s no guarantee we can catch all the defects before a helicopter goes to a customer. It would be better to keep them from happening in the first place.” “You got it, girl,” said Avona. “Let’s assume now that we have data-mined our process data base, using vector analysis of course, prioritized the causes of defects, and successfully eliminated one of the top causes. Our improved process involves only three dice.” After completing another 1000 simulations, Avona showed them the capability analysis of the improved process (Figure 8). Figure 8 Process capability analysis of the new, improved “three dice” process.
“Average defects per unit, using the calibrated eyeball method, the mean marked on the small distribution curve and the 0-25 scale just below it in Figure 8, have gone from 14 to about 10. This is an average savings of $400,000 per helicopter. The standard deviation is about 3. Cpk is 1.05. The number of defects on 831 parts per million (PPM) will be ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
203
‘Above USL’. In other words, we’ll make money on all but one helicopter in a thousand.” “This does look a lot better than the old process,” said Tom. “But what if the average drifts up over time?” “That’s a very good question,” said Avona. “That’s exactly why we can’t afford to be satisfied with a Cpk that is barely over 1. We need to eliminate other causes of defects so that upward drifts don’t result in yield loss. Let’s say we have done that, and our process now involves only two dice.” After they completed another 1000 simulations, Avona showed them the capability analysis of the new process (Figure 9).
Figure 9 Analysis of the ultra-capable “two dice” process.
“Average defects per unit, using the calibrated eyeball method, has gone from 10 to about 7. This is an additional savings of $300,000 per helicopter. The standard deviation is about 2.5. Cpk is 1.75. The number of defects on 78 parts per billion (0.078 PPM) will be ‘Above USL’. In other words, we’ll make money on all but one helicopter in 10 million.” “That’s not all,” added Tom. “With an upper three-sigma limit of about 14 as a control limit, we can detect upward drifts before they cause yield loss.” “Good point,” noted Avona. She did another simulation, a small one this time. She produced a control chart showing
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
204 Sustaining Results what would happen if the causes of defects in the “three dice” process came back (Figure 10).
Figure 10 The first 40 units come from the “two dice” process, the last 10 come from the “three dice” process. The Upper and Lower Control Limits (UCL and LCL) are the three-sigma limits for the “two dice” process.
“A control chart uses the average and the three-sigma limits to monitor a process over time,” explained Avona. “In my simulation, the first 40 units came from our “two dice” process. The last 10 units came from the “three dice” process. Can you see what happened on the chart?” “Yes,” said Dick. “The dots got a lot bigger.” “Thank you, Dick. Actually, I did that myself to distinguish the three-dice units from the two-dice ones. What about the data in relation to the average and upper three-sigma limit?” “Well,” said Mary, “the two-dice units were evenly distributed around the average, and none of them were above the upper three sigma limit. The three-dice units are all above the average, and one of them is above the upper three-sigma limit.” “That’s right,” said Avona. “Those are the two most important rules for interpreting a control chart. These are signals that something has changed. So what you said before was exactly right, Tom. We can use the average and the threesigma limits of the two-dice process to catch an upward drift before it causes any yield loss.” “Also, these rules give us operational definitions of when to initiate troubleshooting,” said Tom. “Otherwise, everyone would have a different interpretation of the same data. Hmm, this reminds me of the ‘standards of evidence’ you’re always ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
205
talking about, Avona. Are control charts related in some way to that?” “Bingo!” exclaimed Avona. “Allow me to explain.” “OK,” said Mary, “but can we take a break first?” Monitoring Practices and Profits After the break, Avona showed the team a table of financial results (Table 1).
Table 1 Quarterly financial report (thousands of dollars).
“In many companies, the Executive Committee agonizes over numbers like these every quarter. They make bar charts (Figure 11). They try to figure out what went wrong in Quarter 5, and who to blame. They try to take credit for Quarter 13, and come up with imaginative explanations.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
206 Sustaining Results
Figure 11 Quarterly financial results (thousands of dollars).
“Don’t we do the same thing?” asked Mary. “We used to,” answered Avona, “but not any more. Not since Rotcev took over. He immediately insisted that we apply standards of evidence everywhere, not just in manufacturing.” “Rotcev is almost as enthusiastic about this stuff as you are, Avona,” commented Dick. “But I didn’t realize you could use it on financial data. I thought the accountants had their own special ways of doing things.” “They did,” said Avona. “That was the problem. They could make the numbers say pretty much whatever our previous CEO wanted them to say.” “To do this right,” continued Avona, “we need to put the numbers into a data matrix (Table 2). Then we can apply a vector analysis. “Looking only at totals is a big problem with traditional cost accounting variance analysis. For example, if we look only at quarterly totals, we lose all the information in the monthly numbers. “If we had weekly data, we would start with that. Looking only at monthly or quarterly totals would lose all the information in the week-to-week variations. “With vector analysis, we use all the information in whatever data we have. The analysis in Table 2 is exactly the same as if we were comparing 15 ways of doing something. The profit ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
207
Table 2 Data matrix and vector analysis for quarterly review of monthly financial data.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
208 Sustaining Results signal vector contains all the information about differences among the 15 quarters. The noise vector contains all the information in the month-to-month variations. “Here is the cornerstone of evidence for this vector analysis (Figure 12). It gives the lengths of all the vectors.
Figure 12 The cornerstone of evidence for the vector analysis in Table 2. The numbers in parentheses are the lengths of the vectors. The three long vectors are not drawn to scale.
DA TA AV E
VAR IA (185TION )
RAW DATA (3356)
(335
2)
RA
GE
NOISE (162)
(33
50)
IT OF AL R P GN SI (90)
“Who can tell me if there are any significant differences among the 15 quarters?” “There aren’t any,” Mary quickly replied. “The p-value is 0.792. It doesn’t meet any standard of evidence. The apparent quarter-to-quarter changes are just noise, not signals.” “That’s right,” said Avona. “Now, the F ratio basically compares the degree of variability in the profit signal vector to the degree of variability in the noise vector. We reached our conclusion because the F ratio wasn’t large enough to achieve a standard of evidence. In other words, the profit signal variation wasn’t large enough compared to the noise variation. “We can confirm this visually by plotting the profit signal and noise vectors together on a single graph (Figure 13). “The solid line is the profit signal vector and the dotted line is the noise vector. They are plotted in time sequence. The overall degrees of variability are about the same.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
209
Figure 13 The numbers in the profit signal vector are plotted as the solid line. Each of these is an average of 3 numbers in the variation vector. To make the visual comparison statistically valid, the numbers in the noise vector were first divided by the square root of 3. Don’t blame us, it’s a Law of the Universe. The adjusted noise numbers are plotted as the dotted line.
“Creating a graphical comparison like this is a little trickier than it looks. To make the comparison statistically valid, I had to divide the numbers in the noise vector by the square root of 3. This is because each number in the profit signal vector is an average of 3 numbers in the variation vector. I know this is confusing, but it’s a Law of the Universe. “Anyway, in 1924 a man named Walter Shewhart was trying to come up with a good graphical method for analyzing data over time when there is a natural or logical way of grouping the data. For example, we grouped our raw monthly data by calendar quarters. That made sense because the Executive Committee reviews it on a quarterly basis. Shewhart called these rational sub-groups.5 “Instead of plotting the profit signal vector and adjusted noise vector on top of each other, Shewhart decided it would be better to plot just the profit signal numbers, and use horizontal lines to represent the upper and lower three-sigma limits of the adjusted noise numbers. Also, he decided to add the data average vector to the profit signal and noise vectors. He felt this would be easier to interpret. “In other words, he invented what we now call the X-bar chart control (Figure 14). “This control chart tells us the same thing as the F ratio: the quarter-to-quarter changes are just noise.” “I have a question,” said Dick. “Do we have to do the vector analysis all over again every quarter?” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
210 Sustaining Results
Figure 14 The dots are the averages of the monthly revenues in each quarter, not the totals. The centerline is the grand average of all the monthly numbers. The Upper Control Limit (UCL) is 3 noise standard deviations above the average. The Lower Control Limit (LCL) is 3 noise standard deviations below the average.
“That’s a good question,” answered Avona. “Fortunately, the answer is no. Once we have a good baseline, like we have in this example, we hold the control limits constant and just plot the new numbers as time goes by.” “I guess the fundamental things you’ve taught us really do apply,” said Dick. “OK, here’s your quiz. What are two ‘events’ on this chart that would indicate a real change of some kind?” “A point outside the control limits,” said Tom. “A bunch of points in a row above the center line,” said Mary. “Right on both counts,” said Avona. “Remember, Mary, it could also be a bunch of points below the center line. And, by the way, the usual requirement is eight in a row for a statistical signal.”6 “This is great stuff,” said Mary. “But I’ve been wondering: aren’t we still losing some of the information in the month-tomonth changes?” “Excellent point,” said Avona. “Shewhart was aware of this problem. His solution was to plot the standard deviations of the subgroups on their own control chart (Figure 15). The two charts together give us a complete picture of what’s going on over time.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
211
Figure 15 The dots are the standard deviations of the monthly revenues in each quarter. The centerline is the average of the standard deviations. The Upper and Lower Control Limits (UCL and LCL) are three-sigma limits based on the standard deviation of the standard deviations. Strange, but true.
Taking Action After the others left, Avona realized there was another basic fact about control charts that she needed to teach them. It wasn’t about how to set up the charts, or how to interpret them. She felt that was pretty easy. She knew from experience that control charts were all too often used as “window dressing”. Maybe “wallpaper” is a better analogy. In manufacturing at least, she knew that control charts add real value only when they are used as a basis for action. She also knew that reacting to control chart signals was a process, just like any other business activity. In order to add value, the reaction process must be defined and documented. It must be improved over time. She had found the tools of Process Mapping to be ideal for these tasks. In her experience, it worked best to have teams of operators, supervisors, maintenance technicians, engineers and managers develop the reaction plans together. She had a reaction plan “skeleton” she always used to get them started (Figure 16). The question “Signal?” refers to one or more pre-defined signals on one or more control charts. The charts and signals are defined by the team that develops the plan. The term ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
212 Sustaining Results
Figure 16 A generic reaction plan “skeleton” for a manufacturing or service process.
“escalate” means to raise the level of the investigation by bringing in someone with greater expertise. Ideally, the manufacturing or service process is stopped until “Continue” is reached. Figure 17 shows an actual example of a reaction plan for a lot-based manufacturing process.
Figure 17 An example of a reaction plan for a manufacturing process.
In this example, the team decided to confirm a control chart signal by immediately taking a second sample from the same lot. If the second sample does not show a signal, the ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
213
occurrence is documented and the lot moves on to the next operation. If the second sample does show a control chart signal, the manufacturing process is put on hold while the Operator goes through a pre-determined checklist. The checklists in a reaction plan are determined by the team that develops the plan. That is why it is so important that all vocations are represented on the team: operator, supervisor, maintenance technician, engineer, and managers. If the operator solves the problem, the occurrence is documented and the lot moves on to the next operation. Otherwise, the supervisor is called in. It may be necessary to bring in the engineer, or the maintenance technician, or even the manager. The important point is that the manufacturing process remains on hold until one of two things happen: 1. The problem is solved. 2. Someone of sufficiently high authority makes the decision to resume manufacturing while the problem is being worked on. The keys to the success of reaction plans are: (a) Orderly and consistent evidence-based response to problems as they occur. (b) Visibility of problems throughout the organization, appropriate to their level of severity. (c) Evidence-based decisions made at the appropriate levels of responsibility throughout the organization. A disciplined approach like this is a bitter pill at first. Supervisors and managers object to the loss of production time. After a few weeks or months, the same supervisors and managers are singing the praises of their reaction plans. Invariably, they have seen their unplanned downtime plummet. Problems are being fixed right away, instead of being ignored until they become catastrophes. These short-term economic benefits are overshadowed by long-term improvements in process capability. The old ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
214 Sustaining Results “four dice” process gives way to the “three dice” process. We think, “That’s great, but that’s it. It’s impossible to get any better.” But we keep following our reaction plan, refining it as we learn new things. One day we wake up and, to our great astonishment, the impossible has happened. We find ourselves with an ultra-capable “two dice” process. We find our competitors using us as the benchmark. You are asking yourself, is this really possible? Was Pythagoras right about right triangles? Is the earth spherical, and does it revolve around the sun? Does multiplication work? Do gravity and electricity exist? Do airplanes fly? Can you buy things on a computer and have them delivered to your door? Are vectors and hyperspace real? All the evidence we have says “yes”. Closing Arguments “The idea of control involves action for the purpose of achieving a desired result.” Walter A. Shewhart. Statistical Method from the Viewpoint of Quality Control, 1939.7
Endnotes Scott, William G. and, Hart, David K. Organizational Values in America. New Brunswick, Transaction Publishers, 1991. Page 139. 1
Committee on Quality of Healthcare in America. Kohn, Linda T., Corrigan, Janet M., Donaldson, Molla S. Editors. To Err is Human, Building a Safer Health System. Washington, D.C. National Academy Press. 2001. 2
http://www.student.math.uwaterloo.ca/~stat231/stat231_ 01_02/w02/section3/fi4.4.pdf and http://www.ralentz.com/ old/space/feynman-report.html 3
http://www.uri.edu/artsci/ecn/mead/306a/Tuftegifs/ Tufte3.html 4
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Sustaining Results
215
Shewhart, Walter A. Economic Control of Quality of Manufactured Product, New York, D. Van Norstrand Company, Inc. 1931. Republished in 1980 by American Society for Quality Control. 5
AT& T Technologies. Statistical Quality Control Handbook, copyright 1956 by Western Electric. Copyright renewed by AT&T Technologies, Inc. 1984. 6
Shewhart, Walter A. Statistical Method from the Viewpoint of Quality Control. New York, Dover Publications, Inc. 1986. 7
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
216 Sustaining Results
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Chapter 8 The Three Rs
E
ducation and training are the first steps in building an organization founded on evidence-based decisions and the New Management Equation. Knowledge and skill necessarily change the nature of authority.1 Trust, decency and respect replace fear and favor as social adhesives.2
American history provides an excellent road map for redefining the 3 Rs—Reading, wRiting, and aRithmetic—to Reading, wRiting, and vectoR analysis. John Adams wrestled with evidence as we all do. “Facts are stubborn things; and whatever may be our wishes, our inclinations, or the dictates of our passions, they cannot alter the state of facts and evidence.”3 Adams and his colleagues were as passionate about intellectual liberty as they were about freedom. “Liberty cannot be preserved without a general knowledge among the people.” Thomas Jefferson wrote not only to Adams, but to all of us in his 1821 autobiography. “We thought that on this subject, a systematical plan of general education should be proposed, and I was requested to undertake it. I accordingly prepared three bills for the Revisal, proposing three distinct grades of education, teaching all classes.”4 Jefferson’s friend and ghostwriter got, characteristically, to the point. “Education should be on the spot; and the best method… I call for the education of one million and thirty thousand children.” This was a bold proposal. Those of us who enjoy the rare combined privilege of Untied States ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
218 The Three Rs citizenship and an American public education can thank Thomas Paine.5 Paine’s outlandishly impractical investment scheme turned out to be the bargain of the millennium. We predict that in the new millennium, Paine’s good deal can yield even better bottom line business results.
Six Sigma’s Hidden Factory “I have been thinking about what you have taught us Avona,” said Dick. “Obviously I had a hard time understanding those spreadsheet tables. Once you showed me the Pareto chart that rank-ordered the factors, my light finally went on.” Roctev the CEO, Avona, Tom and Mary listened. They were smiling. “Thanks for not making fun of me when my lights were out,” said Dick. “Numbers make me nervous. Then I get embarrassed. Then I say dumb things I wish I could take back. “Anyway, I drew a flow diagram yesterday. That darned picture kept me awake all night long. I was driving my wife crazy tossing around. So I got up at 3 AM and came into work.” Roctev, Avona, Tom, and Mary looked at Dick’s map (Figure 1). “You are one brave guy,” complimented Roctev. “The last time an employee told me I was full of baloney was when I was a Vice President of Marketing.” Everyone stared at Roctev. “I can see exactly what you mean by my comfort zone. I began getting uncomfortable in 1986. Before that, I had it all. Big office. Big desk. Four phone lines. Two secretaries and a million-dollar advertising budget. Heck, if I couldn’t make myself look better than everyone else at the annual review, I would have had a real problem. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
The Three Rs
Figure 1 The hidden factory of traditional Six Sigma. Projects are delayed and deferred, not because of cost accounting, finance or data analysis. Bold proposals can make people so uncomfortable that they would rather waste money than upset the status quo.
219
“One day I was working a night shift to ‘get close to my employees.’ I sat down next to a clerk. I believe she was 19. I do remember she was a sophomore at USC. I remember that because USC stood for the University of Southern Colorado, not the University of Southern California. She was none too pleased to have me pay her a social visit. “So, after about 30 minutes of chit chat, and since the lobby was vacant at one AM, she told me she had homework to do. I tried to keep up the conversation by expressing an interest. She shut her mathematics text and looked right at me.” “You know, you guys in senior management don’t have a clue, do you.” she said. “Ummm. What do you mean exactly?” “Well I read that bull poop memo about your recent management decision. You know the one with all the numbers?” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
220 The Three Rs “Yes. I do.” “Well, anybody who has taken Statistics 101 at USC can tell you don’t even know how to do an Analysis of Variance.” “I had taken Statistics 101 as a foreign exchange student at Baldwin-Wallace College in 1969. I never did learn what Analysis of Variance meant, so I said, ‘Well actually I did take that class but I got a C in it and that was a gift. Please show me what you mean.’” “Here, I will show you.” “She pulled out a piece of graph paper, a calculator, a ruler, and a pencil. She drew me a picture. Our layoff plan to save money and our $11 million senior management data analysis on the supposed need for a massive building program just got an F. Her whole show took about five minutes. I thanked her and excused myself. “I figured if a 19-year-old could see through the faulty reasoning behind the decisions our CEO, CFO, COO and me were making, maybe the other 1,000 employees could too. I did not sleep well. But, I did decide to confront my math phobia. I learned how to draw control charts. My colleagues chose to keep on bamboozling. Many of them still are. They all got promoted. “One thing led to another. Now I am a CEO. Instead of a 19year-old kid with spunk, I have a top-flight team. You.” “Wow,” said Mary. “Whew,” said Dick. “I thought you’d go seriously supersonic if you thought I was saying you blame accounting and finance instead of stepping up to the plate and just saying, ‘This change scares the heck out of me.’” Roctev looked at Dick, “So, you say we have this gigantic, Six Sigma hidden factory of rework. Just for the sake of the argument you thought we would have, let’s say your map is true. My comfort zone is the problem that stops projects. As CEO, I am the reason for Six Sigma project rework. What should I do? What would you do?” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
The Three Rs
221
“I would start phasing out the Six Sigma bureaucracy.” “What!” cried Tom and Mary who had just mounted and framed their Black Belt certificates. Avona chuckled quietly and looked at her shoelaces. Roctev nodded his head. “Dick, you just earned your American Society for Quality Six Sigma Black Belt certification. Do you mean to tell me that you are proposing to give it all up for the good of the company?” asked Roctev. “Let me get this straight Dick,” said Tom with a stern look on his face. “Are you saying that Black Belts aren’t needed?” “No. No.” replied Dick. “You are an expert. You are a teacher. We need experts and good teachers. But people respect you and Mary because of what you know and do, not because of your numbered certificates. “All I am saying is, I think we need everybody. Everybody has a brain. All the people we work with have imaginations. Avona, her models, simulations, and the software have made Six Sigma so simple, everybody can contribute. “Six Sigma is simply the use of evidence-based decisions. That idea is as old as Aristotle. All this Black Belt and Green Belt stuff is overhead. It would be less costly, and more effective if we called our Six Sigma program the Three Rs.” “Huh?” said Mary. “You know. We could have some fun with it. Reading, wRiting, and vectoR analysis,” suggested Dick. “Maybe it could be Reading, wRiting, and Refraction. Whatever. We could just call Six Sigma ‘literacy’. Let’s hire people who are literate, or who want to be literate. I even used a cube to outline the three factor interaction.” (See Figure 2) “Literacy?” “Yes. Let’s call the way we work literacy and be done with it.” “Let’s go get a latte, chai and biscotti,” said Avona. “I’m buying.” ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
222 The Three Rs
Figure 2 Literacy now refers to people who know how to read, write and vector-analyze numbers.
vectoR analysis wRiting
Reading
Our Proposal A global workforce that is literate in the Three Rs of the New Management Equation is an excellent value. It is far less costly than the alternatives. But there is a cost. “New arts destroy the old. See the investment of capital in aqueducts, made useless by hydraulics; fortifications by gunpowder; roads and canals, by railways; sails, by steam; steam by electricity,” wrote Ralph Waldo Emerson.6 His observations ring true as we watch vacuum tubes made almost useless by transistors, transistors by silicon chips; palpation by Magnetic Resonance Imaging; auscultation by ultrasound; poisonous purple foxglove seed remedies by quality controlled digitalis; telegrams, by wireless communication; typewriters by computer keyboards; the cost accounting variance analysis by the Analysis of Variance; spreadsheets by the data matrix software.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
The Three Rs
223
In 1992, while grappling with the hand calculations, ruler, pencil, and Xeroxed chart template required to produce a statistical process control chart, the Chief Executive Officer of Northwest Hospital in Seattle at that time, criticized cost accounting by quoting Emerson. His daring Total Quality Management (TQM) observation is intrepid today.7 “A foolish consistency is the hobgoblin of little minds, adored by little statesmen and philosophers and divines.” Though none of us were able to articulate a proposal for improving the cost accounting variance analysis back then— that being a vector analysis applied to a data matrix—we came to discover that the rest of Emerson’s quote was prophetic. Sometimes you just get lucky. “With consistency a great soul has nothing to do. He may well concern himself with a shadow on the wall. Speak what you think in hard words and tomorrow speak what to-morrow thinks in hard words again, though it contradict every thing you said to-day— ‘Ah, so you are sure to be misunderstood,’—Is it so bad then to be misunderstood? Pythagoras was misunderstood, and Socrates, and Jesus, and Luther, and Copernicus, and Galileo, and Newton and every pure and wise spirit who ever took flesh. To be great is to be misunderstood.”8 Each age, as Emerson pointed out, must write its own books. The books of an older generation will not fit ours. Motorola’s Six Sigma business initiative was designed at a time when a dual 5.25-inch floppy disk drive IBM computer with an amber screen was an executive luxury. Harvard Graphics bar charts on a dot matrix printer were breakthrough technology. When General Electric got a hold of Six Sigma, the Internet and Windows 95 were new. 9600 Baud was a fast connection. Time flies. Can it be our own college students were babies in the eighties? Great Caesar’s Ghost! We are old men. How can this be? We have grey hair. There are bald spots on the back of our heads. We are wearing progressive lens glasses. Our Les Paul, Stratocaster and PRS guitars and Cry Baby Wah-Wah pedals are antiques for sale on EBay. When did this happen? Do we look as funky as a judo gi and as old as a Six Sigma acronym? ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
224 The Three Rs
Yes. We do. We must work, and work very hard, to stay young. We must change with the times. We must learn, unlearn and relearn how to do things. We must try to age gracefully. We must negotiate and we, all of us, must get to Yes together.9
Endnotes Wood, Gordon S. The Radicalism of the American Revolution. New York, Alfred A. Knoph. 1992. page 189. 1
Wood, Gordon S. The Radicalism of the American Revolution. New York, Alfred A. Knoph. 1992. page 189. 2
http://www.dropbears.com/b/broughsbooks/history/articles/ john_adams_quotations.htm 3
Jeffereson, Thomas. The Life and Selected Writings of Thomas Jefferson, edited by Adrienne Koch and William Peden. New York, Random House. 1944. Page 48. 4
Paine, Thomas. Collected Writings. New York. Library of America. Pages 630-633. 5
Emerson, Ralph Waldo. Circles from The Portable Emerson, edited by Carl Bode in collaboration with Malcolm Cowley. New York. Penguin Books. Page 229. 6
Sloan, M. Daniel and Torpey, Jodi B. Success Stories in Lowering Health Care Costs by Improving Health Care Quality. Milwaukee, ASQ Quality Press. 1995. Pages 87-97. 7
8
Emerson, Ralph Waldo. Self-Reliance.
Fisher, Roger, and Ury, William, Getting to Yes, Negotiating Agreement Without Giving In. New York, Penguin Books. 1981. 9
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
I. Glossary of Terms: Data Matrix, Vector Analysis And Evidence-based Decisions ANOVA – Acronym for Analysis of Variance, Fisher’s general term for the various forms of vector analysis he developed. Confidence level – Obtained by subtracting the p-value from the number 1 then multiplying by 100. It is a measure of the strength of evidence in the data against the null hypothesis. Cornerstone of Evidence – This is a generalized tetrahedron representing a vector analysis. Each of the four faces is a generalized right triangle. The six sides or edges represent the raw data vector and the five possible vector components of variation that can be broken out of any set of raw data. Data matrix – An array of numbers or labels in rows and columns. Each row is an object, entity or event for which we have collected data. Each column is one of the variables we have measured or observed. Data vector – A stack of numbers or labels treated as a single entity. A column in a data matrix is a vector. It is a point in n-dimensional space, where n is the number of rows in the data matrix. DMAIC – This is an acronym for Design, Measure, Analyze, Improve and Control, which is the Six Sigma project cycle. Factor – A controlled variable in a designed experiment. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
226 Appendices F ratio – A measure of the strength of evidence in the data against the null hypothesis. A statistic proportional to the ratio of squared length of the profit signal vector to the squared length of the noise vector. New Management Equation – Our name for the Pythagorean Theorem. Noise – The chance, normal, common, random, statistical variation found everywhere in Nature. It is a Law of the Universe. P-value – The probability of getting, by chance alone, an F ratio as large as the one we got. A p-value less than 0.15 gives a ‘preponderance of evidence’ against the null hypothesis. A p-value less than 0.05 gives ‘clear and convincing’ evidence against the null hypothesis. A p-value less than 0.01 gives evidence ‘beyond a reasonable doubt’ against the null hypothesis. Profit SignalTM – Quantifies and rank orders which factors impact any business, manufacturing, or service process. It is the vector at the bottom right-hand, forward corner of the tetrahedron. The ratio of the length of this vector to the length of the noise vector in a correct analysis yields the F ratio that measures the strength of evidence. Profit signal vector – Same as profit signal. Pythagorean Theorem – The square of the long side of a right triangle is equal to the sum of the squares of the other two sides. c2 = a2+ b2. Tetrahedron – A three-dimensional figure with four triangular faces and six edges. Vector – An arrow that defines magnitude and direction, connecting one point in space with another. Vector analysis – The process of breaking up a raw data vector into perpendicular vector components of variation.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
227
II. The Business Bookshelf 1. Aristotle’s books on Logic, Eudemian Ethics, and Politics are essential. These texts outline the sequential, Inductive/ Deductive cycle of the scientific method: Hypothesis, Experiment, and Test Hypothesis. Aristotle’s cycle is the foundation for all science and Walter Shewhart’s original Plan, Do, Study, Act cycle and M. Daniel Sloan’s IDEA Cycle: Induction, Deduction, Evaluation, Action. Posterior Analytics suggests that the triangle signifies truth. Eudemian Ethics details the links between a respect for the individual, knowledge, the pursuit of happiness, virtue, and a good social order. 2. David Hume’s A Treatise of Human Understanding, 1739, emphasizes the importance of sequential perceptions. Hume’s ideas, writing, and thinking reflect the British personality in applied science. Bacon, Nightingale, Newton, Darwin, Fisher, and Box are names that can be culturally linked to Hume’s work. 3. The Declaration of Independence, 1776. This classic document addresses life, liberty, the pursuit of happiness, justice, and a good social order. Jefferson, Franklin, Washington, Adams, and other American revolutionaries were Hume’s contemporaries. They communicated. The study of philosophy, science, and mathematics were integral to their lives. 4. Immanuel Kant’s Critique of Pure Reason, 1781, tackles the complexity of quality. The circular logic of knowing, the quality of judgments, relationships, and modality are dealt with in one difficult, challenging text. Einstein specifically honored this work as an inspirational force in his work. 5. Albert Einstein’s little book, Relativity, 1917. Einstein introduces the idea that “the evolution of an empirical science is a continuous process of induction.” Dr. Einstein’s ideas on time, the sequential order of perceptions, measurement, and analysis are landmarks. He specifically describes his use of probability, the Pythagorean Theorem, and the Cartesian coordinate system. He provides a complete listing of applied ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
228 Appendices science thought leaders: Euclid, Galileo, Kepler, Descartes, Gauss, Hume, and Kant. 6. Ronald A. Fisher’s works from 1913-1935. (A good resource for finding them is Collected Papers, Volumes 1-5 J.H. Bennet, Ed. The University of Adelaide, 1971-1974). At age 25, Fisher’s 1915 Biometrika, 10, paper entitled, “Frequency Distribution of the Values of the Correlation Coefficient in Samples from an Indefinitely Large Population” introduces the idea of using geometry to represent statistical samples. The Pythagorean Theorem or New Management Equation is a Generalization. It applies to samples of any size. Fisher’s 1921 Metron paper, “On the Probable Error of a Coefficient of Correlation Deduced from a Small Sample,” explains the logarithmic transformation of the correlation coeffient r that leads to a near normal distribution. He presented a table tabulating the transformation for each value of r. Statistical Methods for Research Workers, 1924. This book details the practical application of a circular, inductive and deductive logic cycle. The thirteenth edition credits W. Edwards Deming for the extension of the z Table to the 0.1 level of accuracy. The Design of Experiments, 1935, is a phenomenal, seminal work. “Inductive inference is the only process known to us by which essentially new knowledge comes into the world.” The importance of experimental observations must be connected to the “precise, deductive reasoning” of Euclidean geometry. 7. Clarence Irving Lewis, Mind and the World Order, Outline of a Theory of Knowledge. 1929. This book inspired Walter Shewhart. The philosophy of conceptual pragmatism led to the development of the field of Six Sigma quality improvement. 8. Walter A. Shewhart’s Economic Control of Quality of Manufactured Product, 1931. This book includes illustrations and ideas from Fisher’s work and his own unique perspective on the importance of sequential data analysis. Induction precedes deduction. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
229
“The Nature and Origin of Standards of Quality,” was written in 1935 and was published in the January 1958 issue of The Bell System Technical Journal. He describes the character of the continuous improvement cycle as legislative, executive, and judicial. Statistical Method from the Viewpoint of Quality Control, 1939, is taken from a series of US Agricultural Department lectures delivered at the invitation of W. Edwards Deming. Pages 44 and 45 contain the graphic illustration of a continuous improvement cycle: Hypothesis (Legislative in nature), Experiment (Executive in character), Test Hypothesis (Judicial). Induction precedes deduction. 9. W. Edwards Deming’s Some Theory on Sampling, 1950, is a noteworthy historical book. It was directly affected by Fisher and Shewhart’s work. The first chapter addresses the primary importance of the design of an experiment. Deming details the geometry of sample variances on page 62. On the “Distinction between Enumerative and AnalyticSurveys,” The American Statistical Association Journal, June 1953, comes directly from Some Theory on Sampling. This article shows the futility of using random samples for analyzing a dynamic process. “On a Classification of the Problems of Statistical Inference,” June 1942, Number 218, Volume 37, gives Deming’s vision of a quality controlled health care system. Out of the Crisis, 1986. Here are fourteen points to ponder for a good social order in the workplace to ponder. His PDCA improvement cycle dates to Aristotle. Deming cites the influence of Shewhart, Clarence Irving Lewis, and Fisher. It is curious to note that Deming’s 1951 understanding of the importance of a designed experiment and the economy/ geometry of sample, is absent from this work. 10. Ludwig von Bertalanffy’s General System Theory, 1968. This is a best-of-class book on systems thinking. 11. George Box, William G. Hunter, and J. Stuart Hunter, Statistics for Experimenters, An Introduction to Design, Data Analysis, and Model Building, 1978. This is a master work of applied science. The pictures R.A. Fisher imagined are drawn. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
230 Appendices Many of the important algebraic expressions Fisher wrote are translated. Somehow Fisher’s ideas are simplified. Induction precedes deduction. One of the book’s essential main points, Fisher’s main point, is hidden from view. The Pythagorean Theorem provides sound theory for all standartd statistical theory. 12. Steve deShazer’s Clues: Investigating Solutions in Brief Therapy, 1988, is the only therapy model and/or psychological theory we know of that was developed using probability theory, inductive reasoning, and flow diagrams. This model for rapid improvement works well in systems of any size. Mr. deShazer formally opposes a focus on defects, defectives, and problems. Rather one should focus on solutions and doing more of what works. 13. Darrell Huff, How to Lie With Statistics. This is the definitive 20th Century work on the Big Bamboozle. 14. Roger Fisher and William Ury, Getting to Yes, Negotiating Agreement without Giving In. This is the handbook for teaching people how to bring Six Sigma breakthroughs to fruition.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
231
III. Evidence-Based Decisions, Inc. Six Sigma Black Belt/ Expert 16 Class Curriculum Outline Black Belt Core Study Texts and Free On-Line Resource book 1. Profit Signals, How Evidence-based Decisions Power Six Sigma Breakthroughs, M. Daniel Sloan and Russell A. Boyles PhD, Evidence-based Decisions, Inc., Sloan Consulting and Westview Analytics, 2003. 2. Getting to Yes, Negotiating Agreement Without Giving In by Roger Fisher and William Ury. (1991) ISBN 0-14-015735-2 3. Getting Ready to Negotiate, The Getting to Yes Workbook by Roger Fisher and Danny Ertel. (1995) ISBN 0-14-023531-0 4. How to Lie With Statistics, Darrell Huff, (1954), ISBN 0393-31072 5. Learning to See Lean Value Stream Mapping work book http://www.lean.org/Lean/Bookstore/ProductDetails.cfm?Sele ctedProductID=9 6. Engineering Statistics Handbook, Free PDF download on-line Internet Resource, http://www.itl.nist.gov/div898/ handbook/index.htm 7. Optional Show Stopper: Paper Flight, Complete, easy to follow instructions for making 48 different models that fly. by Jack Botermans. (1984) ISBN 0-8050-0500-5 Must Read, Master Black Belt, Bedrock Classics: 1. Economic Control of Quality of Manufactured Product, W. A. Shewhart. (1932) ASQ Quality Press, Milwaukee, Wisconsin. 2. Statistics for Experimenters, Box, Hunter and Hunter.(1978) ISBN 0-471-09315-7
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
232 Appendices Software Recommendations: Superior software is essential to breakthrough improvements and bottom line business results. Our course is published for students using Adobe Acrobat 5.0. Therefore, the de facto standard, Portable Document Format (PDF), is a requirement for printing, reading, note taking and electronic file attachments. JMP 5.0 http://www.jmpdiscovery.com/index.html As of September 2003, We believe this vector analysis program is the best in class. It is capable of handling virtually all of the analysis work required in Six Sigma breakthrough projects. Another application, Minitab, is also available. http:// www.minitab.com/ We happily accommodate customers who prefer this excellent application. Microsoft Excel. Six Sigma leaders must know how to use Excel and its add-ins. Crystal Ball by Decisioneering. http://decisioneering.com This multi-variate, financial simulation tool is superb for enlisting and retaining finance leader support. This tool can be an excellent guide for project selection. Quality America’s Excel SPC-IV add-in. http:// qualityamerica.com/ Ease-of-use and a short learning curve makes this program desirable for some executive champions. Our course is distinguished by the speed with which Black Belt candidates produce bottom line business results. The rigor and relevance of the course content are structured around the proven Six Sigma DMAIC cycle: Define, Measure, Analyze, Improve and Control. Course content covers the American Society for Quality’s (ASQ) Six Sigma Body of Knowledge and uses Bloom’s taxonomy of knowledge: Knowledge Black Belt Experts must be able to recognize terminology, definitions, ideas, principles and methods.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
233
Comprehension Experts must be able to understand tables, reports, diagrams, and directions. Application Experts must be able to apply principles, methods, and concepts on the job. Analysis Experts must be able to break down data and information. Statistical reasoning, analysis, and computing literacy are key. Synthesis Experts must expose unseen and informative patterns. Evaluation Black Belt Experts must be able to make judgments regarding the value of proposed ideas and solutions.
Black Belt Course Outline 1. Defining Six Sigma: Introduction, Overview, and History – A Six Sigma Gestalt 1.1. Learning Objectives: Theory and practice. 1.2. Introductions. 1.3. The 5-Minute PhD: Vector analysis applied to a data matrix. 1.4. The Complete Six Sigma Tool Kit: Categorical Catapult Experiment: 23 Designed Experiment (DOE), the ANOVA, Scatter Diagrams, Regression, Correlation, Histograms, Pareto charts, Control Charts, Inductive and Deductive reasoning. 1.5. Four Essentials in a thorough, 6 Sigma Analysis. JMP 5.0. (or Minitab 13) software navigation are introduced. 1.5.1. Calculate the Mean- Recognize that the mode and median exist 1.5.2. Calculate the Standard Deviation: s and sigma, σ. 1.5.3. Calculate Improbability – The F ratio ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
234 Appendices 1.5.4. Graph Data in meaningful ways that illustrate the mean, standard deviation and probability information. 1.6. Six Sigma: History, philosophy, goals and models. 1.6.1. The scientific method: Hypothesis, Experiment, Test Hypothesis. 1.6.2. PDSA or PDCA: Plan, Do, Study, Act or Plan, Do, Check, Act 1.6.3. The IDEA cycle: Induction, Deduction, Evaluation and Action. 1.6.4. DMAIC : Define, Measure, Analyze, Improve, Control. 1.7. Standards of Evidence: Evidence-based Profitability Principles. Vector Analysis applied to a Data Matrix 1.7.1. Analogy (1931-2003): Legal System Decisions 1.7.2. Analogy: Management System Decisions 1.7.3. Interactive Dialogue: Assessing evidence in your corporate culture. 1.7.3.1. Where you are today? 1.7.3.2. Where you want to be in your future? 1.8. An Enterprise View: Suppliers, Inputs, Process, Outputs and Customers 1.8.1 Y = f (X1, X2….Xn) 1.9. The Six Sigma Lucrative Projects Results Map 1.10. Lucrative Project Selection 1.10.1. Calculating the Priority Projects Using Excel matrix 1.10.2. Selecting and Leveraging Projects 1.10.3. Brainstorming 1.10.4. S.M.A.R.T. projects 1.10.5. Specific, Measurable, Achievable, Relevant, and Time Bounded. 1.10.6. Project Charters and Planning Tools Gantt and Performance Evaluation and Review Technique (PERT) Charts 1.11. Designed Experiment Homework. Every class participant will complete his or her first breakthrough project this evening. Results will be recorded and analyzed using JMP or Minitab for class presentations during in class 2. In class demonstrations are mandatory.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
235
2. Define: Organizational Responsibilities and Financial Six Sigma. 2.1. Homework Experiment Presentations using software 2.1.1. Designed experimentation demonstrations from home. Typically these are spread out through the entire day. By the end of the day people have memorized software keystrokes for either Minitab or JMP. Both are easy to master. Both yield identical answers. They are reliable as the sunrise and sunset. 2.1.2. Getting to Yes workbook reports. 2.1.3. How to Lie with Statistics reading assignments and discussion. 2.2. Learning Objectives 2.3. The Continuous Catapult Experiment: 23 Designed Experiment (DOE) 2.3.1. Accuracy and Precision. 2.3.2. Predicting the future with the Profiler. 2.4. Six Sigma Language, Leadership and Job Descriptions. “Kickin’ the heck out of variation,” led to the martial arts metaphor. 2.4.1. Executive 2.4.2. Champions 2.4.3. Master Black Belts 2.4.4. Black Belts 2.4.5. Green Belts 2.5. Linking Organizational Goals and Objectives to Six Sigma 2.5.1 What is different about 6 Sigma and other problem solving tools? 2.5.2. Closeda nd Open Loop Feedback Systems 2.5.3. SWOT analysis of Sub-optimizing systems. Class dialogue on cultural norms and issues related to this topic. 2.5.3.1. Strengths 2.5.3.2. Weaknesses 2.5.3.3. Opportunities 2.5.3.4. Threats 2.6. The New Management Equation - Old Equation Comparison 2.7.Profit Signals workshop to Include Chief Financial Officers and/or Controllers 2.7.1. Crystal Ball budget building Decisioneering Tutorial Review 2.7.2. Futura Apartments Tutorial 2.7.3. Vision Research Tutorial 2.7.4. Hands on corporate example and demonstration ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
236 Appendices 2.8. Project Documentation: Data, analysis, and evidence do not speak for themselves. Visualize and plan your breakthrough project presentation. 2.8.1. Spreadsheets 2.8.2. Story Boards 2.8.3. Phased Reviewed 2.8.4. Management Reviews 2.8.5. Executive Team Presentations 2.8.6. Homework. Design, build, and fly paper airplanes according to your experimental array with your team. Begin building a Crystal Ball model related to potential Six Sigma projects. 3. Defining Six Sigma Project Selection and Benchmarking 3.1..Paper Airplane Homework Presentations 3.1.1 A Complete Six Sigma Pilot Project – Synectic Experiment Paper Flight, Complete, easy to follow instructions for making 48 different models that fly, Jack Botermans. (1984) ISBN 0-8050-0500-5 3.1.2. Debriefing, Analogies, and Analysis. 3.1.2.1. Intuitive and counter-intuitive solutions. 3.1.2.2. Iterative learning and fun. 3.1.3. Project Timelines 3.1.4. Statistical Software Application practice. 3.2. Learning Objectives 3.2.1. A Catapult 25 DMAIC Experiment. 3.2.2. Predicting the Future with categorical and continuous variables. 3.2.3. Profiler: Optimization and Desirability 3.3. The 5 Whys 3.4. Six Sigma is a Business Initiative NOT a quality initiative. The American Society for Quality’s Black Belt test is discussed. As of 2003, there were no questions related to vector analysis or the data matrix. Consequently, we cover the entire list of recommended tools.. 3.5. Negotiation techniques for Success: Getting to Yes. 3.5.1. Practical Applications. 3.5.2. Wise, Efficient, Build Relationships and BATNA 3.5.3 10 Principles for Getting to Yes 3.5.4. Dialogue discussion. 3.6. SIPOC Diagrams (Supplier, Inputs, Process, Outputs, and Customer) 3.6.1. Brainstorm and draw one. 3.6.2. S.M.A.R.T. Projects and the SIPOC diagram. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
237
3.6.3. Specific, Measurable, Achievable, Relevant, and Time Bounded. 3.7. Project Charters and Paper Work. 3.7.1. Process Characterization and Optimization. 3.7.2. Brainstorming Critical To Quality Flight Standards 3.7.2.1. Voice of the Customer (VOC) 3.7.2.2. Ground Rules for Nominal Group Technique 3.8. Benchmarking – Process Elements and Boundaries. DMAIC is what your customers expect to see. Frame your reports accordingly. 3.8.1. Defining - Design for Six Sigma 3.8.2. Measurement – Performance Metrics and Documentation 3.8.3. Analysis: Mean, Standard Deviation, Probability, Graph 3.8.4. Improvement 3.8.5. Control 3.8.6. Internal Best Practices using the complete Six Sigma tool kit. 3.8.7. Comparing Machines, Production Lines, Plants, and Shifts 3.8.8. Plant Visits and interviews. 3.8.9. Literature Searches: Internet and Company. 3.8.10. Independent Evaluations and public Financial Reports. 3.8.11. Product Tear Downs and published books. 3.9. Textbook DMAIC breakthrough Case Study presentation. 3.10. Project homework and reading assignments set. 4. Defining Process and System Capabilities 4.1. Review of Homework and Reading 4.2. Learning Objectives 4.3. The Complete Six Sigma Tool Kit: Vector Analysis Applied to a Data Matrix. Repetition for mastery using candy M&M Sampling, Sorting, and Analyzing. Hands-on Define, Measure, and Analyze Experiments. Practice with JMP 5.0 Software Application 4.3.1. Populations versus Samples 4.3.2. Operational Definitions – Critical to Quality Characteristics 4.3.3. Flow Diagramming the Production Process 4.3.4. Sampling our population of candy. 4.3.5. Histograms ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
238 Appendices 4.3.6. Pareto Charts 4.3.7. Control Charts 4.3.8. Scatter Diagrams and Correlation Coefficients 4.3.9. 23 Designed Experiment: Comparing the value of systematic observation with simple arithmetic counts. Understanding the context of multiple variables is the key to breakthrough improvement projects. 4.4. The DMAIC Breakthrough Chart 4.4.1. Juran’s Trilogy 4.4.2. Shewhart’s P-Chart 4.5. Defects Per Unit 4.5.1. Calculating Defects per Million (DPU) Opportunities 4.5.2. Motorola’s classic, proof reading example. 4.6. Six Sigma Values. 4.7. Cp and Cpk 4.7.1. Practical Applications using Dice 4.7.2. JMP 5.0 or Minitab Calculation Practice 4.7.3. Confidence Interval introduction. 8 4.8. 2 Helicopter Designed Experiment 4.8.1. Emphasis of key concept. Compare M&Ms Enumerative Sampling with two-level, eight factor DOE analytic sampling. 4.9. Project Selection Focused Homework on Process Capability 4.9.1. Calculate and graph Cpk for all 16 copters. 4.9.2. Calculate and graph Cpk for select individual copters. 4.10. Homework: Read Quality Function Deployment white papers for report. 4.10.1. Project selection updates including Crystal Ball model. 4.10.2. Present results of tools applied in daily work. 5. Define: Negotiation, Quality Function Deployment, and Data Mining Training 5.1. Homework reports, presentations, and dialogue. 5.2. Learning Objectives: Vector Analysis applied to a data matrix and Evidence-based decisions. 5.3. Observe Designed Experiments DMAIC Demonstrations by Students 5.3.1. Mean, Standard Deviation, Probability, and Graphed Results. 5.4. Helicopter 23 Confirmation Experiment ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
239
5.4.1. Iterations and efficient learning. 5.4.2. How does this analogy apply to your work? 5.5. Change Agents and Team Leadership: Pythagoras, Aristotle, to Frederick Douglas and Harriett Tubman to 2004. 5.5.1. Cultural Influences 5.5.2. Innovation Adoption Model 5.5.3. Diffusion of Innovation 5.5.4. Adoption Process 5.5.5. Force Field Analysis – Forces Fighting Change 5.5.6. Change Agent Methods 5.5.7. Understanding and overcoming Road blocks 5.5.8. Negotiation – Getting to YES. 5.5.9. Motivation 5.5.10. Communication 5.6. Building a House of Quality - A One proven method of encouraging concurrent engineering. 5.6.1. Using and Excel Template 5.6.2. Functional Requirements and Robust Design 5.6.3. Design for X (DFX): Design Constraints, design for manufacturability, design for test, design for maintainability. 5.6.4. The Whats 5.6.5. The Hows 5.6.6. “Correlation matrix” Trade Offs 5.6.7. The Four Phases, The Four Houses of Quality. 5.6.8. KANO Model of Quality 5.7. Excel Data Sorting Function – A brief history of data mining. 5.7.1. Orthogonal Arrays 5.7.2. Homogeneous Fields and Records, Columns and Rows 5.7.3. 23 DOE Data Mining Demonstration and practice. 5.7.4. A correct vector analysis: Thorough 6 Sigma Analysis: The average, standard deviation, probability, and analytic graph. 5.8. Homework: Outline project selections for class presentation. 5.8.1. Bring data in spreadsheet formatted for data (sorting) mining practice.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
240 Appendices 6. Measuring Value: Rolled Throughput Yield Metrics, & Costs of Quality 6.1. Homework presentations and review of data mining strategy. 6.2. Learning Objectives 6.3. The 24 Quincunx Machine Experiment for JMP or Minitab practice. 6.3.1. Observe the machine. 6.3.2. DMAIC Comprehensive definition process to determine variables and outcomes.. 6.3.3. Central Limit Theorem Simulations using the machine and Decisioneering’s computerized demonstration model. 6.3.4. Uncovering the “Hidden Factory” 6.4. Drawing the Value Stream - Lean Flow Charting Fundamentals 6.4.1. Drive Down Costs Red Bead Sampling Game – Drive Down Costs 6.4.2. Interactive role playing using a game of historical significance. 6.5. Rolled Throughput Yield (RTY) 6.5.1. Poisson Computer Simulation on Defects per Unit 6.6. Thought Process Mapping 6.6.1. Categorical Thinking 6.6.2. Universal Standards of Measurement. 6.7. Critical to Quality Tree 6.7.1. Identifying Critical to Quality Characteristics (CTQ) 6.7.2. Customer needs, Drivers, Quantified CTQ 6.7.3. Relevant to business in financial, quality, and productivity terms. 6.8. Collecting, Sorting, Developing and Translating Customer Information 6.8.1. Surveys: Telephone, mailing, interview 6.9. Brainstorming 6.10. Cause and Effect Diagrams 6.11. Affinity Diagram Experiment 6.12. Costs of Poor Quality 6.12.1. Internal Failures 6.12.2. External Failures 6.12.3. Appraisal Costs 6.12.4. Prevention Costs 6.12.4.1. Detailed walk through of an exemplary Cost of Quality corporate report. Excel Spreadsheet template available. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
241
6.12.5. Quality Cost Statement by Product Line 6.12.6. Taguchi Loss Function Example 6.12.7. Phillip Crosby’s Rule of 3 6.13. Homework Focus on Quality Costs Project Results Measure: Process Mapping 7.1. Introduction 7.1.1. Workshop Purpose and Agenda 7.1.2. Learning Objectives 7.1.3. Homework Reports 7.2 Process and System Concepts 7.2.1. Process Model (SIPOC) 7.2.2. Why use a process model? 7.2.3. Systems and Processes. 7.2.4. Systems Thinking 7.2.5. Definitions 7.2.6. Process Categories 7.2.7. Goals of Process Design 7.2.8. A Primary Objective 7.2.9. What makes a process reliable? 7.2.10. “Global Process Requirements 7.3. Documenting Processes 7.3.1. Why be concerned with information? 7.3.2. What is this common process? 7.3.3. Why document a process? 7.3.4. Structure your information 7.3.5. Balance document needs 7.3.6. Document design rules 7.3.7. A documentation survey tool 7.4. Techniques for process mapping 7.4.1. What is process mapping? 7.4.2. Why use process maps (flow diagrams)? 7.4.3. The mapping method 7.4.4. Define the process 7.4.5. Process model revisited 7.4.6. A Process Definition Tool 7.4.7. Define the Process 7.4.8. What is your purpose? 7.4.9. Process Customers 7.4.10. Process Boundaries 7.4.11. Outline for process definition 7.4.12. Exercise in process definition 7.4.13. Flow charting the primary process 7.5.14. What is a parallel process? 7.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
242 Appendices 7.4.15. Adopt and use standard symbols 7.4.16. Other useful symbols 7.4.17. Example 7.4.18. Writing good narrative 7.4.19. Exercise: Primary Process 7.4.20. Flow charting alternative paths 7.4.21. Example 7.4.22. Exercise: Alternative paths 7.4.23. Add control points 7.4.24. Example 7.4.25. The decision question 7.4.26. Controls: Some considerations 7.4.27. Exercise: Control Points 7.4.28. Responsibility matrix 7.4.29. Exercise: Define responsibilities 7.5. Using alternate formats for process mapping 7.5.1. Types of maps 7.5.2. Simple flow chart 7.5.3. Top-down flow chart 7.5.4. Cross-functional flow chart 7.5.6. PERT chart 7.5.7. Decision tree 7.5.8. Data flow diagram 7.5.9. Geography flow diagram 7.5.10. Standardized Process Chart 7.5.11. Exercise: Remap 7.5.12. Finish the flow chart 7.5.13. Characteristics of a good flow chart 7.5.14. Key implementation points 7.6. Using maps to improve and streamline processes 7.6.1. Goals of process analysis 7.6.2. Elimination targets: waste, rework, delays, reverse loops, and needless complexity. 7.6.3. Technique #1: Value assessment 7.6.4. Technique #2: Standardize 7.6.5. Technique #3: Using the map 7.6.6. Technique #4: Early control 7.6.7. Technique #5: Prevention 7.6.8. Technique #6: Analyze inputs 7.6.9. A process analysis tool. 7.7.10. Homework 8. Measure: The Productive Team Member 8.1. Introduction 8.1.1 Workshop purpose and agenda ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
243
8.1.2. Learning objectives 8.1.3. Homework report (six sigma project progress) 8.2. Characteristics of Effective Teams 8.2.1. Box Of Stuff exercise 8.2.2. Teams vs Groups 8.2.3. What things the Team must Manage 8.2.4. Inputs for a Successful Team 8.2.5. Sense Of Urgency—Good Or Bad? 8.2.6. Outputs of a Successful Team 8.2.7. Stages of Team Development 8.2.8 (Murder Mystery exercise) 8.2.9 Four Stages of Team Development 8.2.9.1. Norms and Team Development 8.2.9.2. A Sample “Code of Cooperation” 8.2.9.3. Overcoming Hindrances to Team Performance 8.2.9.4. Circle In The Square exercise 8.2.9.5. Competition versus Cooperation 8.2.9.6. Signs of Team Trouble 8.2.9.7. Groupthink 8.2.9.8. Ground Rules for Consensus 8.2.9.8. Five Approaches To Getting Unstuck 8.2.9.9. Team Roles and Responsibilities Meeting Management and Leader skills 8.2.9.10. Member role and responsibilities 8.2.9.11. Improving Communication 8.2.9.12. The Johari Window 8.2.9.13. Communication Model 8.2.9.14. Types of Feedback 8.2.9.15. Practicing Feedback 8.2.9.16. Building “I-Statements” 8.2.9.17. Communication Breakdown 8.2.9.18. How To Correct Bad Listening Habits 8.2.9.19. Barriers to Good Listening 8.2.9.11. Learning Style Inventory 8.2.9.12. Strategies for Managing Change 8.2.9.13. Principles of Large-System Change 2.2.9.14. Change vs. Transition 8.2.9.15. External Forces For Change 8.2.9.16. Internal Forces For Change 8.2.9.17. The Four Room Apartment 8.2.9.18. Improving Team Performance 8.2.9.19. Team Self-Evaluation 8.2.9.20. Close
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
244 Appendices 8.2.9. Assign Homework (Six Sigma Project Progress using appropriate tools) Visit http://www.fmeca.com/ Read as much as you can prior to the next class. 9. Measuring the Process: Failure Mode Effects Analysis FMEA Workshop. Criticality is included and emphasized. 9.1. History. 9.2. Definitions and Acronyms. 9.3. Walk through of the entire FMEA process will include group work tools and methods introduced in effective team member class. 9.4. Review output of product. 9.5. Homework: Present project progress and estimated dollar savings using tools. 10. Analyze: Exploring, Summarizing, and Predicting using data. 10.1. Homework Review – Focus on Project Financial Results 10.2. Learning Objectives 10.2.1. Be able to give examples of continuous, categorical, count, pass/fail, and life data. 10.2.2. Use correct graphics to summarize measurement data. 10.2.3. Use software to explore data basesdatabases. 10.2.4. Explain central limit theorem using coin tosses. 10.2.5. Fit a normal distribution to measurement data and assess goodness of fit. 10.2.6. Review process capability concepts with working exercise. 10.3. Review: Types of Data 10.3.1. Traditional taxonomy 10.3.1.1. Nominal, ordinal, interval, ratio 10.3.2. More useful modern taxonomy 10.3.2.1. Attribute = categorical = discrete = nominal 10.3.2.2. Ordinal 10.3.2.3. Continuous = measurement = parameter = variable 10.3.2.4. Time to failure = life data 10.4. Collecting, Recording and Analyzing Measurement Data 10.4.1. Real-world examples 10.5. Review of graphics for measurement data 10.5.1. Stem and leaf diagram ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
245
10.5.2. Frequency histogram and Cumulative Distribution Function (CDF) 10.5.3. Boxplots 10.6. Review of descriptive statistics for measurement data 10.6.1. Minimum, maximum and range 10.6.2. Mean and standard deviation 10.6.3. Plus and minus three standard deviations 10.6.4. Central Limit Theorem, coin tosses and Process Capability 10.7. JMP Data Exploration Exercises 10.7.1. Entering data in rows and columns 10.7.2. Generating descriptive statistics and graphics 10.7.3. Producing a report: Integrating with Microsoft Word, Excel, and Power Point 10.8. Workshop: Wooden sticks, calipers, data entry, and worker variation. 10.8.0.1. Yield calculations for one-sided specs 10.8.0.2. Yield calculations for two-sided specs 10.8.0.3. Cumulative or “rolled throughput” yield (Review and Reinforcement) 10.8.0.4. Gauge Reproducibility and Repeatability Studies and Practice 10.9. Homework Focused on Project Results 11. Analyze: Inductive Reasoning Part 1 – Quantifying uncertainty in measurement systems (Formerly known as Hypothesis Testing) 11.1. Homework Review Focused on Project Results 11.2. Learning Objectives 11.2.1. Explain relationships between processes and populations. 11.2.2. Identify default statistical models for measurement, pass/fail, count and life data. 11.2.3. Express real-world problems in terms of statistical models and population parameters. 11.2.4. Use Confidence Intervals to characterize or test a process in terms of mean, standard deviation, three sigma limits, capability indices, fraction defective or reliability. 11.3. Measurement Systems 11.3.1. Definition of a measurement system 11.3.2. Population sampling 11.3.3. Process Sampling 11.3.4. Measurement objectives
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
246 Appendices 11.4. Measurement Uncertainty 11.4 1. The Normal, binomial, Hyper-geometric, Poisson, and Weibull distribution models 11.4.2. Accuracy and precision (Review and Reinforcement) 11.4.3. The role of calibration procedures 11.4.4. Repeatability and Reproducibility (R&R) 11.4.4.1 Repeatability: dependability of the gauge 11.4.4.2 Reproducibility: dependability of gauge operators and environment 11.4.5. Examples and exercises: Calibration and Calibration Control: Penny for your Thoughts workshop exercise. 11.5. The Seven Habits of Highly Statistical People: Quantifying Uncertainty. 11.5.1. Statistical Inference 11.5.3. The law of likelihood and likelihood function. 11.5.4. Interval Estimation 11.5.5. Confidence and Evidence 11.6. Characterizing and Testing exercises 11.6.1. The “one-sided” fallacy. 11.6.2. Pass/Fail 11.6.3. Interpreting Opinion Polls 11.6.4. Sample Size Calculations 11.7. Chi Square and t distributions 11.8. Homework Focused on Project Results 12. Analyze – Inductive Reasoning Part II 12.1. Homework Review Focus on Project Results 12.2. Learning Objectives 12.2.1. Recognize statistical problems when they occur, and be able to classify them as testing an objective, comparing processes, or relating variables. 12.2.2. Identify appropriate null hypotheses for testing an objective, comparing processes, and relating variables. 12.2.3. Choose appropriate test procedures based on type of problem and type of data. 12.2.4. Use p-values to interpret the results of statistical tests. 12.2.5. Explain the difference between correlation and regression. 12.3. Statistical Hypotheses and Process Hypotheses 12.3.1. The null hypothesis. 12.3.2. Fair coin tosses. 12.3.3. P-values ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
247
12.3.4. Z statistics and the Z transformation 12.3.5. P values from Z statistics 12.3.6. P values from z or Chi squared distributions. 12.4. ANOVA The geometry of analysis 12.4.1. Likelihood ratio 12.4.2. Degrees of Freedom 12.4.3. P Values from the F statistic 12.5. Relating variables. 13. Analyze - Model Building, Data Mining, and Linear Regression 13.1. Homework and Six Sigma Project Progress Review 13.2. Quantifying the Strength of Evidence 13.2.1. Hypothesis testing revisited 13.2.2. Law of likelihood in comparison problems 13.2.3. Confidence interval for a difference 13.2.4. P-values 13.2.4.1. Mathematical definition 13.2.4.2. Operational interpretation 13.3. Sample Size Calculations 13.3.1. Smallest difference of practical significance 13.3.2. Power of detection 13.3.3. Example: comparing two opinion polls 13.4. Pass-Fail Data 13.4.1. Likelihood ratio test for equality of two or more Binomial proportions 13.4.2. Test for equality of two or more Binomial proportions (valid only for large sample sizes) 13.4.3. z test for equality of two Binomial proportions (valid only for large sample sizes) 13.5. Number of Defects 13.5.1. Likelihood ratio test for equality of two or more Poisson means 13.5.2. Test for equality of two or more Poisson means (valid only for large sample sizes) 13.5.3. z test for equality of two Poisson means (valid only for large sample sizes) 13.6. Continuous Measurements 13.6.1. F test for equality of two Normal standard deviations 13.6.2. t test for equality of two Normal means 13.6.3. F test for equality of two or more Normal means (Analysis of Variance) (valid only if all standard deviations are the same) ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
248 Appendices 13.7. Life Data (Time to Failure) 13.7.1. Likelihood ratio test for equality of two or more Weibull distributions 13.8. Chi-square tests 13.8.1. Interpreting the table of the Chi-square distribution 13.8.2. Tests of association in contingency tables 13.9. Workshop: Pennies for Your Thought 13.10. Regression Analysis 13.10.1. Scatter Diagrams (Review and Reinforcement) 13.10.2. Correlation is not causation 13.10.3. Linear Regression Models 13.10.3.1. “All models are wrong, some are useful.” 13.10.3.2. Straight-line regression 13.10.3.3. Multiple regression 13.10.3.4 Polynomial regression 13.10.4. Fitting Regression Models 13.10.4.1. The least squares estimates 13.10.4.2. The RMS error 13.10.4.3. Testing for significance of predictor variables 13.10.4.4. Predicted mean values 13.10.4.5. Confidence intervals for predicted mean values 13.10.4.6. Confidence intervals for predicted individual value 13.10.5. Regression diagnostics 13.10.6. The dangers of R2 13.10.6.1. Testing for lack of fit 13.10.6.2. Residual plots 13.10.6.3. JMP exercises 13.10.7. Workshop: Pennies for Your Thought 13.10.8. Homework with Project Focus 14. Improve – Experimental Design and Analysis 14.1. Learning Objectives 14.1.1. Be able to explain the difference between optimization and screening experiments. 14.1.2. Calculate sample sizes for optimization experiments. 14.1.3. Create matrices for optimization experiments. 14.1.4. Analyze data from optimization experiments. 14.1.5. Interpret and apply results from optimization experiments. 14.2. Homework Review Focused on Projects 14.3. Introduction to Experimentation ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
249
14.3.1 Why should I do experiments? 14.3.2 When should I do experiments? 14.4. Concepts and Definitions 14.4.1. DOE Terminology 14.4.2. Experimental Unit 14.4.3. Sample Size 14.4.4. Response 14.4.5. Factor 14.4.6. Level 14.4.7. Design Point 14.4.8. Design matrix 14.5. Types of factors 14.5.1. Continuous 14.5.2. Categorical 14.5.3. Control 14.5.4. Noise 14.6. Do not experiment with one factor at a time! (OFAT Review and Reinforcment) 14.7. Design principles 14.7.1. Bold strategy 14.7.2. Factorial structure 14.7.3. Control group 14.7.4. Replication 14.7.5. Randomization 14.7.6. Blocking 14.8. Experiments with All Factors at Two Levels 14.8.1. Examples 14.8.2. JMP Steps 14.8.3. Exercises 14.9. Basic Design Process 14.9.1 JMP Steps 14.9.2 Exercises 14.10. Screening Experiments 14.10.1. Examples, Modified Design Process 14.11. Workshop: The Funnel Process 15. Improve - Process Optimization and Control 15.1. Learning Objectives 15.1.1. Describe iterative strategy for experimentation 15.1.2. Perform multiple response analysis. 15.1.3. Calculate sample sizes for robust optimization experiments. 15.1.4. Create matrices for robust optimization experiments. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
250 Appendices 15.1.5. Analyze data from robust optimization experiments 15.1.6. Understand common cause and special cause variation. 15.1.7. Describe a Reaction plan to out of control conditions. 15.2. Review of Designed Experiments Homework 15.3. The Process of Experimentation 15.3.1. The experimental cycle 15.3.2. Types of experiments 15.3.3. Strategies for experimentation 15.4. Statistical Modeling 15.4.1. Standard assumptions 15.4.2. The method of least squares 15.4.3. Models for continuous factors 15.4.4. Models for categorical factors 15.5. Statistical Testing 15.5.1. Testing model coefficients 15.5.2. Testing for lack of fit 15.5.3. Exercises 15.5.4. Predicted values and residuals 15.5.5. Exercises 15.6. Multi-level Optimization Experiments 15.7. Response surface analysis 15.7.1. Quadratic models for continuous factors 15.7.2. Continuous × categorical interactions 15.7.3. Example and JMP exercises 15.8. Design process 15.8.1. Quadratic models for continuous factors 15.8.2. Continuous × categorical interactions 15.8.3. Sample size calculations 15.8.4. Example and JMP exercises 15.9. Workshop: the Funnel Process using robust optimization and quality control, process improvements 15.10. Homework: Design of Experiments Project Focus: Report project results in DMAIC format for final class. 15.11. 16. Control - Optimization Experiments and Statistical Process Control 16.1. Learning Objectives 16.1.1. Multiple response analysis. 16.1.2. Rational sub-grouping. 16.1.3. Establishing baselines 16.1.4. Monitoring low failure rates. ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
251
16.1.5. Multivariate statistical process control 16.1.6. Short-Run SPC 16.2. Review of Designed Experiments Homework 16.3. Multiple Response Optimization 16.3.1. Example 16.3.2. “Multiple responses” is the rule, not the exception 16.3.3. Optimizing one response at a time will not work 16.4. Desirability functions 16.4.1. The three types of response objective 16.4.2. Constructing a desirability function for each response 16.4.3. Constructing the overall desirability 16.4.4. Maximizing the overall desirability 16.4.5. Optimizing over subsets of the design region 16.4.6. JMP exercises 16.5. Robust Optimization Experiments 16.5.1. The concept of robust optimization 16.5.1.1. Optimize the mean 16.5.1.2. Minimize the variance 16.5.1.3. Examples 16.5.2. Strategy for design of robust optimization experiments 16.5.2.1. Identify key noise variables 16.5.2.2. Define noise factor 16.5.2.3. Include noise factor in the design 16.5.3. Strategy for analysis of robust optimization experiments 16.5.3.1. Apply multiple response technique 16.5.3.2. Maximize overall desirability 16.5.3.3. Minimizes variability for a given mean 16.5.3.4. Seeks best combination of close-to-target mean and low variability 16.5.4. Statistical Process Control as a mind set and strategy. 16.5.4.1. Acceptance sampling and broken promises. 16.5.4.2. Hands on SPC experiments and software practice Workshop: the Funnel Process Exercises Summaries for Quick Reference 16.5.5. Thought process for designing an experiment 16.5.6. More on Sample size calculations 16.6. Homework: Design of Experiments Project Focused Report project results in DMAIC format.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
252 Appendices
IV. Profit Signals Production Notes We consciously chose to demonstrate Senior Master Black Belt, Six Sigma level knowledge and skills in every aspect of the production of this book. Our lean production system included two authors. When necessary, we retained the illustration services of expert contractors, just-in-time. We produced the electronic versions of our book independently. Kinko’s prints the four color cover, perfect bound paperback version on demand in a pull-production system. We carry only the inventory we need for personal, corporate use. The Internet, computing power, and software allowed us to complete the entire writing and production of the book in 90 days. This is the classic Six Sigma project time line. We began by creating the Profit Signals title on June 18. We completed the work in PDF format on September 11, 2003. Though it was an entirely Chance coincidence, the completion of our book on this day was an appropriate way to celebrate liberty, freedom of speech, equality, applied science, democratic values, art and the pursuit of happiness. Evidence-based decisions are as important to world peace as they are to prosperity. The applications that played primary roles are as follows: • Microsoft Word® 2000 and 2002 were our primary composition tools. • JMP 5.0® statistical software, manufactured by SAS, was our favored analytic program. We also use Minitab with clients who have that standard. • Microsoft Excel® was used for spreadsheet screen captures and some graphics. Using Excel for data matrix vector analysis shows the amount work required before a spreadsheet behaves like a reliable, rules-driven, software analysis application.
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Appendices
253
• Microsoft Explorer was the web browser we used for Internet research. • Microsoft Power Point® was frequently used by Russell for first draft, technical drawings. • Adobe Illustrator® 10 transformed all illustrations into EPS files for production. •Adobe Photoshop® 7.0 was used for certain photographic and graphic illustrations. • Adobe Acrobat® 6.0 Professional helped us disseminate copies for review. • Adobe In Design® 2.0.2 allowed us to design, layout and construct our book. • Crystal Ball by Decisioneering® was the Excel addin we used to make this spreadsheet behave like a data matrix. • Process Model®, a data matrix based flow diagramming program, was used to create flow diagrams. • Quality America ‘s Excel add-in, Statistical Process Control program was used to produce a control chart. • Dell desktop and laptop computers, a HewlettPackard LaserJet 1300 and an hp officejet v40xi jet printer produced hard copy for old fashioned proof reading and review. In our opinion, it is not only a reasonable expectation for Black Belts, Master Black Belts and Executive Champions to use a similar list of programs in their daily work, it is essential to Six Sigma powered project breakthroughs. Profound thanks are due to our wives, Lynne and Michelle, and our wonderful children, Austin and Molly. Patience is their virtue. We love them. In addition to the entire Adobe products technical support team, four individuals went well beyond the call of duty as we ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
254 Appendices produced Profit Signals. Jack Benham introduced us in July of 2002. Good on ya’ matey. Onwards and upwards. Without Jack’s vision, leadership and masterful management skill there would be no Profit Signals. Cheryl Payseno, our friend, colleague, nurse, and former hospital administrator volunteered her case study on Breaking the Time Barrier. She also encouraged us to tackle the cost accounting variance and break-even thinking head on with the Premise’s second illustration, Figure 2. Our friend, colleague, and final copy proof-reader Bethany Quillinan stepped into the fray to help us see our words through yet another set of eyes. She did a Six Sigma quality job on a pressure packed deadline. Finally, our friend Bill Moore, the President of MedCath, Incorporated, Hospital Division, volunteered invaluable editorial support. The specificity of his constructive criticisms and the solutions he proposed strengthened the quality of our work immeasurably. So, “Thank you very, very, very much Lynne, Austin, Michelle, Molly, Jack, Cheryl, Bethany and Bill.”
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Index
A Adams, John 217 Aladdin 179 analogy 44, 113, 153, 154, 174 analysis 9, 10, 13, 21, 33, 51 Analysis of Variance 37, 53, 58, 222 ANOVA 37, 55, 88 Archimedes 46 Aristotle 152, 161, 175, 221
B Bamboozle 56, 57 belt grinding 142 Bernstein, Peter L. 191 Black Belt 19, 37, 76, 79, 90 Box, George E.P. 115, 153 break-even analysis 11, 179
C CABG 34, 136 Calder, Alexander 165 Case Studies 21, 117 Cohen, Bernard 91 Confidence Level 81, 132 control chart 110, 203, 223 cornerstone of evidence 10, 14, 92, 118, 208 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
256 Index correlation 177 Corrugated Copters 22, 23, 194 cost-accounting variance analysis 11, 12, 15, 21, 53, 119, 177, 179 Cost of Poor Quality 111 Cpk 114, 194 credulity 50 critical thinking 57 Critical to Quality 99 CTQ 99 cube 38 cynicism 118
D Darwin, Charles 178 data matrix 9, 11, 14, 16, 18, 19, 20, 21, 22, 29 data matrix geometry 124 da Vinci, Leonardo 43 defects per million 92 degrees of freedom 65, 167, 168 Delusions 56 Design of Experiments 44, 48 differences 32, 53, 54, 74 Disney, Walt 47, 178 Disraeli 118 DMAIC 21, 96, 130
E Einstein, Albert 43, 46, 91 Einthoven, Willem 33 EKG 33 emergency department 128 Emerson, Ralph Waldo 222 Euclid 46 evidence-based decision 10, 13, 18, 23 Executive Committee 209
F Fads and Fallacies 85 Failure Mode Effects Analysis (FMEA) 112 feedback 136 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Index
257
Feigenbaum, Armand V. 110 Feynman, Richard P. 194 fields 59 fingerprints 178 Fisher, Ronald A. 43 Five-Minute PhD 20, 152
G G. Charter Harrison 179 GAAP 55 Galileo 223 Galton, Francis 55, 178 Galvin, Robert 87 Gantt, Henry L. 97 Generalization 9, 19, 31, 178 generalization 9, 16 Generally Accepted Accounting Principles 57 General Electric 110, 223 George E.P. Box 21 Gosset, William 32 Gould, Stephen Jay 177 Guinness 32
H Harrison, G. Charter 179 Hidden Factory 218 hidden factory 108, 111 Hill, Sir Austin Bradford 135 Huff, Darrell 52 Hunter, William 74 Hunter, J. Stuart 74 hyperspace 38, 39, 43, 60
I Imagineering 31
J JCAHO 128 Jefferson, Thomas 217 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
258 Index JMP 131 Joint Commission 128
K Kaizen-blitz 145 Keats, John 50 knowledge 36, 39
L law of the universe 9, 66, 81, 158 lean 108 Length Of Stay 130
M Mackay, Charles 56 main effect 39 Marconi 43 math phobia 220 Matreshka 106 Maxwell, James Clerk 91 measurements 9, 19, 153 Michelangelo 43 Minitab 88 Motorola 223 multiplication 15
N NASA 194 Netter, Frank 33 Newton, Isaac 50 New Management Equation 63, 161, 175 Normal distribution 70 n dimensions 31
O OFAT 103
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
Index
259
P P-value 81 p-value 72, Paine, Thomas Paper Bags 14 Pareto chart 127, 140, 171 perpendicular planes 41 PERT 97 Picasso, Pablo 37 predicted values 11, 44, 165 process capability 113 process maps 94 Profit Signals 44 Pythagoras 21 Pythagorean Theorem 13
Q quarterly review 207
R reasoning 81 records 123 refraction 50 regression modeling 177 Ronald Fisher 9, 12 Rothamsted 32 Russian dolls 106
S Sagan, Carl 162 sample size 59 sample standard deviation 64 scientific management 13 Sculpey Clay 10, 164 Shewhart, Walter A. 209 Simulation 100 SIPOC 155 Sisyphus 180 Six Sigma 18, 89 Six Sigma theory 19 ©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003
260 Index Six Sigma tools 19 Smith, Bill 18 spreadsheet 80 spreadsheet analysis 14 standards of evidence 20, 21, 49, 160 Stories 50 straight-line prediction 181 straw man 76 strength of evidence 83
T Taylor, Frederick W. 84, 103 tetrahedron 10, 152, 165 Themis 83 Three Rs 23 Transparency 13 Turrell, James 47 Twain, Mark 59
V variation 10 vector 10, 60, 152 vector analysis 9, 10, 13, 70, 76, 158, 171, 208
©
M. Daniel Sloan and Russell A. Boyles, All Rights Reserved, 2003