Six Sigma and The University: Teaching, Research, and Meso-Analysis

SIX SIGMA AND THE UNIVERSITY: TEACHING,
RESEARCH, AND MESO-ANALYSIS
DISSERTATION
Presented in Partial Fulfillment of the Requirements for
The Degree Doctor of Philosophy
In the Graduate School of The Ohio State University
By
James E. Brady, B.S., M.S., MBA, P.E.
The Ohio State University
2005
Dissertation Committee: Approved by
Dr. Theodore T. Allen, Adviser

Graduate Program in Industrial
Dr. Allen R. Miller and Systems Engineering
Dr. Clark A. Mount-Campbell

ABSTRACT
Six Sigma was introduced by industry practitioners and consultants as a means
to improve any given company’s competitive position. Its acceptance by industry has
been widespread over the past two decades, yet academic research on Six Sigma has
been surprisingly limited. Further, most of the research has been focused on the tools
and statistical techniques used in Six Sigma. Its relationship with university activities
including teaching, research, and service is not clear. The purpose of this dissertation
is to explore selected aspects of the relationship between Six Sigma and universities
more fully. In doing so, there is an attempt to answer these fundamental questions:
(i) What is Six Sigma? (ii) What roles can academics usefully play in relation to Six
Sigma? and (iii) How can academia help companies to better use the new project
related data sources created by Six Sigma. Results here divide into three chapters.
First, the literature on Six Sigma is reviewed and synthesized. This includes detailed
descriptions of research trends with an emphasis on establishing its relationship to
quality management theory and topics for future research. Secondly, case base training
is examined as a method to improve Six Sigma education and increase usage on the
job among university student learners. Third, with Six Sigma’s emphasis on
management by data and project based data collection, industry is starting to
ii
accumulate many large databases of “meta-data” concerning the successes or failures
of individual quality improvement projects. We propose methods specifically for
making use of the project data and illustrate their application using 39 case studies
form a mid-western manufacturing firm.
iii
ACKNOWLEDGMENTS
Reflecting on my long and continuing education process, I find it difficult to
adequately acknowledge the incredible support I have received form so many people
I am sincerely grateful to my advisor, Dr. Allen, for all of his continuous assistance
and direction. I would like to thank my other committee members: Dr. Mount-
Campbell, and Dr. Miller as well as Chaitanya Joshi who helped in the early
formulation of some of the goals. I appreciate Lucent Technologies, Inc. and LaBarge,
Inc for the support they provided during my recent studies. I owe a debt to my Aunt
Joan for instilling in me the value and love of learning, my children and grandchildren
for the sacrifice of their time. Finally, I would like to thank my wife. This dissertation
is dedicated to her endless encouragement and support over these many years.
iv
VITA
1951 ……………………………………………… Born – Tucson, Arizona
1974 ……………………………………………… BS Aeronautical Engineering

Wichita State University
Wichita, Kansas
1974-1976 ……………………………………….. US Air Force

W-PAFB, Ohio
1977 ……………………………………………… MS Mechanical Engineering

University of Kansas
Lawrence, Kansas
1977 -2001………………………………………… Lucent Technologies Bell Labs,

AT&T, Sandia National Labs,
Western Electric
Various locations and subsidiaries
1980 ……………………………………………… MBA Marketing

Avila College
Kansas City, Missouri
1997 - Present …………………………………… Ohio State University

Columbus, Ohio
2001-Present ……………………………………… LaBarge, Inc

Joplin, Missouri
FIELDS OF STUDY
Major Field: Industrial and Systems Engineering
Minor Field: Computational Methods
v
TABLE OF CONTENTS
Abstract ………………………………………………………………………………. i
Acknowledgements ………………………………………………………………….. ii
VITA …………………………………………………………………………………. v
List of Tables ………………………………………………………………………. viii
List of Figures ……………………………………………………………………….. x
Chapter 1 Introduction ................................................................................................ 1

1.1 Overview .............................................................................................................. 1
1.2 Plan of Study ........................................................................................................ 3
Chapter 2 Six Sigma Overview and History .............................................................. 5
2.1 Overview of Six Sigma ........................................................................................ 5
2.2 History of Six Sigma .......................................................................................... 12
Chapter 3 Literature Review..................................................................................... 19
3.1 Introduction ........................................................................................................ 19
3.2 Literature Review Methods and List of Articles ................................................ 19
3.3 The Classification Scheme ................................................................................. 22
3.4 Literature Trends ................................................................................................ 24
3.5 Literature Research Topics and Methods ........................................................... 28
3.6 Success Factors................................................................................................... 35
3.7 Six Sigma in the context of Management Theory.............................................. 36
3.8 The Academic Contribution of Six Sigma ......................................................... 37
3.9 Defining Six Sigma ............................................................................................ 39
3.10 The Quality Performance Model ...................................................................... 42
3.11 Review of Empirical Evaluations of Six Sigma ............................................... 44
3.12 Synthesis and Future Research......................................................................... 46
3.13 Literature Review Summary ............................................................................ 50
Chapter 4 Case Study Based Training ..................................................................... 53
4.1 Introduction ........................................................................................................ 54
4.2 Case Study Exercises.......................................................................................... 58
4.3 Sample Case Study “Improving the Yields of Printed Circuit Boards”............. 59
4.4 The Problem Statement ...................................................................................... 60
vi
4.5 Background Information .................................................................................... 60
4.6 In-class Exercise for PCB Study ........................................................................ 63
4.7 The Team Actions and Results........................................................................... 64
4.8 Student Feedback ............................................................................................... 68
Chapter 5 The challenge of Six Sigma Project Meso-Analysis............................... 70
5.1 Introduction ........................................................................................................ 70
5.2 Potential Use of Database................................................................................... 72
5.3 Database Example .............................................................................................. 73
5.4 General Procedure .............................................................................................. 74
5.5 Data Collection................................................................................................... 75
5.6 Statistical Process Control.................................................................................. 79
5.7 Regression .......................................................................................................... 84
5.8 Regression Data Analysis................................................................................... 87
5.9 Regression Results ............................................................................................. 94
5.10 Markov Decision Processes.............................................................................. 96
5.11 Markov Decision Process Analysis................................................................ 103
5.12 Sample MDP Example ................................................................................... 115
5.13 Conclusions .................................................................................................... 122
Chapter 6 Conclusions and Future Research ........................................................ 126
6.1 Overview .......................................................................................................... 126
6.2 Summary of Findings ....................................................................................... 127
6.3 Limitations and Future Research...................................................................... 130
Appendix A The Article Database.………………………………………………. 133
Appendix B Project Database……………………………………………………. 140
Bibliography………………………………………………………………………. 143
vii
LIST OF TABLES
Table 1 List of journals or proceedings with at least one article in the study. ............. 21
Table 2 Descriptors used to classify articles. ............................................................... 24
Table 3 Tabulations of articles by focus and authorship.............................................. 30
Table 4 Literature pertinent to evaluation of Six Sigma’s effects on firm performance.
(BP – Business Performance, C – Core, SSC – Six Sigma Core, SSI – Six Sigma
Infrastructure, OP – Operational Performance, QP – Quality Performance) ....... 45
Table 5 The yields achieved for 16 weeks prior to the initial teams activities. ........... 61
Table 6 The initial team's predicted yield improvements by adjusting each factor. .... 62
Table 7 The performance after the implementation of the initial team's
recommendations.................................................................................................. 62
Table 8 The yields for the 5 weeks subsequent to the initial intervention. .................. 65
Table 9 The data for a Pareto chart of the data from week 1 to week 16..................... 66
Table 10 The data from the fractional factorial design in four factors......................... 67
Table 11 The confirmation runs establishing the process shift/improvement.............. 68
Table 12 Definition of variables................................................................................... 77
Table 13 Summary statistics for the 39 projects, duration team memberrs and profits78
Table 14 Characteristics of the Study Variables .......................................................... 79
Table 15 Summary statistics for model. ....................................................................... 89
Table 16 Coefficient estimates ..................................................................................... 89
Table 17 Regression Checklist. .................................................................................... 91
Table 18 Summary statistics for Simple model............................................................ 94
Table 19 Coefficients for Simple model. ..................................................................... 94
Table 20 Costs and rewards (in $K), rt(st,a), of applying actions (a) in different states
(st).Coefficient estimates .................................................................................... 104
Table 21 Assumed transition probabilities for applying DOE, pt(st+1= j|st= i,a = $$
DOE)................................................................................................................... 105
Table 22 Optimal decision policy for five decision periods....................................... 106
Table 23 The expected reward, ERt*(st = i), in $K as a function of period and state. 106
Table 24 Component Method of Six Sigma. .............................................................. 108
Table 25 State Description Mapped to DMAIC for Sample MDP. ........................... 115
Table 26 Actions and Rewards for Sample MDP. ..................................................... 115
Table 27 Tabulation of State Transitions for Sample MDP. ...................................... 117
Table 28 Bayesian Estimate of State Transitions for Sample MDP........................... 118
Table 29 State Transitions for Sample MDP.............................................................. 120
Table 30 Optimal Policy for Sample MDP. ............................................................... 121
viii
Table 31 Comparison of strengths and limitations for meso-analysis methods used. 124
ix
LIST OF FIGURES
Figure 1 The yearly number of Six Sigma related articles and their authorship. ......... 26
Figure 2 The percentage of articles focused on manufacturing tropics........................ 27
Figure 3 Percentages of articles sponsored by difference societies or areas................ 28
Figure 4 Pareto chart of articles by research approach ................................................ 31
Figure 5 Journal impact factors associated with articles by different types of authors.32
Figure 6 Journal impact factors of publications pertinent to business sectors. ............ 33
Figure 7 Journal impact factors associated with articles focusing on people (Pe), Tools
and Techniques (To), Systems (Sy), or a combination of these. .......................... 34
Figure 8 Journal impact factors associated with articles focusing on philosophy (Ph),
practices (Pr), Tools and Techniques (To), or a combination of these................. 35
Figure 9 Percentages of articles mentioning each of 14 success factors...................... 36
Figure 10 Extended quality performance model of Garvin.......................................... 43
Figure 11 Control chart for the entire study period...................................................... 65
Figure 12 Normal Probability Chart for Six Sigma Projects........................................ 80
Figure 13 EWMA Control Chart for first 25 Six Sigma Projects. ............................... 82
Figure 14 EWMA Control Chart for Six Sigma Projects............................................. 83
Figure 15 Main Effects Plot of Regression Model. ...................................................... 88
Figure 16 3-D plot of the results.................................................................................. 90
Figure 17 Normal sores vs. residuals. .......................................................................... 91
Figure 18 Main Effects Plot of Simple Regression Model........................................... 93
Figure 19 Flow Chart of DMAIC Six Sigma Project. ................................................ 110
Figure 20 Morkov Decision Process for Six Sigma. .................................................. 114
x
CHAPTER 1
INTRODUCTION
1.1 Overview
Six Sigma was born approximately two decades ago as a process improvement
philosophy to help improve business financial performance. It was developed in
industry and spread largely by professional consultants. Since its introduction it has
found its way into most sectors of today’s business society. Mostly led by practitioners
Six Sigma has acquired a strong prescriptive stance with practices often being
advocated as universally applicable, “one size fits all”. Yet, Six Sigma has already
spawned a large number of published articles in pier reviewed journals. One of the
main objectives of this dissertation is to review this information and identify
opportunities for additional contributions for the academy.
While Six Sigma has made a big impact on industry, the relationship between
the university, industry and Six Sigma has not been studied. University-industry
partnerships are expected to contribute to the advancement of knowledge and
innovation in the production of competitive products, processes and services and
ultimately contribute to the welfare of society.
1
The major problems addressed in this dissertation are:
1. With reference to past academic contributions, what is Six Sigma? Also,
what are the on-going research trends?
2. What are the implications of Six Sigma philosophy and methods for
university education?
3. What methods should be used to mine the new databases about project
financial results? Also, what insights can be gained form studying data at a
real company?
This dissertation addresses these three aspects of the university-industry
relationship with Six Sigma. For the first question, the literature covering a fourteen
year timeframe describing the trends, sources, and findings of Six Sigma is reviewed.
The aim is to provide a description of the Six Sigma literature with an emphasis on
establishing its relationship to quality management theory and to business practices in
general and to identify topics for future research.
Secondly, case base training is examined as a method to improve Six Sigma
education and increase usage on the job among university student learners. From the
literature review, one of the major themes and “success factors” identified by authors
of Six Sigma articles is education of practitioners in the use of statistical tools. Six
Sigma differs form most quality management systems by its emphases on training
what people need to know to implement improvement programs effectively. A subset
of this topic is the education of non-statisticians in the use of statistical tools. A major
2
concept than becomes how to effectively educate participants in the necessary tools
and techniques of Six Sigma.
Third, with Six Sigma’s emphasis on management by data and project based
data collection, industry is starting to accumulate a large database on quality
improvement programs. Quality/cost improvement project data was collected over a
30 month period at a U.S. based mid-west manufacturing firm. A study was
undertaken to look at this database in a way that could help management decision-
making applied to improvement projects. Three levels of decision-making are defined
as:
Micro – related to the use of individual statistical methods.
Meso – supervisor level decision-making about method selection and timing.
Macro – dealing with overall quality programs and stock performance.
This study investigates the use of Six Sigma databases in Meso-Analysis for decision-
making.
1.2 Plan of Study

The remainder of the study is organized as follows:
• Chapter 2 is an overview and history of Six Sigma. It describes how
financial accountability for each project is established by the literature
as a key component of Six Sigma philosophy and methods. As a result,
databases about the financial successes or failures of individual projects
have been generated in many companies for the first time.
3
• Chapter 3 covers a review of the literature concerning Six Sigma from
1990 through 2003. This chapter presents a classification scheme for
the articles, trends, topics covered, contributions, and tabulates
identified success factors for Six Sigma. A definition of Six Sigma is
given based on a synthesis of the literature.
• Chapter 4 examines case based instruction in classes on SPC and DOE
at Ohio State University. The approach is illustrated with an exercise
based on an actual case study of increasing yields for manufacturing
electronic components.
• Chapter 5 describes methods for mining the Six Sigma project
databases appearing at many companies and the properties of these new
data sources. The results of a study based on 39 quality improvement
projects conducted at a U.S. based mid-west manufacturing firm are
used to investigate how this new database might be used to help
operational management in the area of quality management practices.
• Chapter 6 summarizes the findings, limitations, and conclusions of the
study and discusses potential areas for future research.
4
CHAPTER 2
SIX SIGMA OVERVIEW AND HISTORY
2.1 Overview of Six Sigma

One definition of “Six Sigma” is a target for quality characteristics of units
produced by the engineered system being improved (Shina, 2002; Tadikamalla, 1994).
It is a rating that signifies “best in class”, with only 3.4 defects per million units or
operations. A part or item is classified as defective if the desired measurement,
denoted by X, is outside the upper or lower specification limits (USL or LSL). In
addition to specifying the USL and LSL, a target value is specified, which typically is
the midpoint between the USL and LSL.
The symbol sigma (σ) is a letter in the Greek alphabet used by modern people
to describe variability. In Six Sigma, the common measurement index is defects per
million opportunities and can include anything from a component, piece of material,
line of code, an administrative form, time frame or distance. A sigma quality level
offers an indicator of how often defects are likely to occur, where a higher sigma
quality level indicates a process that is less likely to create defects. Consequently, as
sigma level of quality increases, product reliability improves, the need for testing and
5
inspection diminishes, work in progress declines, cycle time goes down, costs go
down, and customer satisfaction goes up.
Six sigma is a condition of the generalized formula for process capability,
which is defined as the ability of a process to turn out a good product. It is a
relationship of product specifications to manufacturing variability, measured in terms
of Cp or Cpk, or expressed as a numerical index. Six sigma is equivalent to Cp=2 or
Cpk=1.5.The definition of the capability of the process or Cp is:
specification width(or design tolerance)

Cp = (1.1)
process capability(or totalprocess var iation)
Specifically,
USL − LSL
Cp = (1.2)
6σ (total process range from − 3σ to + 3σ )
This formula can be expressed conceptually as,
product specifications
Cp = (1.3)
manufacturing var iability
The equation for Cpk is:
⎡USL − X X − LSL ⎤
Cpk = min ⎢ , ⎥ (1.4)
⎣ 3σ 3σ ⎦
Six sigma is achieved when the product specifications are at ± 6σ of the
manufacturing process corresponding to Cp=2 or Cpk=1.5. Design engineers normally
set the product specifications, whereas manufacturing engineers are responsible for
6
production variability. The object of increasing the process capability to six sigma is
twofold: either increase the product specifications by widening them, or reducing the
manufacturing variability. Either effort can have a positive effect on reaching six
sigma.
An alternative and more common definition for “Six Sigma” methods, implied
by Pande and Holpp (2001) and Watson (2002a), is a series of ordered activities with
associated component methods (Allen, 2003). Six Sigma is a disciplined and
quantitative approach involving setting up a system and process for the improvement
of defined metrics in manufacturing, service, or financial processes. The approach
drives the overall process of selecting the right projects based on an organization’s
business goals and selecting and training the right people to obtain the results.
Improvement projects follow a disciplined process defined by a system of five macro
phases. These component methods derive from statistics, marketing, and optimization
and are sequenced as Define, Measure, Analyze, Improve, and Control (DMAIC). In
design projects the specifics of the DMAIC steps are often modified to DMADV;
Define customer requirements and goals for the product.

Measure and match performance to customer requirements.
Analyze and assess product design.
Design and implement new product.
Verify results and maintain performance.
7
The phases of DMAIC are described by Rasis, Gitlow and Popovich (2003a
and 2003b) as follows:
Define Phase: Define the project’s objectives by identifying customer
requirements often called “CTQs” “critical to quality”, develop a team charter and
define process map.
• Identify the process or product for improvement, identify customers
and translate the customer’s needs into CTQs.
• The team charter involves selection of team members and defining of
roles, developing the problem and goal statements, determining project
scope, setting project milestones and preparing a business case to gain
management support.
• Do a high level process map connecting the customer to the process.
Measure Phase: Measure the existing systems. Establish valid and reliable
metrics to help monitor progress towards the project goals. Customer expectations
are defined to determine “out of specification” conditions.
• Identify and describe the potential critical processes/products. List
and describe all of the potential critical processes obtained from
brainstorming sessions, historical data, yield reports, failure
analysis reports, analysis of line fallout and model the potential
problems.
8
• Perform measurement system analysis. Determine precision,
accuracy, repeatability and reproducibility of each instrument of
gauge used in order to ensure that they are capable.
Analyze Phase: Analyze the system to identify ways to eliminate the gap
between the current performance of the system or process and the desired goal. In
this phase, project teams explore underlying reasons for defects. They use
statistical analysis to examine potential variables affecting the outcome and seek to
identify the most significant root causes. Then, they develop a prioritized list of
factors influencing the desired outcome.
• Isolate and verify the critical processes. Narrow the potential list of
problems to the vital few. Identify the input/output relationship
which directly affects specific problems. Verify potential causes of
process variability and product problems.
• Perform process and measurement system capability studies.
Identify and define the limitations of the processes. Ensure that the
processes are capable of achieving their maximum potential.
Identify and remove all variation due to special causes. Determine
what the realistic specifications are. Determine confidence
intervals. A process is to be considered capable when it is in
control, predictable, and stable.
Improve Phase: In this phase, project teams seek the optimal solution and
develop and test a plan of action for implementing and confirming the solution.
9
The process is modified and the outcome is measured to determine whether the
revised method produces results within customer expectations.
• Conduct design of experiment. Select design of experiment factors
and levels, Plan design of experiment execution. Perform design of
experiment to find out the most significant factor.
• Implement variability reduction designs/assessments. Implement
permanent corrective action for preventing special cause variations.
Demonstrate process stability and predictability.
Control Phase: Control the new system. Ongoing measures are implemented
to keep the problem form recurring. Institutionalize the improved system by
modifying policies, procedures, operating instructions and other management
systems.
• Specify process control methods. Establish on-going controls for
the process based on prevention of special cause variation using
statistical process control techniques.
• Document the improvement processes. Record all the
processes/steps in improvement phase using the decision tree and
reaction plan.
The methods are generally taught in the context of system improvement
projects and expertise is often characterized by an analogy to karate “belts”: “black
belt”, “green belt”, etc. For thousands of participants at the lowest “green belt” level
of accreditation, one of the main benefits of “Six Sigma” training is that it simplifies
10
(through restriction) the sequence and choice of available techniques to apply to a
particular case. Therefore, the value of the six sigma movement derives partly from
standardization of problem solving methods and partly in how it guides people to
suggest which techniques to apply in which order to an improvement project.
Contributing to the widespread deployment of Six Sigma is an abundance of
anecdotal evidence attributing quality, productivity, and costs benefits to this
particular quality improvement initiative. Scientific evidence to lead credence to the
anecdotal evidence has been rather limited and exists primarily as small-sample case
studies. Moreover, while the empirical results from these case studies have generally
been atheoretical in nature, their conduct had not been governed by rigorous, a priori
theory development.
Bisgaard and Freiesleben (2000) showed how defect elimination and
prevention associated with a Six Sigma program can improve financial results. Their
view was to do or not to do a project. Hild, Sanders, and Cooper (2000) discussed the
different structures of Six Sigma based on processes type, continuous or discrete.
Sanders and Hild (2000a and 2000b) outlined the importance of considering
organizational issues in the structuring of successful Six Sigma projects. Snee (2001a)
similarly stated that one of the keys was to understand the environment in order to
tailor Six Sigma projects; type of company (manufacturing or service), type of
function (operations, transactional, administrative, or new product development) and
type of industry (assembly, processing, chemical, etc.). Pyzdek (2001b) pointed out
11
the importance of selecting the right individual to act as project leader even before
they are trained.
Pande, Neuman, and Cavanagh (2000) contributed probably the most complete
and explicit version of the Six Sigma methods. Yet, even their version of the
methodology leaves considerable latitude to the implementers to tailor approaches to
applications and to their own tastes. This lack of standardization of the methodologies
explains, at least in part, why the American Society for Quality did not have a
certification process until 2001.
2.2 History of Six Sigma

According to Folaron (2003) Six Sigma can arguably trace its roots as far back
as 1798 and Eli Whitney. One of his greatest contributions to modern manufacturing
was the introduction of his revolutionary uniformity system in the mass production of
muskets. He proved it was possible to produce interchangeable parts that were similar
enough in fit and function to allow for random selection of parts in the assembly.
Throughout the next century, quality involved objective methods of measuring and
assuring dimensional consistency.
With the advent of the moving assembly line by Ford in the early 1900’s it was
critical to predetermine the consistency of incoming parts to prevent a slow down or
stoppage of the assembly line. In addition with the increased production rates, it
became cost prohibitive to measure each part. This necessitated the development of
12
methods to monitor the part producing process for consistency and the use of sampling
(Folaron, 2003).
In 1924, Walter A. Shewhart from Bell Telephone Laboratories, proposed the
concept of using statistical charts to control the variables of products manufactured at
Western Electric. This was the beginning of statistical quality control (Small, 1956).
The role of the quality inspector changed with the statistically based control charts
form one of identifying and sorting defective product to one of monitoring the stability
of the process and identifying when it had changed. Improved product quality resulted
by early detection of the change and appropriate corrective action.
Dr. Shewhart kept on with his efforts and applied the fundamentals of statistical
quality control to industry. This lead to the modern attention to the use of statistical
tools for the manufacture of products and process originated prior to and during World
War II, when the United States of America geared up to a massive buildup of
machinery and arms to successfully conclude the war. The need to manage the myriad
of complex weapon systems and their varied and distributed defense contractors led to
the evolution of the system of Statistical Quality Control (SQC), a set of tools that
culminated in the military standards for subcontracting, such as MIL-Std 105 (Shina,
2002). The basis of the SQC process was the use of 3 sigma limits, which yields a rate
of 2700 defective parts per million (PPM).
Prior to that period, large U.S. companies established a quality strategy of
vertical integration (Shina, 2002). In order to maintain and manage quality, companies
had to control all of the resources used in the product. U.S. companies slowly realized
13
that quality improvements depended on the realization of two major elements. First,
they have to be quantifiable and measurable, and second all elements that make the
company successful must be implemented. These are superior pricing, delivery,
performance, reliability, and customer satisfaction.
The Western Electric manufacturing company is noteworthy during this time
because it was the breeding ground for many quality leaders, not only Shewhart but
Joseph Juran, Edwards Deming and Kaoru Ishikawa all worked there at some time
(Dimock, 1977). After World War II, numerous manufacturing experts where involved
with the rebuilding of the Japanese business infrastructure. Two prominent individuals
were Deming and Juran. Deming promoted the use of the plan-do-check-act (PDCA)
cycle of continuous improvement. Later Juran introduced the concepts of project by
project quality improvement.
Any discussion on quality today will most likely cite at least one from the
group of Deming, Juran, Crosby, Feigenbaum, and Ishikawa, if not all. They certainly
represent the preponderance of information about quality. In fact, the concepts of
TQM are mainly based on their works on quality.
Their collective philosophies can be summarized as follows:
• Management leadership and employee participation in the new
philophy (Deming, 1986). Make quality the concern of everyone in the
company (Crosby, 1980; Juran, 1989; Feigenbaum, 1991).
• Emphasis on meeting the requirements of both the internal (Crosby,
1980; Feigenbaum, 1991) and external customers (Ishikawa, 1985).
14
• Eliminate non-conformance, Appraise conformance to standards. Have
zero defect standard of performance (Crosby, 1980). Reduce cost of
appraisal, prevention and failure (Feigenbaum, 1991).
• Use statistical and quantitative control methods. Implement problem
solving using Quality Control Circles, Plan-Do-Study-Act cycle, and
Quality Assurance (Deming, 1986; Ishikawa, 1985).
• Search continually to improve process and product (Demin, 1986).
Quality is a continuous program (Crosby, 1980: Feigenbaum, 1991).
Adding to this group, Bill Smith, Motorola Vice President and Senior Quality
Assurance Manager, is widely regarded as the father of Six Sigma, Shina (2002),
although several have played key roles in promoting this phrase including Harry
(1994) and Pande, Neuman, and Cavanagh (2000).
According to Shina (2002) before, January 15, 1987, Six Sigma was solely a
statistical term. Since then, the Six Sigma crusade, which began at Motorola, has
spread to other companies which are continually striving for excellence. While it is
progressing, it has extended and evolved from a problem-solving technique to a
quality strategy and ultimately into a sophisticated quality philosophy. However, this
unique philosophy only became well known after GE’s Jack Welch made it a central
focus of his business strategy in 1995. Today, Six Sigma is considered one of the
fastest growing business management system in industry (Cook, 1990; Gill, 1990;
Rayner, 1990; Behara et al.,1995; and Maguire, 1999b). Since its initial development
and deployment at Motorola, Six Sigma has influenced virtually every sector of the
15
economy, from manufacturing to service and from the largest to the smallest
organization. Today, Six Sigma processes are being executed in a vast array of
organizations and in a wide variety of functions (Breyfogle, 1999).
The evolution of what is known as Six Sigma began in the late 1970s, when a
Japanese firm took over a Motorola factory that manufactured television sets in the
United States and the Japanese promptly set about making drastic changes to the way
the factory operated (Folaron, 2003). Under Japanese management, the factory was
soon producing TV sets with 1/20th the number of defects they had produced under
Motorola management. Finally, Motorola recognized it quality was awful. Since then
Motorola management decided to take quality seriously (Main, 1994; Pyzdek 2000).
When Bob Galvin became Motorola’s CEO in 1981, he challenged his company to
achieve a tenfold improvement in performance over a five-year period.
During 1985, Bill Smith wrote an internal quality research report which caught
the attention of Bob Galvin, Smith discovered the correlation between how well a
product did in its field life and how much rework had been required during the
manufacturing process. He also found that products that were built with fewer
nonconformities were the ones that performed the best after delivery to the customer
(Harry, 1998; Maguire, 1999b). Although Motorola executives agreed with Smith’s
supposition, the challenge then became how to create practical ways to eliminate the
defects. Smith developed a four-stage problem-solving approach: Measure, Analyze,
Improve, Control (MAIC). Later, the MAIC discipline became the roadmap for
DMAIC and achieving Six Sigma quality.
16
On January 15, 1987, Galvin launched a long term quality program, called
“The Six Sigma Quality Program”. The program was a corporate program which
established Six Sigma as the required capability level standard. This new standard was
to be used in everything, that is, in products, processes, services, and administration.
After implementing Six Sigma, in 1988, Motorola was among the first recipients of
the Malcom Baldrige National Quality Award.
At that time, enamored by Motorola’s success, several other companies, such
as Texas Instruments, began a similar pursuit, But, it wasn’t until late 1993 that Six
Sigma really began to transform business. That’s the year Allied Signal (Honeywell)
and its CEO, Larry Bossidy, decided to adopt Six Sigma.
By adequately selecting the right Six Sigma projects and promptly providing
the right support for them, Bossidy suggested that high level executives should also
understand Six Sigma tools. At Allied Signal, an entire system of leadership and
support systems began to form around the statistical problem solving tools of Six
Sigma.
Not long after Allied Signal began its pursuit of Six Sigma quality, Jack
Welch, then Chairman and Chief Executive of General Electric, influenced by
Bossidy, began to get interested in Six Sigma.
Some argue that many of the tools used with Six Sigma are not new
(Tadikamalla, 1994: Hahn et. Al, 1999; Watson, 2000). However, while Six Sigma
uses conventional methods, its application is anything but conventional. Instead it
stresses the impotence of searching for a new way of thinking and doing (Maguire,
17
1999b). In fact, Six Sigma defines a clear roadmap to achieve Total Quality
(Balkeslee, 1999; Breyfogle, 1999; Pyzdek, 2000; Harry, 1998; Young, 2001).
Six Sigma’s approach and deployment makes it distinguishable from other
quality initiatives. The Six Sigma approach involves the use of statistical tools within
a structured methodology for gaining the knowledge needed to achieve better, faster,
and less expensive products and services than the competition. The repeated,
disciplined application of the master strategy on project after project, where the
projects are selected based on key business objectives, is what drives dollars to the
bottom line, resulting in impressive profits. Moreover, fueled by the bottom line
improvement, top management will continuously be committed to this approach, the
work culture will be constantly nurtured, then customers will definitely be satisfied
and Total Quality will ultimately be achieved.
18
CHAPTER 3
LITERATURE REVIEW
3.1 Introduction
Although Six Sigma originated in industry, it has inspired a
considerable amount of academic literature. This chapter reviews this literature
covering a fourteen-year timeframe (1990 through 2003) describing the trends,
sources, and findings. The chapter also seeks to synthesize the literature, with an
emphasis on establishing its relationship to quality management theory and topics for
future research. In doing so, there is an attempt to answer the fundamental questions:
(i) What is Six Sigma? (ii) What are its impacts on operational performance? and (iii)
What roles can academics usefully play in relation to Six Sigma?
3.2 Literature Review Methods and List of Articles

The list of articles was derived from a Science Citation Index (SCI) Expanded
search spanning the time period from 1990 through 2003. Five descriptors were used:
Six Sigma, quality systems, quality improvement, quality management, and quality
19
meta-model. The test of each article was reviewed in order to eliminate those that were
clearly not related to “Six Sigma” improvement strategies. For example, articles were
removed that focused on detailed synthesis of chemicals and used the term Six Sigma
in an unrelated context. Also, a small number of articles were included from
magazines and conference proceedings that were subjectively assessed to be academic
in character. The list of journals, proceedings, and magazines that provided at least
one relevant article is shown in Table 1. Two hundred and one (201) articles were
identified and covered in the review. The terms used to categorize these articles are
defined in section 3.3. Overall, it is not claimed that the list of articles is exhaustive,
only that the associated database serves as a reasonably comprehensive list for
understanding Six Sigma related research.
The focus of this review was in three areas. First, what is the definition of Six
Sigma? A solid definitional foundation must exist before rigorous analysis can be
undertaken. Second, what is the impact of Six Sigma on a firm’s performance? Six
Sigma reportedly has saved millions of dollars for such companies as GE, but can its
benefits be quantified in the literature? Third, what has been reported on how to
implement Six Sigma in a real business setting? Is there guidance for the practitioner
on the how and why to successfully implement a Six Sigma program?
20
Accreditation and Quality Assurance Journal of Evaluation in Clinical Practice
AIAA-2002-1471 Journal of Healthcare Management
Annual Quality Congress Transactions Journal of Management Engineering
Annual Reliability and Maintainability Journal of Manufacturing Science and
----Symposium Proceedings ----Engineering Transactions of the
Archives of Pathology & Laboratory Medicine ----ASME
Assembly Automation Journal of Mechanical Design
Aviation Week and Space Technology Journal of Operations Management
Building Research and Information Journal of Quality and Participation
Business Management Journal of Quality Technology
Business Month Journal of The IES
Cancer Journal Lecture Notes In Computer Science
Chemical Engineering Communications Manufacturing Engineering
Chemical Engineering Progress Milbank Quarterly
Chemical Week Proceedings of the 2001 Winter Simulation
Clinical Chemistry ----Conference
Computers & Industrial Engineering Proceedings of the 2002 Winter Simulation
Computers In Industry ----Conference
Control Engineering Proceedings of the ASME Design
Electronic Business ----Engineering Technical Conference
Fortune Professional Engineering
Genetic Engineering News Quality and Reliability Engineering
Hospitals & Health Networks ----International
Hydrocarbon Processing Quality Digest
IEEE Engineering Management Review Quality Engineering
IEEE Software Quality Management in Health Care
IEEE Transactions on Neural Networks Quality Progress
IEEE Transactions on Semiconductor R&D Magazine
----Manufacturing Radiology
IIE Solutions Research-Technology Management
Industrial Management & Data Systems Six Sigma Forum Magazine
International Journal of Production Research Technometrics
International Journal of Quality & Reliability The American Statistician
----Management The Physics Teacher
International Journal of Quality Science Therapeutic Apheresis
Journal of American Geriatrics Society Total Quality Management
Journal of Applied Statistics Total Quality Management & Business
Journal of Engineering Design ---- Excellence
Training & Development
Table 1 List of journals or proceedings with at least one article in the study.
21
3.3 The Classification Scheme
Articles were classified using the eleven descriptors in Table 2. Authors
represented either academic institutions or industrial companies or constituted a team
with representatives from both. Many articles contain definitions of the phases
Define, Measure, Analyze, Improve, and Control (DMAIC) but most did not.
Two schemes were used to evaluate the primary topic(s) of each article.
Oakland (1989) divided quality issues roughly into systems, practices, people, or other
focused, without providing precise definitions of these terms. Zain, Dale and Kehoe
(2001) followed Oakland in using this division to classify articles (version 1). Sousa
and Voss (2002) developed a modified scheme based on philosophies, practices, tools
and techniques, and other (version 2). Sousa and Voss (2002) defined “philosophy” as
“an approach to management,” and practices as “an observable facet of a philosophy
and it is through them that managers work to realize organizational improvements.”
Those authors also described “tools and techniques” as “core elements” with examples
being process control and Pareto analysis.
The Science Citation Index (SCI) provides a number called “journal impact
factor” that is a ratio between the citations to articles in a journal to the average
number of citations to journals in that field. The impact factor can be viewed as a
rough evaluation of the academic quality of the journal. Many articles made explicit
reference to either the manufacturing or service sector issues, while others offered
general contributions. A common feature of articles was mentioning 3.4 defects per
million opportunities in relation to the definition of Six Sigma.
22
The articles were each affiliated with one of the following sponsoring societies
or areas of study: the American Institute of Chemical Engineers (AICHE), the
American Society of Mechanical Engineers (ASME), the applied statistics area
including publications sponsored by the American Society of Quality (ASQ) and the
American Statistical Association (ASA), the Institute of Industrial Engineers (IIE), the
operations research or management science (OR/MS) area including publications
sponsored by the Institute for Operations Research and Management Science
(INFORMS) and other related journals, or the medical area in general including the
American Medical Association (AMA).
Following Zain, Dale, and Kehoe (2001), articles were classified as focused on
case studies, survey results, literature review, comparative analysis, or theoretical with
application. A sizable fraction of the articles investigated the factors contributing to
the success of Six Sigma implementations. For those articles, the specific success
factors mentioned were tabulated. The terminology used to describe the success
factors was standardized to correspond to the dimensions of quality management
practice in Sousa and Voss (2002) whenever possible.
Finally, articles that recommended the usage of one or more practices without
clarifying conditions in which this practice has provable properties were classified as
“speculative” in nature. This classification differentiated these articles from others
without specific recommendations for practices or having rigorous statistical or
optimization justifications.
23
Descriptor Source Levels
Authorship Brady and Allen Industrial (I), Academic (A), or Both (I A)
Define DMAIC Brady and Allen Yes (Y) or No (N)
Topics Version 1 Oakland (1989) Systems (Sy), Tools and Techniques (To), and
People (Pe)
Topics Version 2 Sousa et al. (2002) Philosophy (Ph), Practices (Pr), Tools and
Techniques (To), and Other
Industrial Sector Zain et al. (2001) Manufacturing (M), Service (Se), or General
(G)
Journal Impact Science Citation Index 0.13 to 4.76
Factor
Mention of 3.4 ppm Brady and Allen Yes (Y) or No (N)
Research Approach Zain et al. (2001) Case Study (Ca), Comparative (Co), Survey
(Su), Literature Review (R), or Theoretical with
Application (TA)
Society or Area Brady and Allen AIChe, ASME, ASQ, IIE, INFORMS, or
Medical
Success factors Brady and Allen All combinations of 13 possible factors
Speculative in Brady and Allen Yes (Y) or No (N)
Nature
Table 2 Descriptors used to classify articles.
3.4 Literature Trends

In this section, a characterization of the database of articles using statistics
derived from the classifiers described in the last section is presented. Goals include
the identification of trends including those that relate to the authorship of articles and
the subjects addressed. There is also an investigation of the interaction of authorship
with research focus and a tabulation of the associated sponsoring societies or areas of
study. Finally, results relating to success factors including a tabulation of the success
factors cited most often in the literature are discussed.
Figure 1 plots the number of articles verses the year. Based on the number of
papers there is little doubt that the subject is actively reported. The plot suggests two
24
findings. First, the number of articles by industrial authors peaked in 2000. It is
hypothesize that this declining trend was influenced by condemnations of Six Sigma
in the popular press such as Clifford (2001) in Fortune Magazine. Second, at the same
time, interest among academics continued to grow in 2003. Over the entire search
period, 69.2% of the authors had industry affiliations and 30.8% had academic
affiliations. These proportions have been changing to the point where 53% of the
authors reviewed in 2003 were associated with a university or college. This trend in
authorship from industry dominated to academic dominated is not surprising
considering the industrial origins of Six Sigma. It would be anticipated that the
growing interest in Six Sigma from the academic arena would add rigor and
theoretical understanding to the subject.
25
45
40
Academia
35
Industry
30
Number of Articles
25
20
15
10
0
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
Year
Figure 1 The yearly number of Six Sigma related articles and their authorship.
Another trend is the diversification of research topics from primarily
manufacturing focused to more general in nature as indicated by Figure 2. This trend
is characterized with increased emphasis on generic issues and on service related
business sectors. Particularly, there is increased emphasis on generic statistical tools
such as design of experiments (DOE), probabilistic design, and statistical process
control (SPC) in the context of Six Sigma, e.g., Mason and Yong (2000), Coleman et.
al. (2001), McCarthy and Stauffer (2001), Koch (2002) and Goh (2002). Figure 3
charts the percentages of articles associated with different areas. The fact that the
26
earliest medical related publication in the database is Buck (1998) supports the finding
that the medical area is playing an important role in increasing the topic diversity. Six
Sigma has been advocated as universally applicable to organizations, a one-size-fits-
all. Rigorous academic studies are needed to question the universal validity of Six
Sigma across such dimensions as industry sector, company size, country/culture,
product/process/service, and frequency of product change or introduction.
100%
90%
% of Manufacturing Articles
80%
70%
60%
50%
40%
30%
20%
10%
0%
1990 1992 1994 1996 1998 2000 2002
Figure 2 The percentage of articles focused on manufacturing tropics.
Figure 3 also indicates that the applied statistics journals such as the Journal of
Quality Technology, Quality Engineering, Quality Management Journal, Quality
Progress, and Technometrics dominate scholarly publications on Six Sigma. Note
that Technometrics is sponsored jointly by ASQ and the American Statistical
Association (ASA). This dominance is perhaps surprising considering the
27
multidisciplinary nature of Six Sigma as noted by Hahn, Doganaksoy, and Hoerl
(2000) and others.
IEEE
OR+MS
Medical
ASME
AICHE
Applied Statistics
0% 10% 20% 30% 40% 50%
Figure 3 Percentages of articles sponsored by difference societies or areas.
3.5 Literature Research Topics and Methods

Next, the topics and research approaches of the articles in the database are
examined. We begin by focusing on the topics covered and the dependence of the
number of articles and scholarly impact on the authorship. Then, the methods used in
relation to scholarly impact are investigated.
Table 3 contains a cross tabulation of the topic and authorship variables. It
supports three findings. First, the majority of articles focused on either philosophy or
28
systems topics. The percentages on these topics were 54.7% and 66.2% respectively.
In general, papers in these categories provided a general description of Six Sigma and
advocated its use, e.g., Rayner (1990), Harry (1998), Snee (2001) and Does et. al.
(2002). It was found that 32% of the total articles are in this category are introductory
in nature. Overall, only 6% of the articles were written at the practices level which
Sousa and Voss (2002) argued are most useful for stimulating actual organizational
improvements. Academic authors wrote about practices with higher frequency (10%).
Second, Table 3 also shows that academics were more likely to choose topics
amenable to theoretical study such as tools and techniques and practitioners were more
likely to present philosophical or systems level contributions. For example, only 17%
of articles by exclusively industrial authors concerned tools and techniques compared
with 36% of the articles by exclusively academics.
The literature has focused primarily in the two areas of philosophy or tools and
techniques. These areas of study are not helpful for development of usable models to
aid decision makers in the implementation of Six Sigma. More research in needed into
the linkage between the practices involved in Six Sigma.
29
Topics Version 1 from
Oakland (1989) Academic Industrial Mixed Totals
Systems 23 83 4 110
Tools and Techniques 20 28 5 53
People and Systems 3 18 2 23
People and Tools and
Techniques 4 5 0 9
Systems and Tools and
Techniques 0 5 1 6
Topics Version 2 from

Sousa and Voss (2002) Academic Industrial Mixed Totals
Philosophy 26 101 6 133
Tools and Techniques 18 25 5 48
Philosophy and Tools and
Techniques 1 6 1 8
Philosophy and Practices 3 3 0 6
Practices 1 2 0 3
Practices and Tools and
Techniques 1 2 0 3
Totals 50 139 12 201
Table 3 Tabulations of articles by focus and authorship.
Figure 4 is a Pareto chart of the number of articles associated with the different
research methods. The papers containing case studies constituted a sizable fraction of
papers on all topics. For example, 40% of the papers classified as philosophy focused
contained case studies with no new tools and techniques and 50% of the papers
exclusively on tools and techniques contained a case study with no new tools and
techniques. Of the articles with case studies the majority contained only a single case.
These articles for the most part have been descriptive in nature lacking rigor. Those
articles classified as theoretical with application where written primarily by
academics. It was noted that they offered more solid foundation but mostly dealt with
30
tools and techniques and where not helpful in developing a model on how and why six
Sigma works.
# of Articles
0 10 20 30 40 50 60 70 80
Case Study
Research Approach
Theoretical with Application
Survey
Comparitive
Literature Review
Other
Figure 4 Pareto chart of articles by research approach .
As described in Section 3.3, journal impact factors were developed by the
Science Citation Index (SCI) to provide a rough measure of journal quality or impact.
Figure 5 is a box and whisker plot of the journal impact factors associated with the
articles in the database. It shows that, not surprisingly, academic authors tended to
publish in journals with higher scholarly impact. Again, these dealt mostly at the
technique level and not the level of practices, which is most helpful to those decision
makers looking for guidance. Figure 6 is a box and whisker plot of the journal impact
31
factors associated with articles associated with either manufacturing or service sectors
of business or of generic interest. The plot shows that service related publications
have the highest scholarly impact. This can be attributed to the relatively greater
impact associated with the specific journals Clinical Chemistry, Radiology, and the
Journal of the American Geriatrics Society. The associated articles covered topics
classified as systems and tools and techniques.
4
Impact factor
0
Academic Industry Mixed
Figure 5 Journal impact factors associated with articles by different types of authors.
32
5
Impact factor 4
0
General Manufacturing Service
Figure 6 Journal impact factors of publications pertinent to business sectors.
A sizable fraction of the articles in the most prestigious statistics journals
concerned the tools that Six Sigma black belts “should” know, e.g., the discussion of
Hoerl (2001a) and Hoerl (2001b) and Montgomery, Lawson, and Molnau (2001).
These articles were classified into people combined with tools and techniques in the
Oakland (1989) scheme and philosophy and tools using the Sousa and Voss (2002)
scheme. These articles caused both categories to be associated with the highest
median journal impact factors. Figure 7 and Figure 8 provide box and whisker plots
of the impact factors associated with the research topics. Surprisingly, the topic
33
associated with the least impact is “practices.” Sousa and Voss (2002) argued that this
topic was the one most likely to stimulate organizational change.
Two other important themes related to people topics. First, articles focusing
on the cultural implications or management actions included Sanders and Hild(2000b)
and Wiklund (2002). Second, leadership and training are also popular themes (e.g.,
see Hahn, 1999, and Hoerl, 2001). Many of these articles examined of success factors
as we describe next.
4
Impact factor
0
Pe Sy Pe To Sy Sy To To
Figure 7 Journal impact factors associated with articles focusing on people (Pe), Tools
and Techniques (To), Systems (Sy), or a combination of these.
34
5
4
Impact factor
3
0
Ph Ph Pr Ph To Pr Pr To To
Figure 8 Journal impact factors associated with articles focusing on philosophy (Ph),
practices (Pr), Tools and Techniques (To), or a combination of these.
3.6 Success Factors

In the database, 26.9% of the articles made a claim to the “success factors”
necessary for Six Sigma to succeed. Fourteen separate factors can be found among the
literature, some being in conflict with each other, Figure 9. Half would be classified as
infrastructure, half would be classified as core as defined by Sousa and Voss (2002).
The most commonly cited factor is “Top Management Commitment”, which could be
said of any organization initiative. These success factors are offered for the most part
without rigorous proof. The second most common factor was training. This high lights
that training is an integral part of Six Sigma.
35
Critical thinking
Adaptable system
Change management
Goal based approach
Right project leadership

Listed Success Factors
Customer focused
Project selection
Team involvement
Bottom line focus
Forming the right team
Structured approach
Data system
Team Training
Top Management Commitment
0.0% 5.0% 10.0% 15.0% 20.0% 25.0% 30.0%
Figure 9 Percentages of articles mentioning each of 14 success factors.
3.7 Six Sigma in the context of Management Theory

Six Sigma was developed by industry practitioners at Motorola who were not
primarily interested in academic contributions. It is not surprising, then, that the role
of Six Sigma in the context of management theory is obscure and, as noted in
Linderman et. al. (2002), only a small fraction of the Six Sigma literature has been
devoted to theory. In this section, the description of Six Sigma to bring its
contribution into clearer focus is synthesized. Then, modifications appropriate for the
36
evaluation of Six Sigma to the quality performance model of Garvin (1987) are
suggested.
3.8 The Academic Contribution of Six Sigma

This section begins by reviewing four facts established previously about Six
Sigma. First, it can be debated whether or not the principle of establishing monetary
justification for applying the Six Sigma method belongs in the definition. Yet,
monetary justification of projects assuredly is associated with Six Sigma. Second, Six
Sigma is relatively specific in nature in relation to the pantheon of quality
management practices. This fact is established by the definition in Linderman et. al.
(2002) of Six Sigma as a “method”. Also, 24% of articles defined the DMAIC phases
and many of the most popular books on Six Sigma associate specific core statistical
methods with phases (e.g., Breyfogle, 2003, Harry 1998a, and Pande et al., 2001).
Third, the books and training materials associated with Six Sigma are relatively
vocational in nature. For example, Hahn, Doganaksoy, and Standard (2001) wrote
that the aim is not to train “statistical experts”. Fourth, the most important success
factors associated with six sigma were believed to be (1) top management
commitment and (2) multidisciplinary team training.
We begin by connecting the emphasis on monetary justification with achieving
top management commitment. As noted by Hahn and Hoerl (1998), money is the
language spoken by management and key to getting projects funded (paraphrasing).
37
Next, we connect greater specificity and relatively vocational materials with training
large numbers of practitioners from multiple disciplines. Intuitively, greater
specificity about what “should” be used and when it “should” be used in the context of
a project, combined with the omission of complicated theory, would seem appropriate
for motivating adult learners to use the methods.
Combining these observations, we conclude that widespread multidisciplinary
usage of statistical techniques is the implied goal of Six Sigma and its main
contribution to the business world. Academically, we see three related contributions.
First, the bottom line and multi-phase nature of Six Sigma has likely increased the
scope of research to embrace total projects and not just the portions associated with
the application of a single statistical method. This explains the interest on modeling
quality savings in Bisgaard and Feriesleben (2000) and why over one third of the
articles contained case studies.
Second, the relatively greater emphasis on specific core (SC) methods and
specific infrastructure (SI) has spawned considerable academic discussion with greater
specificity. For example, there is a substantial academic thread focused on what tools
“should” be learned and used by Six Sigma trained participants or “black belts” (e.g.,
see Hoerl 2001a and the related discussion in the Journal of Quality Technology).
While the discussion of training materials is not new to the quality literature, the
emphasis on relatively specific references to the phases of projects is somewhat new.
Third, Six Sigma has caused many people from multiple disciplines to become
aware of and apply statistical methods. It is perhaps remarkable that 69% of authors
38
of academically relevant publications had industry affiliations. While, in general, Six
Sigma practitioners have learned only standard methods, they constitute a large
potential market for research and, perhaps, new methods. Distinguishing features of
this market include that participants: (1) are relatively practical and focused on
business results, (2) need techniques for predicting the bottom line impacts of projects
before they embark upon them, and (3) apply statistical methods without, in general,
being experts in statistics.
In Section 3.10, we discuss the implications of these findings on the roles that
academics can most usefully play in relation to Six Sigma. Next, where the specific
infrastructure and core elements of Six Sigma fit into the quality performance model is
discussed.
3.9 Defining Six Sigma

Linderman et. al. (2003) emphasized the need for a common definition of Six
Sigma and proposed:
“Six Sigma is an organized and systematic method for strategic
process improvement and new product and service development that relies
on statistical methods and the scientific method to make dramatic
reductions in customer defined defect rates.”
Those authors further acknowledged that “the name Six Sigma suggests a goal” of
less than 3.4 defects per million opportunities (DPMO) for every process. However,
39
Linderman et. al. (2003) did not include this principle in the definition because, “Six
Sigma advocates establishing goals based on customer requirements.”
One concern with the Linderman et. al. (2003) definition of Six Sigma as a
“method” is that the definition leaves out philosophy and principles. For example,
Dean and Bowen (1994) defined quality management to include techniques and a set
of principles and practices. We suggest that emphasis on monetary gains in Harry
(1998a), Hahn et. al. (1999), Bisgaard and Freiesleben (2000), and other seminal
literature warrants the following addition: “The Six Sigma method only fully
commences a project after establishing adequate monetary justification.”
Montgomery (2001) argues that it is this focus on the bottom line that keeps
management interested while its predecessors like Total Quality Management (TQM)
are “dead”.
Virtually all popular books and training materials describe statistical methods
much more vocationally than standard statistical texts (Breyfogle, 2003, Harry, 1998a,
and Pande et al., 2001). Specifically, these books and training materials omit much of
the associated theory and include, in some cases, simplified versions of standard
statistical methods. Further, Hahn, Doganaksoy, and Standard (2001) wrote that the
related education goals are not to train “statistics experts” but only to give the
“knowledge essential to…obtaining business results.” We therefore propose to add
the following principle to the definition of Six Sigma: “Practitioners applying Six
Sigma can and should benefit from applying statistical methods without the aid of
statistical experts.”
40
Another concern with the Linderman et. al. (2003) definition is that it may be
unnecessarily vague. It can be argued from a review of the literature this vagueness in
the definition of Six Sigma is partly in an attempt to avoid controversy as most authors
have historically been practitioners advocating its use. We submit that there is
sufficient consensus within the Six Sigma literature to offer the following additional
details about the Six Sigma method in its definition:
The Six Sigma method for completed projects usually but not always
includes as its phases either Define, Measure, Analyze, Improve, and
Control (DMAIC) for process improvement or Define, Measure, Analyze,
Design, and Verify (DMADV) for new product and service development,
Design for Six Sigma (DFSS).
Widely read books such as Harry (1998a) and Pande et al. (2001) clearly imply
that this refinement is part of the definition of Six Sigma. Again, Linderman et. al.
(2003) points out that this definition may only be useful for improvement efforts on
complex challenging projects. For simple tasks; the so called “low hanging fruit” this
ridged approach may not create substantial benefits. The use of an “organized and
systematic method for strategic process improvement … that relies on statistical
methods … to make dramatic reductions in customer defined defect rates” would still
apply.
Harry (1998a), Pande et al. (2001), and others also imply that multiple
techniques are often used in applying Six Sigma. Therefore, the definition of Six
41
Sigma as “a method” complicates reference to the techniques used in its application.
We propose to refer to these techniques as “sub-methods” to clarify their relative
scope to that of Six Sigma. The existence of “sub-methods” helps to connote the idea
that Six Sigma is broader than its definition as a method might imply. Also, Six
Sigma then becomes more like a “practice” than a “core method” as defined by Sousa
and Voss (2002). Sousa and Voss (2002) also defined “infrastructure practices” as
those that create “an environment supportive of the use of core practices”. With these
definitions in mind, it becomes apparent that both of the principles above are
associated with what might be called specific “Six Sigma infrastructure” (SSI)
practices.
It is also unmistakable from reading the most popular books on six sigma
(Breyfogle, 2003, Harry, 1998a, and Pande et al., 2001), and others that there is a
strong attempt to associate sub-methods with specific phases of the application of Six
Sigma. For example, the application of gauge R&R would generally not be
considered appropriate in the Define phase. However, to our knowledge no specific
associations have currently received sufficient consensus to become part of the
definition of Six Sigma. Also, any specific set of associations could justifiably be
viewed as undesirable restrictions by some portion of Six Sigma users.
3.10 The Quality Performance Model

Much research has focused on the relationship of quality management
practices with the various aspects of firm performance (Voss and Sousa, 2002).
42
Garvin (1984) introduced a quality performance model to set up an empirical
examination of the separate effects of management practices on internal process
quality and product quality performance (QP) and their effects on operational
performance (OP) and business performance (BP). In reviewing the literature on Six
Sigma, it is helpful to place Six Sigma into this diagram.
In our definition of Six Sigma in Section 2.1, we identified principles and
methods associated with “Six Sigma infrastructure” (SSI) and “Six Sigma core” (SSC)
quality management practices. It was argued that the method of Six Sigma is itself a
quality practice while sharing some characteristics with a core method. Figure 10
shows the placement of these specific core practices and infrastructure in the Garvin
(1984) model.
Internal Process Operational

Quality Performance
Quality Management Practice
Six Sigma
Practice
Six Sigma
Infrastructure Business
Performance
Six Sigma
Product Quality
Performance
Figure 10 Extended quality performance model of Garvin.
43
3.11 Review of Empirical Evaluations of Six Sigma
In this section, the literature relating to the performance evaluation of Six
Sigma is briefly reviewed. Only a small fraction of articles in the database pertain to
an empirical model or evaluation with scope greater than estimating the savings
associated with a single case study. Table 4 below lists five of these articles with
reference to Six Sigma core (SSC) and Six Sigma infrastructure (SSI) practices. The
other acronyms used are referenced in the extended quality performance model in
Figure 10. The fourth article by Gautreau et al. (1997) does not make specific
reference to Six Sigma but was included because it addresses issues related to the
inclusion of specific methods in the context of quality projects. This seems relevant
given Six Sigma’s emphasis on specific core methods and business outcomes.
Table 4 contains a description of the roles each article plays for empirical
validation in relation to Six Sigma. Goh et al. (2003) examined stock performance
associated with announcements of Six Sigma programs and dates of quality awards.
They found hints of short-lived abnormal returns but no significant evidence of short
or long term returns. Another data driven meta-analysis that we found was Lee
(2002), which was based on survey data. The associated surveys indicated positive
self assessments of the value of the company’ own Six Sigma efforts. Also, to our
knowledge, the impacts of specific core sub-method selection on bottom line impacts
has not been studied empirically, with Gautreau, Yacout, and Hall (1997) providing
one of few relevant theory based modeling approaches.
44
Study Quality Study Main findings
performance method
model
Bisgaard and SSI→BP Economic Fraction
Feriesleben brake-even nonconforming and
(2000) analysis unnecessary activities can
model significantly influence cost
and reduce profit
Gautreau, Yacout, C→QP→BP Partially Decision based model
and Hall (1997) Observed of process improvement
Markov activities, e.g., do nothing
Decision or inspect and separate can
Process each be optimal depending
on assumptions
Goh et al. (2003) SSI/SSC→BP Hypothesis The majority of firms
(stocks) testing show positive returns after
announcing Six Sigma
programs but no statistical
significance was
established
Lee (2002) SSI/SSC→BP Survey of 106 Top management
firms commitment, project
selection, team leader,
training and the specific
tools used effect business
results
Linderman, SSC→QP/OP Goal-theoretic Types of goals effect
Schroeder, Zaheer →BP model quality and operational
and Choo (2003) performance that effect
business results
Table 4 Literature pertinent to evaluation of Six Sigma’s effects on firm performance.

(BP – Business Performance, C – Core, SSC – Six Sigma Core, SSI – Six Sigma
Infrastructure, OP – Operational Performance, QP – Quality Performance)
45
Considering the emphasis on modeling profits to justify each project, it is not
surprising at some attempts to provide meta-modeling tools, e.g, see Bisgaard and
Feriesleben (2000). Yet, Bisgaard and Feriesleben (2000) explore prediction of
product value under simple, generic assumptions. As they themselves suggest, more
research on related topics will likely be needed for broad applicability.
3.12 Synthesis and Future Research

In this chapter, a definition of Six Sigma and characterized the associated body
of academic literature is proposed. Also Six Sigma’s contributions to scholarly
research and its relationship to quality management theory was clarified. Reflecting
on the findings, we next return to the questions: (i) What is Six Sigma? (ii) What are
its impacts on operational performance? and (iii) What roles can academics usefully
play in relation to Six Sigma?
Six Sigma is defined as a method usually involving either Define, Measure,
Analyze, Improve, and Control (DMAIC) or Define, Measure, Analyze, Design, and
Verify (DMADV) as phases. This definition of Six Sigma as a method builds on the
one proposed by Linderman et. al. (2003). The inclusion of DMAIC and DMADV in
the definition is supported by the fact that 75% of introductory articles on Six Sigma
reference these structures. Two principles are included in this definition. The first
emphasizes attention to the bottom line in initiating projects. This was supported by
46
comments of seminal writers relating to how Six Sigma differs from Total Quality
Management (TQM), e.g., Harry (2001a) and Montgomery (2001). Also, bottom line
focus was mentioned by 24% of relevant articles as a critical success factor. The
second principle emphasized the training of non-statisticians with minimal theory.
This inclusion is based on remarks by Hahn, Doganaksoy, and Standard (2001) and
others about the goals of Six Sigma training. In addition, relatively frequent mention
of multidisciplinary training as a critical Six Sigma success factor (roughly 24% of
relevant articles) is found. Finally, popular books on Six Sigma such as Breyfogle,
(2003), Harry (1998a), and Pande et al. (2001) noticeably de-emphasized theory.
To a great extent, the financial impact of Six Sigma on operational
performance is well established. This follows because a defining principle of Six
Sigma is that such justifications are needed for each Six Sigma project. For example,
Dupont reported an expected profit from Six Sigma practices of over $1 billion
between 1999 and 2003 (Noble, 2001). Also, Goh et al. (2003) found hints of short
lived abnormal stock performance associated with the decisions to start Six Sigma
programs. Yet, those authors found no statistical significance in relation to these
claims nor evidence of long term effects. Those authors also included specific caveats
about the ability to connect Six Sigma programming effects at divisions with overall
parent company performance. Considering that failure to find significant effects does
not constitute proof, the conclusion is that more work is needed for a thorough
evaluation of the bottom line impacts of Six Sigma.
47
In Section 3.8, we identified three main types of contributions of Six Sigma to
academia embodied in the literature: (1) increased emphasis on complete case studies
compared with single sub-method applications, (2) new, relatively specific core and
infrastructure practices, and (3) the development of a large new market of industrial
non-experts who might be interested in practically oriented research and new methods.
Many authors have proposed areas for further research building on these contributions.
Also, Cooper and Noonan (2003), Linderman et. al. (2003), and Snee (1999) suggest
that, in general, too much research has been focused on descriptions of practice rather
than on theory development that is of use to managers and scholars.
We provide only a sampling of selected suggestions, which we divide into
three groups. First, Sousa and Voss (2002) highlighted the need for empirical
justifications of assertions of all types in the quality management literature. In the
context of Six Sigma, statements abound that are unsupported by objective evidence.
Examples include self reported profits, the effects of success factors, and advocacy for
Six Sigma in general. For example, as noted above, the impacts on stock performance
investigated by Goh et al. (2003) are not fully resolved. This suggests a need for
additional data collection and analysis to answer the important question of long term
impacts of decisions to adopt Six Sigma programs.
Second, while over 50% of the articles in the database either explicitly or
implicitly recommend Six Sigma programs, empirical study of the appropriateness of
implementing Six Sigma in specific business contexts has, apparently, not been
investigated. Related, largely unanswered questions include: How can data about any
48
specific company’s management, training programs, or environment be useful in
decision-making about the adoption of Six Sigma programs?
Third, Snee (1999 and 2000a) calls for research to help practitioners identify a
robust set of improvement tools to be used in conjunction with the DMAIC process.
The focus in these recommendations is not so much on new techniques as on refined
techniques associated with specific phases. However, as it was suggested in Section
3.8, new techniques might be relevant to Six Sigma practitioners who are often not
experts in statistics.
Fourth, additional modeling techniques to predict and evaluate the bottom-line
impacts of projects are needed. This follows because of the central importance of
profit related justifications in Six Sigma for initiating decisions on projects. Also,
Bisgaard and Feriesleben (2000) admit that the assumptions associated with their
models have limited scope and may ignore indirect savings. New models that are also
easy-to-use could be developed with broader applicability and improved prediction
accuracy.
It might also be useful to extend profit models to investigate the selection of
specific core methods or sub-methods in specific situations. These efforts could be
combined with empirical investigations to permit the development of prescriptive
models to aid practitioners from different disciplines select the most advantageous
techniques. This could build on research related to the most appropriate methods for
training black belts, e.g., in Hoerl (2001a), by associating the methods more
specifically to phases in a project and to situations.
49
In conclusion, it is proposed that Six Sigma is both a method and two
principles. These principles relate both to building and maintaining management
support and to fostering usage of methods among practitioners who are not experts in
statistics. Trends in the literature include an increasing academic participation and a
broader focus than solely on manufacturing. Only partial consensus about the factors
making Six Sigma effective is found. Opportunities for new research on Six Sigma
including creating more realistic project payback models, clarifying which techniques
are most applicable in which situations, and developing new statistical methods with
clear advantages for business.
3.13 Literature Review Summary

As the nature of research on Six Sigma is difficult to confine to specific
disciplines, the relevant material is scattered across various journals. The search
resulted in the identification of 201 articles published between 1990 and 2003.
Although this review cannot claim to be exhaustive, it does provide reasonable
insights into the state-of-the-art. It is felt that the results presented in this chapter have
several important implications.
Although the industry has an increased interest in Six Sigma implementation
and many companies have gained the profits and advantages from this disciplined
approach, the literature is limited and the research of the impacts of Six Sigma
implantation and factors contributed to Six Sigma success remain unclear. Many
50
articles on the impact analysis of operations performance do not mention the detailed
improvements in the operating areas such as scrap rate, rework rate and so on, but
focus on the overall bottom line impact [Breyfogle (2001), Noble (2001) and Lucas
(2002)]. Therefore, it is necessary to do a deeper and more detailed study in this area.
In addition, only a few articles were found that dealt with factors in the area of
success factor analysis to Six Sigma implantation. Even the existent studies are not
well integrated and the research is mostly anecdotal. Current concepts in the field of
Six Sigma are largely based upon case studies, anecdotal evidence and the
prescriptions of leading “gurus.” Consequently there is little consensus on which
factors are critical to the success of the approach. Most of the articles reported that top
management leadership is the main factor to Six Sigma success [Blakeslee (1999) and
Scalise (2001)]. However, many other factors affecting Six Sigma’s success are
important and need to be better documented.
Even so, there is substantial evidence that Six Sigma and other quality
management systems have a positive effect on the value of a company. Goh (2003),
Hendricks (2000) and Adams (1997) all found improvement of companies financial
performance with the implementation of quality initiatives. Several of the leading
authors have produced broad frameworks for implementing and sustaining competitive
advantages through quality management.
Although some authors have called for theoretic research (see Cooper and
Noonan 2003, Linderman et. al. 2003 and Snee 1999), too much research is focused
on descriptions of practice rather than on theory development that is of use to
51
managers and scholars. The attempt to build a theory of how and why Six Sigma
works is aimed at building a prescriptive model. From this, managers would be able to
identify which activities from which programs are more or less likely to be useful in
their situations, as well as which of their goals would be most affected. With the
future success of corporations riding on the outcome, there has been little theory to
explain the differences between successful and unsuccessful efforts.
52
CHAPTER 4
CASE STUDY BASED TRAINING
This chapter discusses one of the major themes of Six Sigma: training of
practitioners. From the literature, Six Sigma differs from most quality management
systems by its emphases on training what people need to know to implement
improvement programs effectively. A subset of this topic is the training of non-
statisticians in the use of statistical tools. A major concept then becomes how to
effectively train participants in the necessary tools and techniques of Six Sigma. In
this chapter, we describe one case study and associated exercises used in senior and
introductory graduate level engineering courses on SPC and DOE at The Ohio State
University. The role of the case study exercises in the context of material covered in
the lectures is described. Section 4.2 provides an overview of the case study structure.
In Sections 4.3 a case study and associated in-class exercises is presented. Finally, a
description of the student feedback to the case based teaching approach is reviewed.
53
4.1 Introduction
One constant tread in the literature dealing with Six Sigma is the training and
the limited chose that non-statisticians have to make. For thousands of participants at
the lowest “green belt” level of accreditation, one of the main benefits of “Six Sigma”
training is that it simplifies (through restriction) the sequence and choice of available
techniques to apply to a particular case. Therefore, the value of the Six Sigma
movement derives partly from standardization of problem solving methods and partly
in how it guides people to suggest which techniques to apply in which order to an
improvement project.
Drawing a box as a way of teaching judgment is based on one of the most
successful methods that is used to teach skills: breaking a task into small bits and
presenting it to the student with practical, logical steps that he or she can understand
and apply. By simplifying the decision-making model and helping our students to
define and operate within a well-defined “box” of their limitations, we have a practical
way of teaching good judgment.
The first step is to simplify the decision-making model. The purpose of
decision making is to reach the right decision at the right time. We can make the
whole decision-making process practical by applying it to various actual situations.
Now, rather than trying to teach an ambiguous “good judgment” concept, we are
giving our students a practical route to good decisions.
54
Drawing a box for our students involves expanding on the familiar concept of
personal knowledge and capabilities. The first step in the model is the ability of a
student to operate within his or her own capability. The second step involves looking
at a given situation, evaluating current skills, and setting boundaries for that specific
situation. In essence, you’re drawing a box of operating parameters for your student in
that situation. The third and final step is to insist that he or she always operate within
that box.
By drawing the box before the situation a good decision was made before a
crisis forced the student’s hand. Everyone’s box will be different, since we all have
different levels of experience and ability. As proficiency is gained, the size of the box
will change.
Teaching decision making is a daunting- some would say impossible task, but
one that is imperative. By giving students the tools and ability to draw their own box,
we are providing the foundation for good decision making.
Six Sigma has caused many people from multiple disciplines to become aware
of and apply statistical methods. It is perhaps remarkable that 69% of authors of
academically relevant publications had industry affiliations. While, in general, Six
Sigma practitioners have learned only standard methods, they constitute a large
potential market for research and, perhaps, new methods. Distinguishing features of
this market include that participants: (1) are relatively practical and focused on
business results, (2) need techniques for predicting the bottom line impacts of projects
55
before they embark upon them, and (3) apply statistical methods without, in general,
being experts in statistics.
There is increasing recognition in the statistics, engineering, and education
literatures that exercises based on case studies can play an important role in
reinforcing understanding as well as motivating students to learn and apply statistical
process control (SPC) and design of experiments (DOE) techniques, e.g., see Alloway
(1993), Barton and Nowack (1998), Cobb (1992), Nolan and Speed (1999),
Petruccelli, Nandram, and Chen (1995). There is also abundant evidence, e.g., Czitrom
(1999), that most practicing engineers continue to display a lack of motivation to
apply the technology they have learned and in general see their class experiences as
largely irrelevant to their professional life.
Engineering students generally gain deeper understanding of the theory once
they are exposed to the realistic engineering contexts in which the related methods are
applied. The growing literature on the benefits of case-based learning approaches in
engineering, Howell (1996), McKeachie (1993,) and Wankat (1993) gives us insight
into restructuring traditional class outlines into a more productive approach. For these
reasons it is important to devote a substantial fraction of class time (roughly 20%) to
in-class case study based exercises that emphasize the other forms of knowledge and
reinforce the lecture materials. These exercises also seem to aid in establishing
linkages with the students’ lives outside of the university.
A major challenge in using case studies in classes has been that the examples
in standard textbooks generally seem highly contrived to the students and divorced
56
from real life contexts. As described by Bisgaard (1998) and in our own experience, it
is important that the case study examples used in the course are impressive to the
students in their relevance to their own possible careers. Therefore, we have found that
it is helpful to use exercises based closely on case studies from local companies in
which the application of the techniques taught in the course played a key role in
achieving a measurable success. Perhaps the most important purpose of these
exercises is that the students can visualize themselves in the role of facilitating real
world successes.
The case study based exercises were introduced into the two combined senior
and first year graduate level courses at The Ohio State University. The classes meet
twice a week for 10 weeks, each period is for one hour and twenty minutes. Many of
the students have had two quarters in statistics. Topics addressed in the SPC courses
can be divided into four categories: (1) a survey of quality techniques and a review of
probability and inference relating to continuous random variables, (2) control charts
for continuous random variables, (3) review of probability relating to discrete random
variables and attribute control charts, and (4) acceptance sampling and assorted topics
including ISO9000 and “six sigma” methodology. The topics in the DOE class are: (1)
t-testing and ANOVA, (2) screening experiments, (3) response surface experiments,
and (4) robust optimization. In both classes, each category corresponds to
approximately five class periods and one homework assignment. The case study
exercises are timed on the days after each of four assignment due dates, so that the
students are responsible for knowing the material involved.
57
The case studies are true but for proprietary concerns some are modified in
order to maintain the confidentiality of the companies involved. Each case study is
selected from area industries to provide relevant experience and important evidence of
the practicality of the statistical techniques described in the lecture and textbook.
4.2 Case Study Exercises

Each case study has five components: 1) an introduction, 2) a problem
statement, 3) the background information relating to prior attempts to solve the
problem, 4) questions for the students, and 5) the team implementation of course
techniques and results from the actual study. The exercise begins by handing out
written materials summarizing components 1-4.
In the introduction, we use 5 minutes of lecture time to provide a brief
description of the industry in question and some of the standards, issues and
terminology used so the students have some familiarity with the subject. The
introduction also serves the important role of establishing that the students might
realistically find themselves in similar situations in their future careers. In the problem
statement, we use 5 minutes of lecture time to describe the goal, e.g., 5% defective
products from a particular process, and the associated business objective, e.g., avoid a
$2.5 million capital expenditure to meet increased product demand. In our oral
description of the problems, we encourage creative analyses and emphasize that there
is no one solution. An additional 10 minutes are used to describe supplemental
background information which includes the relevant data that was available to the
58
company at the time of the team intervention involving class methods and sufficient
information such that the students can be expected to provide their own analyses and
recommendations. Our goal is to present the problem such that the students should be
able to easily identify themselves as playing the role of consultant and advocate of
SPC and DOE techniques in a real-world context.
Next, we supply abbreviated lists of potential investigations and leading
questions. These questions involve both critiquing the results of earlier investigations
and developing recommendations for future study. The students are divided into
groups of 4-6 and given a 30-minute exercise period to answer the questions. Students
can ask technical questions and when relevant we sketch results on the blackboard
including, e.g., formulas for sample size estimation and additional details requested by
the students. From group input, we use 15 minutes class time to construct composite
answers to questions in class and write these answers on the blackboard. During this
process, we also provide limited feedback to increase the sophistication of the
proposed approaches while preserving the students intent. After the answers from the
class are on the board, we use the remaining class time to describe the actual
company’s actions and results.
4.3 Sample Case Study “Improving the Yields of Printed Circuit

Boards”
This case study exercise is used in both the SPC and the DOE classes and
illustrates how with a small number of quality technology tools highly valuable
59
information can be learned. This case study involves the use of Pareto diagrams, p
charting, hypothesis testing, and fractional factorial designs. The study also illustrates
potential dangers from one-factor-at-a-time experimental approaches.
4.4 The Problem Statement

In early 1998, an electronics manufacturing company introduced to the field a
new advanced product that quickly captured 83% of the market in North America.
During initial production period, yields (the % of product requiring no touchup or
repair) had stabilized in the 70% range. In early 1999 the product was selected as a
major equipment expansion in Asia. In order to meet the increased production
demand, the company needed to purchase additional test and repair equipment at the
cost of $2.5 million or the first test yield had to increase to above 90%. The latter was
the preferred situation due to the substantial savings in capital and production labor
cost thus, the problem was how to increase the yield in a cost-effective manner.
4.5 Background Information

A team of highly regarded engineers was assembled from various design and
manufacturing areas throughout the company. Their task was to recommend ways to
improve the production yield based on their prior knowledge and experience. Table 5
gives the weekly first test yield results for the 16 weeks prior to the team’s activities
with production volume of 1500 units per week.
60
Week Yield Week Yield Week Yield Week Yield
1 71% 5 87% 9 66% 13 63%
2 58% 6 68% 10 70% 14 68%
3 69% 7 71% 11 76% 15 76%
4 77% 8 59% 12 82% 16 67%
Table 5 The yields achieved for 16 weeks prior to the initial teams activities.
Based on their technical knowledge and experience, the assembled
improvement team reviewed the design and production process. They created a list of
15 potential process and design changes for improvement based on their engineering
judgment and anecdotal evidence. With this list in hand, they proceeded to run various
single factor experiments to prove the validity of their plan. Due to perceived time and
cost constraints, only one run of each factor was completed using a sample of 30 units.
Results were compared with the yield from the previous 13 weeks of production. Each
unit could only succeed or fail to meet specifications. Factors that showed a yield
decrease below the 13 week average were discarded along with the experimental
results. Table 6 shows the results of the experiments with yield improvements
predicted by the team.
Based on their analysis of the circuit, the above experimental results and past
experience the improvement team predicted that a yield improvement of 18% would
result from their proposed changes. All of their recommendations were implemented
at the end of week 17. Table 7 gives the weekly first test yields results for the six
weeks of production after the revision.
61
FACTOR YIELD IMPROVEMENT
Replace vendor of oscillator 8.5%
Add capacitor to transistor 8.5%
Add RF absorption material 5.5%
New power feed layout 5.5%
Increase size of ground plane 2.5%
Lower residue flux 2.5%
Change bonding heat sink 2.5%
Solder reflow in air vs. N2 2.5%
Temperature of solder tips 2.5%
Table 6 The initial team's predicted yield improvements by adjusting each factor.
Week Yield Week Yield

17 62% 21 40%
18 49% 22 41%
19 41% 23 45%
20 42%
Table 7 The performance after the implementation of the initial team's

recommendations.
Reviewing the data we see that the yield actually dropped 29%. On week 22 it
was apparent that the proposed process changes were not achieving the desired
outcome. Two additional engineers, with exposure to quality improvement tools
through continued education at local universities and company-sponsored seminars,
were added to this project.
62
4.6 In-class Exercise for PCB Study
In our SPC classes, we then give the students the following questions based on
the premise that the company is hiring them as a quality and manufacturing consultant.
(1) Critique the methods used by the engineers for predicting improvements, e.g., is
there any evidence that they increased the yield? Roughly how many samples would
they need for the standard deviation of the estimated yield to approximately equal 5%,
i.e., they could begin to resolve differences of approximately 5%? (2) What additional
procedures would you recommend to measure the process capability? Include in your
answer what specific types of data or information would you request or collect. (3)
What addition procedures would you recommend to aid in improving the process?
In DOE classes, we ask the students: (1) What size of experiment should they
start with? (2) How should they determine which runs to perform? A representative
from each team presents their group’s answers in class, which are written on the
board. The entire class then comments on the reviews and coherent consulting answers
to the questions are created. In the real study, Pareto Charts were used to direct the
investigation, p-charts were used to measure the common cause variation, and formal
approaches were used to improve the process. Therefore, a wide variety of class
techniques are relevant, presenting life-like ambiguities for the students. If we do not
finish and/or if we feel it is appropriate we include unanswered questions as part of the
homework assignments. After the class discussion we relate in detail what actually
occurred in real life.
63
4.7 The Team Actions and Results
The second team’s first step was to construct a yield attribute control chart (a
yield chart or 1- defective chart “1-p”) with the knowledge of the process change date
(Figure 1). The engineers were able to tell that most of the fluctuations in yield
observed before the team implemented their changes were, as Deming calls it,
common cause variation or random noise. The engineers’ first decision was to revert
back to the original, documented process in place during week 16 since the evidence
that had supported these changes was probably due to random noise within the
process. Table 8 gives the weekly test yields for the five weeks after this occurrence.
This had the effect of restoring the process to its previous in control state with yields
around 75%. The increase in yield shown on the control chart (Figure 11) during this
time frame was discounted as random noise or the “Hawthorn” effect since no known
improvement was implemented. By Hawthorn effect, we mean an effect caused by our
attention to and study of the process. Next the procedure of Pareto charting was
applied to help visualize the problem, Table 9.
64
100%
90%
UCL
80%
70%
Y ie ld CL
60%
50%
LCL
40%
30%
0 10 20 30 40
Subgroup
Figure 11 Control chart for the entire study period.
Week 24 25 26 27 28
Yield 62% 78% 77% 75% 77%
Table 8 The yields for the 5 weeks subsequent to the initial intervention.
65
ACP –30 kHz 6.7%
ACP +30 kHz 5.0%
ACP –60 kHz 2.9%
ACP +60 kHz 2.8%
ACP –90 kHz 2.7%
VDET 1950 mHz 2.4%
VDET 1990 mHz 2.0%
ACP +90 kHz 1.4%
Power supply 1.2%
Gain 1.0%
Output return .8%
Bias voltage .5%
Max current .2%
Other .4%
Table 9 The data for a Pareto chart of the data from week 1 to week 16.
It can clearly be seen from a Pareto graphs that 21.5% of the 30% of total
defects was attributed to one parameter called “ACP”. The engineers concentrated
their efforts on this dominant defect code. The team of experts was reassembled with
the addition of representation from the production workers to identify what variables
might cause this defect. Four factors identified were: (1) transistor performance
distribution (high end of spec or low end of spec), (2) transistor mounting (socket or
solder), (3) input circuit tuning (centered or low end of spec), and (4) transistor heat
sink type (current or new configuration). This last factor was added at the request of a
representative from production. This factor was not considered important by most of
the engineering team. The two lead engineers decided to include this factor as the
66
marginal cost of adding it was small. An eight run fractional factorial experiment was
conducted as shown in Table 10. Each run corresponded to 350 units.
STD Run Yield

7 5 92.7
4 6 71.2
6 7 95.4
2 9 69.0
3 10 72.3
5 11 91.3
8 13 91.5
1 14 79.8
Table 10 The data from the fractional factorial design in four factors.
Using ANOVA and main effects plots, the heat sink type emerged as the main
effect. The two engineers went forward based on the DOE results and recommended
that the process change to the new heat sink. This was implemented during week 29.
Table 11 gives the weekly yield results for the period of time after the recommended
change was implemented. Using the yield charting procedure, the engineers were able
to confirm that the newly designed process produced a stable first pass yield in excess
of 90% thus avoiding the equipment purchase and saving the company $2.5 million.
The main points that we emphasize in our wrap-up of the exercise are the
following. First, the technology of fractional factorial experiments permitted the
fourth factor to be added without additional runs being needed. The importance of this
factor was controversial because it had been suggested the operators. If fractional
67
factorial had not been used, then the additional costs would likely have precluded its
inclusion and the important subsequent discovery. Second, the case also illustrates the
important benefit of a “balance experimental design”, i.e., half of a total number of
experiments are at each level of each factor providing the maximum power to identify
small effects.
Week Yield Week Yield Week Yield
29 87% 34 94% 39 91%
30 96% 35 90% 40 93%
31 94% 36 90% 41 89%
32 96% 37 86% 42 96%
33 91% 38 92%
Table 11 The confirmation runs establishing the process shift/improvement.
4.8 Student Feedback

Students seem to identify with the case study "stories" much better than the
traditional lectures. Before implementation of the case study exercises, end-of quarter
evaluations by students for the two classes had been below the average end-of-quarter
class evaluations in the department. After the implementation of the case study
approach and with minimal other changes, the evaluations climbed well above
average. We attribute this dramatic turn-around mainly to the introduction of the case
68
based method. Further, we conjecture that presenting applications of course methods
in which literally millions of dollars are saved effectively motivates students
concerned about job security and advancement in the highly competitive industrial
marketplace.
It is our belief that these results should generalize to other statistics, operations
research, and engineering design theory related courses. This follows because all of
these courses are fundamentally concerned with the students' usually voluntary
application of formal techniques to solve engineering problems. Since this requires
discipline and motivation, we feel that both practicing the application of techniques in
relatively real-world contexts and disseminating success stories are critical to
maximizing benefits from applying the techniques.
69
CHAPTER 5
THE CHALLENGE OF SIX SIGMA PROJECT
MESO-ANALYSIS
5.1 Introduction
One outcome of the wide spread acceptance of Six Sigma and with its
systematic program is the fact that many databases describing the performance of
improvement projects and the methods used have been generated. Even with its
apparent highly visible success, one area of investigation is the use of this new source
of data, the industry Six Sigma database, for additional insight into the management of
corporate business improvements. How might this database be used to help
operational management in the area of quality management practices? Three levels of
decision-making can be defined:
Micro – dealing with individual statistical methods.
Meso – supervisor level decision-making about method selection and timing.
Macro – related to overall quality programs and stock performance.
70
Martin (1982) pointed out that the availability of certain type of data might
disproportionately influence the problems that are investigated as well as the
conclusions drawn. Reviewing the literature with historical data sources reveals a
large portion concerning macro level decision-making, the question to implement a
Six Sigma methodology at a company, Bisgaard and Freiesleben (2000), Chan and
Spedding (2001), Gautreau, Yacout and Hall (1997), Yacout and Gautreau (2000), and
Yu and Popplewell (1994). This is mostly based on single case studies and anecdotal
information. A second large grouping of articles deals at a micro level, concerning
component tools and techniques for green belts and black belts. These are the terms
used for the individuals working on project implementation. Little work is published
that relates to the meso level of managing and decision-making of Six Sigma, the mid-
management operational decisions regarding Six Sigma and improvements to the
system, Linderman et. al. (2002). The uses of these databases are likely being ignored
at most companies for at least two reasons. First, there is little assistance form
academics in how to make sense of them. Second, the people with the most statistical
expertise are involved in the individual projects and not in cross project evaluation.
Most managers are not statisticians and need help in making sense of the data now
available to them.
According to Juran and Gryna (1980) the activities in companies that assure
quality can be grouped in three processes: quality planning, quality control and quality
improvement. Quality improvement consists of the systematic and proactive pursuit of
improvement opportunities in production processes to increase the quality levels.
71
Typically, quality improvement activities are conducted in projects. This proactive and
project wise nature distinguishes improvement form quality control, which is an on-
line process that is reactive in nature. In Harry (1994) all things are a process. A
central belief of Six Sigma is that the product is a function of the design and the
manufacturing process which must produce it. This is symbolized as Y=f(X), where Y
is characterized as dependent, output, effect, symptom and its role as “to be
monitored”. The X is described as independent, input, cause, problem, and its role as
“to be controlled”. The view is that the emphasis should shift from monitoring Y to
controlling the relevant Xs.
With Juran and Harry in mind, Six Sigma can be viewed as a process and
subject to the same controls and improvement objectives of other processes. Against
this background, the purpose of this study was to look at this growing database in a
way that could help management better-run improvement projects.
5.2 Potential Use of Database

The use of this growing database of project related quality improvement
activities could be useful in the empirical study of some important research questions.
Potential research topics include: the health of a given company’s quality system,
modeling Six Sigma, or the optimality of selection and ordering component methods
associated with Six Sigma. Researchers focus on what they have data and tools for.
Now, new data sources and the associated ability to ask and answer new types of
questions are more readily available. For example, “Is my quality system
72
out-of-control?” “Which method would lead to greatest expected profits in my case?”
“Under what circumstances does it make business sense to terminate a project?” If
these kinds of questions can be systematically explored in the Six Sigma discourse
then important lessons can be learned regarding investment decisions. Moreover, there
will be an increase in critical writings and inquiry on this subject, which will add
depth and meaning to Six Sigma in organizations. To help in this study empirical data
on multiple quality/cost improvement projects at a medium size U.S. manufacturing
company was collected over a 30-month period starting in January 2002. A total of 39
projects were included in the study.
5.3 Database Example

The company used for study is a U.S. based mid-western manufacturing
company, which manufactures components for the aerospace, industrial and defense
industries. It has approximately 1000 employees, annual sales of $170 million, with
six factories located in five states. The data is all derived from one of its six
manufacturing sites. This one site has 250 employees with sales of $40 million.
Quality improvement and cost reduction are important competitive strategies
for this company. The ability to predict project savings and how best to manage
project activities would be advantages to future competitive posture and the long-term
strength of the company.
73
5.4 General Procedure
Over the course of this study data was collected on 20 variables. Two
additional dependent variables are tabulated in the data, which are functions of some
of the other variables. The two variables are Profit, which is Actual Savings minus
cost and Formal Methods (FM) which is any combination of Charter, Process
Mapping, Cause & Effect, Gage R&R, DOE or SPC. Table 12 lists and describes the
variables tracked during this study.
Data was collected on each project by direct observation and interviews with
team members to determine the use of a variable such as DOE or Team Forming. No
attempt was made to measure the degree of use or the successfulness of the use of any
variable. We only were interested if the variable activity took place during the project.
A count was maintained if an activity was used multiple times such as multiple DOE
runs i.e. a screening DOE and an optimization DOE would be recorded as 2 under the
variable heading.
Expected Savings and Actual Savings are based on an 18 month period after
implementation. The products and processes change fairly rapidly in this industry and
it is standard company policy to only look at 18 month to evaluate projects. These
calculations were based on the current monthly production forecast.
Costs were tracked with existing company accounting procedures. All projects
where assigned a work order for the charging of direct and non-direct time spent on a
specific improvement activity. Direct and non-direct labor where charged at the
74
average loaded rate. All direct materials and out side fees (example, laboratory
analysis) where charged to the same work order to capture total cost.
The raw data derived form this study in tabulated in Appendix B.
5.5 Data Collection

The empirical analysis for this exploratory study is based on the data collected
at one manufacturing site over a 30 month time frame. Not all programs followed
DMAIC nor was it the intent. Our intent was to observe unbiased projects in a real
world setting to start to develop a model of improvement activities. Observation and
interviews determined the use or non use of a tool. No measure of the degree of use
was attempted. We only indicated whether or not a tool was used and how often. The
number of times a tool was used during a project was recorded and listed in the data
summary, Appendix B. These projects did not include all activity but where judged to
be project unbiased by the author. As a manager at the facility where the study was
conducted one author could influence the activities. The studied projects where
selected because they were judged to be unbiased by the author.
75
Field Description
Expected savings An estimate of the projects saving over an 18
month period based on the current business
forecast.
Expected time An estimate made at the start of a project as to the
time needed to complete the project
s-short less than 3 months
m-medium between 3 and 9 months
l- long over 9 months
M/I management or self initiated Whether the project was initiated by management
or initiated by team members
Assigned or participative Whether the project was assigned to a team by
management or the members actively chose to
participate
# people Number of team members
EC Economic analysis A formal economic analysis was preformed with
the aid of accounting to identify cost and cost
brake allocations
CH Charter Formally define project scope, define goals and
obtain management support
PM Process Mapping Identify the major process steps, process inputs,
outputs, end and intermediate customers and
requirements; compare the process you think
exists to the process that is actually in place
CE Cause & Effect Fishbone diagram to identify, explore and display
possible causes related to a problem
GR Gage R&R Gage repeatability and reproducibility study
DOE A multifactor Screening or optimization design of
experiment
SPC Any statistical process control charting and
analysis
Table 12 Continued on next page.
76
Table 12 Definition of variables (Continued from last page)
DC Documentation Formally documenting the new process and or

setting and/or implementing a defined control
plan
EA Engineering analysis Deriving conclusions based solely on calculations
or expert opinion
OF one factor experiment A one factor at a time experiment
Time Actual time the project took to completion
Profit A current estimate of the net profit over the next
18 months after implementation based on the
actual project cost and actual savings
Actual Savings A current estimate of the savings over the next 18
months after implementation based on the new
operating process and current business forecast
Cost The actual cost as tracked by the accounting
system based on hours charged to the project,
material and tooling, equipment
Formal Methods A composite factor, if multiple formal methods
were used in a project this was positive
Table 12 Definition of variables
As stated in section 3.11 one of the main principles of Six Sigma Harry
(2001a) and Montgomery (2001) is the emphasis placed on the attention to the bottom
line results in initiating projects. In the literature reviewed, bottom line focus was
mentioned by 24% of relevant articles as a critical success factor. Profit, which is a
combination of cost and actual savings, than becomes the dependent variable. The
remaining eighteen variables are the independent variables. Of these variables, Process
Mapping, Cause & Effect, Gage R&R, DOE, and SPC are related to Formal Methods.
77
These 39 improvement projects generated a total of $4,385,099 in net savings
(profit). The projected savings form these projects ranged rorm $1600 to $2,200,000
with actual net savings of -$220,000 to $3,874,500 over the first 18 moths after
implentation. Cost to implement the projects ranged from $1,000 to $325,500. Fifteen
of the thirty nine projects resulted in negative net savings as reported. Table 13
presents descriptive statistics for project duration, number of team members and
profits.
Factor Median Low High Average

Time 8 months 1 month 30 months 10 months
Profit $3,025 -$220,000 $3,874,500 $112,438
Team 1 1 9 3
members
Table 13 Summary statistics for the 39 projects, duration team memberrs and profits
We looked at directed assignments vs. self-initiated, as one factor identified in
the literature was team participation. Of the projects, 72% were initiated by
management directive. Management assigned the team members in 48.7% of the tasks.
The other teams were self-initiated by the team members. Additional characteristics of
the study samples are given in Table 14.
78
Item Number in Percent of the Study
Study Sample
Economic Analysis 15 38.5%
Formal Charter 28 71.8%
Team Forming Exercise 12 30.8%
Process Mapping 21 53.8%
Cause & Effect 11 28.2%
Gage R&R 7 17.9%
Design of Experiment 11 28.2%
Statistical Process Control 7 17.9%
Formal Documentation 28 71.8%
Engineering Analysis 29 74.4%
One Factor Experiment 10 25.6%
Table 14 Characteristics of the Study Variables
Three potential uses of this database were investigated using analyses by
regression, Markov Decision Process, and SPC. Each analyses tool was used to look at
a different problem that this database might be useful in researching.
5.6 Statistical Process Control

A control chart is one of the primary techniques of Statistical Process Control
(SPC). The control chart is a very useful process monitoring technique; when unusual
sources of variability are present, sample averages will plot outside the control limits.
Control charts can be used to monitor the health of a process improvement
79
methodology (Six Sigma) or changes to a process such as did training have a positive
effect on process improvement activities.
The first step in deriving a process control chart is to check the assumption of
normality. Figure 12 in a normal probability plot of the project data. It can be seen
from an examination that the data is comprised of two populations and two outliers.
2.5
2
1.5
1
0.5
z-score
0
-2000000 -0.5 0 2000000 4000000
-1
-1.5
-2
-2.5
Profit ($)
Figure 12 Normal Probability Chart for Six Sigma Projects.
With the data base in the example the logical subgroup size is n=1. With only
one measurement per subgroup (a project) a subgroup range can not be calculated. The
data is comprised of a small number of non-normal observations. The exponentially
80
weighted moving-average (EWMA) control chart is typically used with individual
observations (Montgomery 1997). The exponentially weighted moving average is
defined as:
Z i = λx i + (1 − λ ) Z i −1 (5.1)
where 0 < λ ≤ 1 is a constant and the starting value is the process target so that
Z0 = μ0 (5.2)
The average of preliminary data can also be used such that
Z0 = x (5.3)
The EWMA control chart is constructed by plotting Zi versus the sample
number i with the following control limits and center line:
UCL = μ 0 + Lσ
λ
(2 − λ )
[1 − (1 − λ ) ]
2i
(5.4)
CL = μ 0 (5.5)
LCL = μ 0 − Lσ
λ
(2 − λ )
[1 − (1 − λ ) ]
2i
(5.6)
According to Montgomery (1997) values of λ in the interval 0.05 ≤ λ ≤ 0.25
work well, with λ=0.05, λ=0.10, and λ=0.25 being popular. L values between 2.6 and
3.0 also work reasonably well. Hunter (1989) has suggested values of λ=0.40 and
L=3.054 to match as closely as possible the performance of a standard Shewhart
control chart with Western Electric rules.
One of the important variables that might be of interest in monitoring the
health of a quality improvement methodology like Six Sigma is savings per project.
81
From the database under investigation and the above equations a control chart in can
be generated. We start with plotting the first 25 points to obtain the control limits as
shown in Figure 13. One out of limit point was discarded after the derivation of this
chart. This one project was a DFSS (Design for Six Sigma) vs. a DMAIC project and
as such was unique. This chart was constructed with based on Hunter (1989) with
λ=0.40 and L=3.054.
2000000
1500000
1000000
Profit ($)
500000 UCL
zi
0 LCL
1 3 5 7 9 11 13 15 17 19 21 23 25
-500000
-1000000
-1500000
Project
Figure 13 EWMA Control Chart for first 25 Six Sigma Projects.
82
EWMA Chart for Profit
150000
100000
50000
Profit $
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
-50000
-100000
Six Sigma Project
Figure 14 EWMA Control Chart for Six Sigma Projects.
The projects are listed in rough chronological order therefore Figure 14 can be
considered a time series graph. One major outlier was removed from the plot.
Traditionally since their introduction by Walter Shewart, control charts have been
used to characterize the behavior of a time series. If a time series displays
unpredictable behavior, then the underlying process which gives rise to the time series
is said to be out-of-control. Likewise a process will be said to be in control when,
83
through the use of past experience, we can predict, at least within limits, how the
process will behave in the future, Shewhart (1931).
This distinction between predictability and unpredictability is important
because prediction is the essence of doing business. Predictability is a great asset for
any process because it makes the manager’s job that much easier. The fact that a time
series remains within the computed limits, and that there is no obvious trend, not any
long sequences of points above or below the central line, suggests that a process may
display a reasonable degree of statistical control. In this case, the process has an
average profit per team of $112,438 and values have varied form a low of -$220,000
to a high of $3,874,500
Of special interest are the last seven projects that took place after Six Sigma
training provided by Honeywell and Raytheon. This would tend to indicate “proof”
that training helps in project improvement. This is limited data but future study with
this same technique could be used to verify if training contributed to a fundamental
change in the process.
5.7 Regression
The popularity of ordinary linear regression models is attributable to its low
computational costs, its intuitive plausibility in a wide variety of circumstances, and
its support by a broad and sophisticated body of statistical inference. Given the data,
the tool of regression can be employed on at least three separate conceptual levels.
84
First, it can be applied mechanically, or descriptively, merely as a means of curve
fitting. Second, it provides a vehicle for hypothesis testing. Third, and most generally,
it provides an environment in which statistical theory; discipline-specific theory and
data may be brought together to increase our understanding of complex physical and
social phenomena. Simply put regression can be used to derive a model that will help
predict average performance for a new combination of inputs and/or to understand
which input changes cause changes in average outputs.
The familiar multiple regression equation is represented by equation 5.7 below:
yest(βest,x) = f1(x)΄βest (5.7)
Where f1(x) is a vector of functions only of the system inputs, x.
In most regression analysis the coefficients of the independent variables are the
major point of interest. The slope of the regression line measures the relationship of
each independent variable to the dependent variable of interest; the regression
coefficients are these slopes. The most interesting parameter in a linear model is
usually the slope. In this equation the slope of Y on X1 has a constant value across the
range of X2. For complex systems researcher may notice that regressions based on
different subsets of the data produce very different results, raising questions of model
stability. This may result in confusing or conflicting recommendations. This may be
caused by interaction between the independent variables and the response. Terms
involving products e.g. f1,7(x) = x1x7, are called interaction terms, a relationship
represented by equation 5.8 below:
yest = β1X1 + β2X2 +β3X1X2 + β0 (5.8)
85
The effect of this first order interaction is to shift the regression line based on
the level of the independent variables producing a family of results. The X1X2
interaction signifies that the regression of Y on X1 depends upon the specific value of
X2 at which the slope of Y on X1 is measured. There is a different line for the
regression of Y on X1 at each and every value of X2. The regressions of Y on X1 at
specific values of X2 form a family of regression lines. An analogy to chemistry, this
has the effect of a buffer. That is while X1 is positively related to Y the strength of this
relationship depends on the level of X2.
Much of the literature on Six Sigma converges on the importance of
management commitment, employee involvement, teamwork, training and customer
expectation among others in Six Sigma implementation. A number of research papers
have been published suggesting key Six Sigma elements and ways to improve the
management of the total quality of the product, process, corporate and customer
supplier chain.
However, most of the available literature considers different factors as an
independent entity affecting the Six Sigma environment. But the extent to which one
factor is present may affect the other factor. The estimate of the net effect of these
interacting factors is responsible for the success of the Six Sigma philosophy.
Quantification of Six Sigma factors and their interdependencies will lead to estimating
the net effect of the Six Sigma environment. The authors are not aware of any
publication in this direction.
86
5.8 Regression Data Analysis
The data was analyzed using a commercial software regression modeling
program. The software is a polynomial regression program based on Microsoft®
EXCEL spreadsheets and has the capability of simple regression models, multiple
regression models or custom modeling. The software regression model features permit
the user to select models automatically or manually. They facilitate fitting high order
terms (i.e. second or third order terms).
The motivation for this study was to evaluate the use of an emerging database
due to the expanded use of Six Sigma methodologies in management decision-making.
We utilized regression techniques which are readily available to management in the
business environment for the analyses. The data was analyzed using a commercial
polynomial software regression modeling program. One of the models that was
derived from the data was an interaction model between the use of Formal Methods
and Engineering Analysis. This model illustrates the possible use of these techniques
and appears helpful in validating the importance of estimating expected savings and
then deciding how formal to be. The model can be represented by equation 5.9:
Profit = 79931 + 0.848*Exp Savings - 76708.7*FT + 1258.21*EA +
2.2*Exp Savings*FT - 4.04*Exp Savings*EA (5.9)
Where:
Exp Savings is the amount of expected savings in dollars,
87
FT is the number of Formal Methods used,
EA is the use of engineering analysis
And Profit is the response.
Figure 15 is the main effects plot produced by the software for this model. It
can be seen that as the expected savings increases the model would suggest a greater
reliance on the use of formal methods. Likewise, based on this limited database, as the
expect savings decrease; a greater reliance on engineering analysis should be used.
The results of the regression model are presented in Table 15, with the
summary statistics given in Table 16. A 3-D plot of the results is included in Figure
16. Last, normal scores vs. residuals are found in Figure 17.
Pseudo Main Effects Plot - Model Predictions
12000000
10000000
8000000
Predicted Response
6000000
4000000
2000000
0
Exp Savings:1600
Savings:2.2e+006
Savings:1.1008e+
TF:0.5
FT:3.5
Time:15.5
k2:I
k1:L
DOE:0
SPC:0
SPC:1
SPC:2
DC:0
DC:1
DC:2
EA:0
EA:2
EA:4
OF:0.5
#people:1
#people:5
#people:9
EC:0.5
CH:0.5
PM:0
PM:1
PM:2
CE:1.5
GR:0
GR:1
GR:2
DOE:2
DOE:4
k1:M
k2:M
k1:S
k3:A
k3:P
-2000000
TF:0
TF:1
FT:0
FT:7
Time:30
EC:0
EC:1
OF:0
OF:1
CH:0
CH:1
CE:0
CE:3
Time:1
-4000000
Exp
-6000000
-8000000
-10000000
Factors
Figure 15 Main Effects Plot of Regression Model.
88
Summary statistics from the model are presented in Table 15. The model
produced an R2 value (observed vs. predicted) of 0.934 and an SSE (Sum of Squared
Errors) of 0.883.
Summary Statistics
Criterion Value
R^2 0.934
R^2 adj 0.924
R^2 predict 0
R^1 0.743
PRESS 4.54E+15
s (est. err.) 171190.53
SSE(LSE)/SSE(LAD) 0.833
Table 15 Summary statistics for model.
Coefficient Estimates
Standard p-
Coefficients Error t Stat value Lower 95% Upper 95% VIF
const 79930 62285 1.28 0.208 -46790 206652
Exp Savings 0.85 0.80 1.06 0.297 -0.78 2.47 101.65
FT -76708 19028 -4.03 0.000 -115422 -37994 2.04
EA 1258 56503 0.02 0.982 -113699 116216 2.22
Exp Savings*FT 2.20 0.17 13.09 0.000 1.86 2.54 222.24
Exp Savings*EA -4.04 0.22 -18.37 0.000 -4.48 -3.59 123.45
Table 16 Coefficient estimates
89
Profit
20000000
15000000
10000000
15000000-20000000
5000000
10000000-15000000
Profit 0 5000000-10000000
-5000000 0-5000000
-10000000 -5000000-0
-15000000 6.2 -10000000--5000000
-20000000 3.1 FT -15000000--10000000
1600.0
-20000000--15000000
490133.3
978666.7
0.0
1467200.0
1955733.3
Exp Savings
Figure 16 3-D plot of the results.
90
Normal Scores vs. Residuals
2.5
2
1.5
1
Normal Scores
0.5
0
-800000 -600000 -400000 -200000-0.5 0 200000 400000
-1
-1.5
-2
-2.5
Residuals
Figure 17 Normal sores vs. residuals.
This model is intuitive and appears to provide a good fit. With higher expected
savings it seems logical to apply more formal methods and obtain higher profit as the
model predicts. With any regression questions should be asked as to the goodness of
fit and evaluated against a checklist such as Table 17.
Issue Evaluation Method Checks

Inputs supports model Variance Inflation Factors (VIFs) 9
Outputs fit the model Outliers 9
Outputs fit the model Residual Plots 9
Table 17 Regression Checklist.
91
An examination of the variance inflation factors (VIF) associated with this
model would indicate that they are unacceptable and this model should not be used for
predictions. Variance inflation factor measures the impact of collinearity among the
inputs in a regression model on the precision of estimation. It expresses the degree to
which collinearity among the predictors degrades the precision of an estimate. In a
regression model we expect a high variance explained (r-square). The higher the
variance explained is, the better the model is. However, if collinearity exists, probably
the variance, standard error, parameter estimates are all inflated. The high variance
might not be a result of good independent predictors, but a mis-specified model that
carries mutually dependent and thus redundant predictors. Scaling the inputs does not
always cause improvements in the VIFs, as in this example problem. Some model
forms may not be suitable to fit the given data.
Another simple model involving training and management direction was
investigated. This model is represented by equation 5.10.
Profit = 13510+38856*M_I+19566*Training (5.10)
Where:
M-I is whether the project was initiated by management or by team members,
Training is before of after Formal Six Sigma Training
And Profit is the response.
92
The main effects plot for this model is shown in Figure 18, with the results of
the regression model presented in Table 19, and the summary statistics given in Table
18.
60000
Predicted Response
40000
20000
0
k1:M k1:I Training:-1 Training:0 Training:1
Figure 18 Main Effects Plot of Simple Regression Model.
93
Summary Statistics
Criterion Value
R^2 0.194
R^2 adj 0.148
R^2 predict 0.0251
R^1 0.102
PRESS 1.33E+11
s (est. err.) 56133.32
SSE(LSE)/SSE(LAD) 0.827
Table 18 Summary statistics for Simple model.
Coefficient Estimates
Standard Lower Upper
Coefficients Error t Stat p-value 95% 95% VIF
-
const 13509.53 12750.01 1.06 0.297 12374.36 39393.43
M/I_I 38856.45 21245.49 1.83 0.076 -4274.20 81987.09 1.06
Training 19566.38 11002.61 1.78 0.084 -2770.10 41902.86 1.06
Table 19 Coefficients for Simple model.
5.9 Regression Results

This model was based on the results of 39 separate projects at one company
while most of the current literature is based on only a single case study. Specifically,
our results indicate the importance of estimating expected savings with an economic
study prior to the project start. With this information management can more
94
effectively decide how formal to be in the approach to the proposed project. From our
data the brake even point is $12,500. If a project is projected to save less than $12,500,
formal Six Sigma methods are not advised. Instead relying on quick engineering input
would tend to generate more profit sooner.
On projects over $12,500, the model states formal quality improvement tools
should be employed. As anticipated savings increase it is more advisable to spend
increasing time with statistical tool to increase the probability of increased profits.
The ability to estimate potential effects of changes on the profitability of
projects is valuable information for policymakers in the decision-making process. This
study demonstrated that utilizing existing data analysis tools to this new management
data source provides useful knowledge that could be applied to help guide in project
management. In this study we compared results of various sized projects and the use
of formal tools. In our case study we found determining the estimate of the economical
value to be important to guide the degree of use of formal tools. Based on the results
of this study, when predicted impact is small a rapid implementation based on
engineering input is best. As projects expand more statistical data improves outcome.
This study had a crossover point of $12,500.
The simple model also tends to show a strong benefit to training. This model
has good VIF values and supports the findings from the SPC findings. Of interest is
the correlation on management initiation of projects. There is still ambiguity in the
results. For example it is not know if people worked harder on projects they initiated
or if due to better familiarity they picked better projects.
95
However, replication of these results for other projects at other companies and
industries is warranted to determine the generalizability of these findings. Natural
extensions of this study would be to expand the scope of the analysis to include other
variables such as multiple manufacturing sites and other statistical methods.
5.10 Markov Decision Processes

Recently, there has been considerable interest in modeling the utility of data
and analysis method, including Six Sigma related methods. For example, Bisgaard and
Freiesleben (2000) used simple financial models to investigate the return on
investment for any process improvement method for a particular context. Yu and
Popplewell (1994) proposed a neural network meta-model as an online decision-
support model for Six Sigma quality improvement efforts, taking into consideration
yield and cost. Chan and Spedding (2001) expanded on this work. Eid, Moghrabi and
Eldin used simulation to compare quality/cost decision scenarios. Goh (2002)
discussed some strategic perspectives of Six Sigma based on organizational security.
A hole in this growing literature is the optimality of selection and ordering
component methods associated with Six Sigma improvement projects. A proposed
method to investigate optimal selection is the use of Markov Decision Processes
(MDP, Puterman, 1994) and Partially Observed Markov Decision Processes (POMDP,
Lovejoy, 1991, and White, 1991).
MDPs are used to model systems in which the evolution is controlled by
available actions which can be chosen by the system in each state. The use of MDP
96
models has encompassed a wide range of applications. They have been widely applied
to inventory control problems which represent one of the earliest areas of application.
MDP models have been used to study equipment maintenance and replacement
problems, (Rust, 1987) and (Golabi, 1993) as well as computer, manufacturing and
communications systems. MDP models have become popular in behavioral ecology
used in a wide range of contests to gain insight into factors influencing animal
behavior. Examples include models of social and hunting behavior of lions (Clark,
1987: Mangel and Clark, 1988), site selection and number of eggs laid by apple
maggots and medflys (Mangel, 1987), daily vertical migration of sockeye salmon and
plankton (Levy and Clark, 1988: Mangel and Clark, 1988), changed in mobility of
spiders in different habitats (Gallespie and Caraco, 1987), and singing versus garaging
tradeoffs in birds (Huston and McNamara, 1986) as well as Games of Chance (Dubins
and Savage, 1965). Kelly and Kennedy (1993) used MPD models in their study of
mate desertion in Copper’s hawks.
Details about Markov decision processes can be found in Puterman (1994).
The Markov decision processes describe a model for sequential decision making under
uncertainty which takes into account both the outcomes of current decisions and future
decision making opportunities. At each decision epoch, the system state provides the
decision maker with all necessary information for choosing an action from the set of
available actions in that state. As a result of choosing an action in a state, two things
happen; the decision maker receives a reward (or cost), and the system evolves to a
possibly different state at the next decision epoch. Both the rewards and transition
97
probabilities depend on the state and the choice of action. As this process evolves
through time, the decision maker receives a sequence of rewards.
At each decision epoch, the decision maker chooses an action in the state
occupied by the system at that time. A policy provides the decision maker with a
prescription for choosing this action in any possible future state. A decision rule
specifies the action to be chosen at a particular time. It may depend on the present
state alone or together with all previous states and actions. A policy is a sequence of
decision rules. Implementing a policy generates a sequence of rewards. The sequential
decision problem is to choose prior to the first decision epoch a policy to maximize a
function of the reward sequence.
The key ingredients of this sequential decision model are the following:
1. A set of decision epochs.
2. A set of system states.
3. A set of available actions.
4. A set of state and action dependent immediate rewards or costs.
5. A set of state and action dependent transition probabilities.
When confronted with a decision, there are a number of different alternatives
(actions) available. Choosing the best action requires a consideration of more than just
the immediate effects of the action. The immediate effects are often easier to calculate
than the long-term effects. Sometimes actions with poor immediate effects can have
better long-term ramifications. To maximize the total expected value function, the
98
right tradeoffs between the immediate rewards and the future gains are needed to yield
the best possible solution.
When making a decision, we need to consider how actions will affect the
system. The state is the way the system currently exists and an action will have the
effect of changing the state. The actions are the set of possible alternatives that can be
chosen. The problem is to know which of these actions to take when in a particular
state of the system. When deciding between different actions, we need to consider how
they will affect the current state. The transitions specify how each of the actions
changes the state. Since an action could have different effects, depending upon the
state, we need to specify the action’s effect for each state. The most powerful aspect of
the Markov decision process is that the effects of an action can be probabilistic. We
could specify the effects of doing action ‘a1’ in state ‘s1’ if there is no question about
how ‘a1’ changes the system. However, many decision processes have actions that are
not this simple. Sometimes an action usually results in state ‘s2’ but occasionally it
might result in state ‘s3’. MDPs allow specifying a set of resulting states and the
probability that each state results.
As an agent moves through its environment, it bases its actions on information
received from a number of sources including sensory input and memories of previous
inputs. This information tells the learner something about the state of the world.
An agent in some state at time t executes an action and receives a reward from
the environment. This is called a decision process. If the next state is dependent only
99
on the current state and action the decision process is said to obey the Markov
property. This is called a Markov decision process (MDP) (Bellman 1957b).
If the set of states and actions are finite, then the problem is called a finite
MDP. We can also distinguish between finite and indefinite horizon problems, where
the task has a natural endpoint, and infinite horizon problems, where the task
continues forever.
Formally an MDP consists of:
o A set of state S, and actions A,
o A transition distribution P ( s t +1 s t , a t ), s t , s t +1 ∈ S , a t ∈ A , and
o A reward distribution P (r t s t , a t ), s t ∈ ℜ, a t ∈ A.
In the above, t indexes the time step, which ranges over a discrete set of points
in time. The transition probability is denoted by Pij (a), Pr( s t +1 = j s t = i, a t = a). The
expected immediate reward received by executing action a in state i by ri (a ) :
{
ri (a) = E r t s t = i, a t = a } (5.11)
The solution to an MDP is called a policy and it simply specifies the best
action to take for each of the states. The goal of solving an MDP is to find a policy
that maximizes the total expected reward received over the course of the task. A policy
tells the learning agent what action to take for each possible state. It is a (possibly
stochastic) mapping π : S → A from states to actions. The policy can be non-
stationary, in which case a different mapping from states to actions can be used at each
100
point in time. Alternatively, it can be stationary, in which case the same mapping is
used at every point in time.
The expected return for a policy π is defined as the total reward that is
expected when following policy π :
⎧n ⎫
Eπ {R t } = Eπ {r t + γ r t +1 + ... + γ n r t + n } = Eπ ⎨∑ γ k r t + k ⎬ (5.12)
⎩ k =0 ⎭
where t is current time. Notice that the expectation is taken with respect to the
policy π .
By assuming that the problem is Markov, we know that an optimal policy need
only be a function of the current state: no other information is required to act optimally
(Howard 1960).
The class of MDPs is a restricted but important class of problems. By
assuming that a problem is Markov, we can ignore the history of the process, and
thereby prevent an exponential increase in the size of the domain of the policy. The
Markov assumption underlies a large proportion of control theory, machine learning
and signal processing.
Dynamic programming is a technique for finding solutions to
optimization problems. It is similar to divide and conquer techniques, where a problem
is broken down into sub-problems that can be solved. To be amenable to dynamic
programming, the optimization problem has to exhibit optimal substructure: An
optimal solution must contain within it optimal solutions to sub-problems.
101
In the case of MDPs, we must find a policy that produces the greatest expected
return. With knowledge of transition probabilities Pij (a) and expected immediate
rewards ri (a), and given a stochastic policy π , we can calculate the expected
discounted return from the current state s:
⎧n ⎫
V π ( s ) = Eπ ⎨∑ γ k r t + k s t = s ⎬ (5.13)
⎩ k =0 ⎭
Here t denotes the current time. The function V π is called the value function
for policy π . The value function gives the expected return that can be achieved by
starting from any state, and then following policy π .
Markov Decision Process approaches involve modeling a sequence of
selections in which (1) the outcomes are random with probabilities that depend upon
the current state (st) at time t, and action taken, a, and (2) the rewards, rt, depend on
the current decision and results form a sequence of future decisions. With a finite
number of decision periods, e.g., days in a project, the optimal policy can be derived
using a recursive approach called dynamic stochastic programming.
To apply MDP to support decision-making, one needs to know the current
system state, st. Depending upon how the states are defined; a team in a process
improvement project might not have complete knowledge of the current state of their
project. Partially Observed Markov Decision Processes (POMDP) permit the user the
flexibility to supply only probabilities that the system is in specific states. Intuitively,
102
the fact that project improvers have only partial knowledge of the system state might
provide important justification for a “one-size-fits-all” approach such as Six Sigma
type methods.
Yet, while MDP models permit the derivation of globally optimal decision
policies, POMDP methods do not generally lead to guarantees that the derived policies
are optimal. Lovejoy (1991) and White (1991) both survey solution methods for these
problems including work in Smallwood and Sondik (1973) and White and Scherer
(1989). Yacout and Gautreau (2000) used POMDP models to compare three quality
assurance policies: do nothing, inspect, or improve. In Gautreau, Yacout, and Hall
(1997), they had already used POMDP directly as a tool for quality improvement. In
our study we only considered unconstrained, discrete time, finite horizon MDPs.
5.11 Markov Decision Process Analysis

Suppose that a manufacturer is currently losing $500K/year running a certain
process but knows a competitor is making $150K/year from running a similar process.
Instead of shutting the process down, the manufacturer decides to invest in an
improvement project with a three month time limit. Assume that the resident quality
expert, a “green belt” six sigma practitioner, has already declared that the process state
(st) must be one of those listed in the left-hand column of Table 20. Also, assume that
the possible actions (a) are listed in the top first row of the table.
103
These actions include applying statistical process control (SPC) p charting, a
relatively inexpensive ($) design of experiments (DOE) application (e.g., a standard
screening method using fractional factorials), and a relatively expensive ($) DOE
method (e.g., a response surface method application). The quality expert estimates the
annualized cost of applying each action for processes in the various states, including
the cost of measurement equipment, record keeping, and training. If either a control
plan is applied or the production action is taken, then the system improvement ends
(all future states are #6).
action (a)
SPC p Control
State (st) Meet Charting $ DOE $$ DOE Optimize Plan Produce
1. Settings lose
money and
-20 -50 -70 -70 -10 -150 -500
capability
unknown
2. Settings lose
money and -20 -30 -40 -70 -10 -10 -200
capability known
3. Measured but
input-output func. -20 -30 -40 -70 -10 -50 -100
not known
4. Input-outputs
func. known but -20 -30 -40 -70 -10 -30 80
not fully opt.
5. Improved
settings found but -20 -30 -40 -70 -10 250 150
not validated
6. System
-20 -30 -40 -70 -10 0 50
improvement ends
Table 20 Costs and rewards (in $K), rt(st,a), of applying actions (a) in different states
(st).Coefficient estimates
104
Assume further that the quality expert has documented that each of the m = 6
actions will cause systems in each of the q = 6 states to transition to each state (st+1 =
1,…,6). Table 21 shows the probabilities for an example action, which is applying an
expensive design of experiments method (a = $$ DOE) such as response surface
methods.
To State (st= i)
1 2 3 4 5 6
1 0.9 0.1 0.0 0.0 0.0 0.0
From 2 0.0 0.2 0.8 0.0 0.0 0.0
State 3 0.0 0.00001 0.1 0.79999 0.1 0.0
(st+1= j) 4 0.0 0.0 0.0 0.5 0.5 0.0
5 0.0 0.0 0.0 0.01 0.99 0.0
6 0.0 0.0 0.0 0.0 0.0 1.0
Table 21 Assumed transition probabilities for applying DOE, pt(st+1= j|st= i,a = $$
DOE).
With this information, an algorithm is generated to find the actions that
globally maximize the expected reward at every decision period for every possible
state. This calculation is based on the following recursion for t = 5,4,…,1 and i =
1,…,q:
ERt*(st = i) = max {rt(st,a) + Σj=1,…,q pt(st = j|st = i,a) ERt+1*(st+1= j)} (5.14)
a
where ERt=6*(s6 = i) = r6(s6 = i,a = Produce) for i = 1,…,q. The results are
shown in Table 22. The quality expert inputs the time period and the state and the
table indicates which method is recommended. A second table also gives the expected
105
reward or return on investment shown in Table 23. Note that if the current state
cannot be observed and one has only the probability of being in a certain state,
solution to global optimality is, in general, computationally difficult (e.g., see
Lovejoy, 1991a).
StateWeeks 1-2 Weeks 3-4 Weeks 5-6 Weeks 7-8 Weeks 9-10 Weeks 11-12
1 Meet Meet Meet Meet Meet Produce
2 SPC SPC SPC SPC Produce Produce
3 $ DOE $ DOE $$ DOE $ DOE $$ DOE Produce
4 Optimize Optimize Optimize Optimize Produce Produce
5 Control Plan Control Plan Control Plan Control Plan Control Plan Produce
6 Produce Produce Produce Produce Produce Produce x
Table 22 Optimal decision policy for five decision periods.
StateWeeks 1-2 Weeks 3-4 Weeks 5-6 Weeks 7-8 Weeks 9-10 Weeks 11-12
1 65.0 -27.6 -101.5 -179.6 -272.2 -500.0
2 175.0 96.8 -5.3 -75.7 -150.0 -200.0
3 307.8 251.3 183.4 67.7 -1.0 -100.0
4 429.4 379.1 327.2 264.5 130.0 80.0
5 500.0 450.0 400.0 350 300.0 150.0
6 500.0 250.0 200.0 150.0 100.0 50.0 x
Table 23 The expected reward, ERt*(st = i), in $K as a function of period and state.
The above illustrates the application of Markov decision process (MPD) to
assist in selecting which methods to use in which order in an improvement project.
The results are subjective in the sense that the transition probabilities associated with
each action, pt(st = j| st = i,a), e.g., the numbers in Table 18 were subjectively
estimated. Therefore, the resulting decision policy may be viewed as a rationalization
for the choice of methods with explicit assumptions. Another possible benefit of using
106
the above method for planning improvement projects is the possibility that the
probabilities can capture knowledge of experts and make this knowledge available to
novices or “green belts”.
A more detailed model might be useful as it relates to the data used in the
regression model and based on the component methods definition of Six Sigma.
Component methods can be associated with a phase of Six Sigma as shown in Table
24.
107
Phase Deliverables Component methods
DEFINE- (Finding the • Establish process Surveys
Ys in the model) responsibilities Focus Groups
• Define system Interviews
(inputs, outputs, Process MAP
customers) Pareto Analysis
• Identify customer
requirements
MEASURE – (Finding • Gap between Process mapping
the Ys in the model) customer Pareto analysis
requirements and Process capability
process capability Gage R&R
Measurement system
analysis
ANALYZE (find the • Cause and effect Cause and effect
factors or x’s in the matrix analysis
model) • Prioritized list of all Process mapping
x’s Benchmarking
• Few vital x’s Histograms
• Statistical analysis Cater diagrams
for significance Run charts
Multivariate charts
Box and whisker
Pareto chart
Regression
IMPROVE - (Define • Target setting for x’s Process mapping
Y=(x,z) and move Simulation
toward Max Y) DOE
CONTROL – Maintain • Control plan Risk assessment
improvements to the SPC
system Process capability
Checklists
Documentation
Table 24 Component Method of Six Sigma.
108
A Six Sigma type method is an arrangement of component methods such that
there is at least one selection of the associated activities that is consistent with a
DMAIC ordering of activities. Also, in Six Sigma type methods multiple components
methods can be used in the contest of the same activity and no method needs to be
performed associated with any of the activities.
The Figure 19 below shows an example of Six Sigma type method that the
team might tentatively lay out in the design phase of a project.
109
Define:
Hold formal meeting
Measure:
Perform gage R&R
Create SPC charts
Perform
benchmarking
Analyze:
Plan and perform DOE
Fit regression model
Apply FMEA
Improve:
Apply formal optimization
Control:
Develop control plan
Continue SPC charting
Figure 19 Flow Chart of DMAIC Six Sigma Project.
110
The optimality of any particular Six Sigma type method in question or Six
Sigma type method for a particular application can be investigated using Markov
decision processes. Possible states could be defined as:
“Right” initial team

Found “right” project
Charter is established
Initial baseline established
Confirmed gauge capability
“Right” initial factors
Confirmed important factors
“Accurate” Yr(x) models
Developed “recommendations”
Establish new baseline
Updated SOPs
Project completed
We have the limited data form the previously covered study of 39
improvement projects and the actions as defined in Table 21. The rewards or costs of
these actions are easy to estimate based on the data. The transition probabilities are not
as easy to estimate based on this limited data. One method for this estimate is a
Bayesian approach based on Dirichlet prior distributions.
Bayesian probabilities are sometimes called subjective probabilities. It is
important to understand exactly what is meant by subjective in this contest. Decision
analyses are often unique. The situation in which one is making the decision may
occur only once. It cannot be replicated, so there is no possibility for measuring
probabilities by repeated sampling. People in the same state might take different action
dependent on skill level, taste or problem particulars. Nevertheless, Bayesian analysis
may be used to compute the probabilities needed to make decision. Because these
111
probabilities cannot be measured by repeated sampling, they are called subjective and
they represent a degree of belief in a particular outcome.
Bayesian analysis is a kind of meta-analysis in which observed data in
combined with a prior belief to end up with a posterior belief. In short, it’s a way to
update a belief with new data. A positive aspect of the Bayesian approach is that it
encapsulates the manner in which research findings are assimilated. Meta-analysis is
literally an analysis of analyses, a synthesis of all research on a particular effect.
If there is no basis in observed data for estimating the prior probability
distribution, then the analyst may simply assume a particular prior distribution. Most
commonly, a noninformative prior distribution is assumed. A classic approach to
multinomial estimation is via the use of the Dirichlet distribution. The Dirichlet
distribution entertains several properties that become very useful in statistical
inference. According to Friedman and Signer (1999), estimates derived using Dirichlet
priors are consistent, the estimate converges with probability one to the true
distribution; conjugate, the posterior distribution is also a Dirichlet distribution; and
can be computed efficiently, all queries of interest have a closed-form solution.
In Bayesian parameter estimation, the prior incorporates prior knowledge or
beliefs about the parameters. As data is gathered, these beliefs do not play a significant
role anymore. More specifically, if the prior is well-behaved, does not assign 0
probability to feasible parameter values, the approach will converge in the limit to the
real values. A Dirichlet prior with parameters β1, … βk is defined as:
P(θ) = α ∏ θiβi-1 (5.15)
112
Then the posterior will have the same form, with parameter βi + Ni:
P(θ|D) = P(θ) P(D|θ) = α ∏ θiβi-1+Ni (5.16)
The property that the posterior distribution follows the same parametric form
as the prior is called conjugacy. The Dirichlet prior is a conjugate family for the
multinomial likelihood. Conjugate families are useful because they can be represented
in closed form, incremental updates to the parameters can be done as data is gathered,
and often there is a closed-form solution for the prediction problem.
With this background a MDP model for Six Sigma based on the previous
collected data and action definitions could take the form of Figure 20.
113
D ev elo pe d "r ec om m e nd atio ns "
C on fir m e d inp or ta nt fac to rs
C on fir m e d ga ug e c a pa bility
Initia l ba s e line es ta blis h ed
"A c c u ra te " Yr ( x ) m o de ls
E s ta blis h n ew ba s e line
C ha r ter is es ta blis h ed
F o un d "r igh t" p ro jec t
"R ig ht" in itial fac to rs

"R ig ht" in itial tea m
U pd ated S O P s
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
0 0 0 0 0 0 0 0 0 0 0 TF TF EC
1 0 0 0 0 0 0 0 0 0 0 EC EC EC TF
1 1 0 0 0 0 0 0 0 0 0 TF TF TF TF SP
1 1 1 0 0 0 0 0 0 0 0 SP SP SP SP SP
1 1 1 .5 0 0 0 0 0 0 0 SP SP SP SP GR
1 1 1 1 0 0 0 0 0 0 0 GR GR GR GR CE
1 1 1 1 1 0 0 0 0 0 0 CE CE CE CE CE
1 1 1 1 1 .5 0 0 0 0 0 CE CE CE CE SD
1 1 1 1 1 1 0 0 0 0 0 SD SD SD SD SD
1 1 1 1 1 1 .2 0 0 0 0 SD SD SD SD SD
1 1 1 1 1 1 .4 0 0 0 0 SD SD SD SD SD
1 1 1 1 1 1 .6 0 0 0 0 SD SD SD SD SD
1 1 1 1 1 1 .8 0 0 0 0 SD SD SD SD RD
1 1 1 1 1 1 1 0 0 0 0 RD RD RD RD RD
1 1 1 1 1 1 1 .3 0 0 0 RD RD RD RD RD
1 1 1 1 1 1 1 .5 0 0 0 RD RD RD RD RD
1 1 1 1 1 1 1 .8 0 0 0 RD RD RD RD OP
1 1 1 1 1 1 1 1 0 0 0 OP OP OP OP SP
1 1 1 1 1 1 1 1 1 0 0 SP SP SP SP SP
1 1 1 1 1 1 1 1 1 0 0 SP SP SP SP CP
1 1 1 1 1 1 1 1 1 .5 0 CP CP CP CP TE
1 1 1 1 1 1 1 1 1 1 .5 CP CP CP TE
1 1 1 1 1 1 1 1 1 1 1 TE TE TE
Figure 20 Morkov Decision Process for Six Sigma.
The actual actions taken might be along the diagonal as indicated. The amount
of data obtained to date has not allowed for the construction of an adequate model and
continues to be an area of research interest for future work.
114
5.12 Sample MDP Example
The data associated with the 30 improvement projects can be used in the
derivation of state/action rewards and the state/action transition probabilities. In this
simplified case six states are defined which map to the standard DMAIC as follows in
Table 25.
State Description Map to DMAIC

1 No Charter Define
2 No Base line Measure
3 Define input-output Analysis
4 No Recommendation Improve
5 Not Controlled Control
6 Production
Table 25 State Description Mapped to DMAIC for Sample MDP.
Ten actions were possible with average cost calculated form the collected data.
These action-rewards are presented in Table 26 with the tabulation of the number of
state transitions by action for the projects displayed in table 27.
Action Cost of action Action Cost of action

Charter $1440 DOE $16000
Team Forming $1440 SPC $750
Process Mapping $2080 Documentation $1500
Cause & Effect $1580 One Factor Experiment $4000
GR&R $1700 Engineering Analysis $4000
Table 26 Actions and Rewards for Sample MDP.
115
Action State to State 1 2 3 4 5 6
Charter 1 12 8 6 2 - -
Charter 2 - - - - - -
Charter 3 - - - - - -
Charter 4 - - - - - -
Charter 5 - - - - - -
Charter 6 - - - - - -
Team Forming 1 1 11
Team Forming 2 - - - - - -
Process Mapping 1 - - 3 3 - -
Process Mapping 2 - 3 5 2 5 -
Process Mapping 3 - - - - - -
Cause & Effect 1 - - - - - -

Cause & Effect 2 - - - - - -
Cause & Effect 3 - - 5 6 1 -
Cause & Effect 4 - - - - - -
Cause & Effect 5 - - - - - -
Cause & Effect 6 - - - - - -
GR&R 1 - - 1 - - -
GR&R 2 - 1 5 - - -
GR&R 3 - - - - - -
GR&R 4 - - - - - -
GR&R 5 - - - - - -
GR&R 6 - - - - - -
Tabulation of State Transitions for Sample MDP (continued on next page)
116
Tabulation of State Transitions (continued from previous page)

DOE 1 - - - - - -
DOE 2 - - - - - -
DOE 3 - - - - - -
DOE 4 - - - - 9 2
DOE 5 - - - - - -
DOE 6 - - - - - -
SPC 1 - - - 1 - -
SPC 2 - - 3 - - -
SPC 3 - - - - - -
SPC 4 - - - - - -
SPC 5 - - - - - -
SPC 6 - - - - - -
- - - - - -
Documentation 1 - - - - - -
Documentation 5 - - - - 3 26
One Factor Experiment 1 - - - - - -

One Factor Experiment 4 - - - 4 6 -
Engineering Analysis 1 - - - 1 - 3
Engineering Analysis 2 - - - - - -
Engineering Analysis 3 - - - 15 1 -
Engineering Analysis 4 - - - 2 10 4
Table 27 Tabulation of State Transitions for Sample MDP.
117
An examination of Table 27 illustrates the limited data available from the
improvement projects and the need for Bayesian estimation. Table 28 gives an
example of the tabulated actual data and Alpha priors which when added together
results in a usable transition matrix for the action Engineering Analysis. Continuing
with the remaining actions, (estimating Alpha priors and adding to the tabulated data)
results in a matrix that can be used to solve for a MDP policy. One possible Bayesian
distribution based on the data from the 39 projects and Alpha priors as described
above is presented in Table 29.
Tabulation State to State 1 2 3 4 5 6

No Charter 1 - - - 1 - 3
No Base line 2 - - - - - -
Define input-output 3 - - - 15 1 -
No Recommendation 4 - - - 2 10 4
Not Controlled 5 - - - - - -
Production 6 - - - - - -
Alpha Priors
No Charter 1 67 3 5 9 5 7
No Base line 2 7 80 6 4 2 1
Define input-output 3 2 3 50 19 6 4
No Recommendation 4 2 2 4 10 50 16
Not Controlled 5 2 5 20 30 40 3
Production 6 0 0 0 0 0 10
Table 28 Bayesian Estimate of State Transitions for Sample MDP.
118
Charter 1 .01 .61 .189 .071 .07 .05
Charter 2 .3 .5 .1 .05 .03 .02
Charter 3 .2 .3 .4 .05 .03 .02
Charter 4 .02 .2 .3 .4 .05 .03
Charter 5 .02 .02 .2 .3 .41 .05
Charter 6 0 0 0 0 0 1
Team Forming 1 .3 .4 .1 .08 .07 .05

Team Forming 2 .3 .5 .1 .05 .03 .02
Team Forming 3 .2 .3 .4 .05 .03 .02
Team Forming 4 .02 .2 .3 .4 .05 .03
Team Forming 5 .02 .03 .2 .3 .4 .05
Team Forming 6 0 0 0 0 0 1
Process Mapping 1 .6 .2 .2 .1 0 0
Process Mapping 2 .07 .16 .3 .13 .3 .04
Process Mapping 3 .1 .1 .5 .2 .05 .05
Process Mapping 4 .05 .1 .1 .5 .2 .05
Process Mapping 5 .05 .05 .1 .1 .5 .2
Process Mapping 6 0 0 0 0 0 1
Cause & Effect 1 .5 .2 .1 .1 .05 .05

Cause & Effect 2 .1 .5 .2 .1 .05 .05
Cause & Effect 3 .05 .07 .35 .4 .08 .05
Cause & Effect 4 .05 .1 .1 .5 .2 .05
Cause & Effect 5 .05 .05 .1 .1 .5 .2
Cause & Effect 6 0 0 0 0 0 1
GR&R 1 .07 .1 .61 .1 .07 .05

GR&R 2 .03 .12 .75 .05 .03 .02
GR&R 3 .1 .1 .5 .2 .05 .05
GR&R 4 .05 .1 .1 .5 .2 .05
GR&R 5 .1 .05 .05 .1 .5 .2
GR&R 6 0 0 0 0 0 1
State Transitions for Sample MDP (continued on next page)
119
State Transitions (continued from previous page)

DOE 1 .4 .3 .17 .05 .05 .03
DOE 2 .02 .4 .3 .2 .05 .03
DOE 3 .02 .05 .1 .75 .05 .03
DOE 4 .02 .02 .04 .05 .73 .14
DOE 5 .02 .05 .2 .3 .4 .03
DOE 6 0 0 0 0 0 1
SPC 1 .64 .01 .07 .2 .07 .01

SPC 2 .01 .07 .8 .07 .04 .01
SPC 3 .01 .07 .8 .07 .04 .01
SPC 4 .01 .02 .07 .8 .07 .04
SPC 5 .01 .01 .04 .07 .07 .8
SPC 6 0 0 0 0 0 1
Documentation 1 .8 .07 .06 .04 .02 .01

Documentation 2 .07 .8 .06 .04 .02 .01
Documentation 3 .02 .07 .8 .06 .04 .01
Documentation 4 .01 .02 .06 .8 .07 .04
Documentation 5 .01 .01 .03 .05 .1 .8
Documentation 6 0 0 0 0 0 1
One Factor Experiment 1 .8 .07 .06 .04 .02 .01

One Factor Experiment 6 0 0 0 0 0 1
Engineering Analysis 1 .67 .03 .05 .1 .05 .1

Engineering Analysis 6 0 0 0 0 0 1
Table 29 State Transitions for Sample MDP.
120
The above Bayesian probabilities and the rewards given in Table 26 where
used in the development of an optimal MDP policy for the six listed states. It was also
assumed that the production state had two different rewards. This was based on real
data gathered from the improvement project. The average reward for a project that
terminated early was $4,700. The average reward for a project that entered production
at epoch 9 was $180,875. The optimal policy calculated with these assumptions is
presented in Table 30.
State Period 1 Period 2 Period 3 Period 4 Period 5

No Charter GR&R GR&R GR&R GR&R GR&R
No Base P. Map P. Map P. Map P. Map P. Map
Input-output DOE DOE DOE DOE DOE
Recomm Eng. Anal. Eng. Anal. Eng. Anal. Eng. Anal. Eng. Anal.
Controls Docs. Docs. Docs. Docs. Docs.
Ready Production Production Production Production Production
Period 6 Period 7 Period 8 Period 9

No Charter Eng. Anal. Eng. Anal. Eng. Anal. Production
No Base P. Map P. Map C & E Mat. Production
Input-output DOE C & E Mat. C & E Mat. Production
Recomm Eng. Anal. Eng. Anal. Eng. Anal. Production
Controls Docs. Docs. SPC Production
Ready Production Production Production Production
Table 30 Optimal Policy for Sample MDP.
This optimal policy has the structure on the standard DMAIC but does not
follow the five phases explicitly. This could arise for numerous reasons. The Alphas
selected may not be reasonable. The validity of the selection would become more
apparent with added real data which would allow an update to the Bayesian estimates.
121
DMAIC may not be optimal in all cases. Again more data is needed to investigate this
possibility. Both Linderman et. al. (2003) and Harry and Crawford (2005) have
indicated that DMAIC may not be optimal for small project, the “low hanging fruit”
and that DMAIC should be focused on complex challenging problems. For simple
tasks such a system may decrease performance. The 39 improvement projects included
in this study included a wide array of programs with estimated savings ranging from a
low of $1,600 to a high of $2.2 million. It is possible that the data should be split into
simple and complex projects. This is an area for future work, relating the size of a
project to its optimal structure.
5.13 Conclusions
Much of the literature related to six sigma focuses on knowledge that
participants at various levels “should know” (Hoerl 2001). Other literature has
addressed organization design to accommodate a “Six Sigma Culture” (Sanders and
Hild 2000, and Hoerl 2001). Goh (2002) has discussed some strategic perspectives of
Six Sigma based on organization needs. The literature on supporting method selection
section in the context of Six Sigma has focused on decisions at a high level such as
whether to perform projects at all including Bisgaard and Freiesleben (2000), Chan
and Spedding (2001), Gautreau, Yacout and Hall (1997), Yacout and Gautreau (2000),
and Yu and Popplewell (1994). In this study some methods for the evaluation of the
expanding Six Sigma database are investigated. These approaches help to express Six
122
Sigma in more quantitative terms, which has more often been expressed in qualitative
terms.
Clearly the list of methods considered here for meso-analysis is incomplete and
each has some advantages and limitations as shown in Table 31.
123
Tool Strength Limitation
Regression Simple and many diagnostics Necessary sample size can become
available large as
n = 4 × [1 + m pre + m act + (m pre × m act )]
MPDs Can address sequencing effects There is strong pressure to define
of method applications and only a small number of states for
interactions between them. simplicity.
They permit the theoretical Because of this pressure assumptions

development of a prescriptive, about the system performance
data driven expert system that depending only on the state are
could advise novices about the critical.
methods relevant to their
situations. Further, the data must be coded into
transitions between these states.
Finally, because of the need to

estimate many transition
probabilities, Bayesian priors are
generally needed for the transition
probabilities.
SPC Can provide simple way to Generally, databases are limited.
evaluate whether changes in
business practices are adversely There is intuitive pressure to chart
or positively affecting quality individual observations because
programs every project is important in some
sense.
There may be a need for the

development of new short run charts
such as EWMA charts.
Table 31 Comparison of strengths and limitations for meso-analysis methods used.
124
To our knowledge no research has focused on the use of this database in this
way. Practical implementation of these methodologies in a systematic manner will
help industry to identify, analyze and evaluate factors and their interdependency,
which would help to understand and unveil the complexity of Six Sigma. This could
open the door to research that more clearly clarifies and capitalizes on the primary
value of the Six Sigma movement which lies in providing a relatively detailed
approach to problem solving (e.g., see Watson, 2000). In addition to providing
algorithms that can be used to generate software for green belts, the research may lead
to substantially more valuable methods than those of the Six Sigma type.
125
CHAPTER 6
CONCLUSIONS AND FUTURE RESEARCH
A summary and conclusion of this study is presented in this chapter. The first
section provides an overview of the research while the second section summarizes the
major research findings. The third section addresses the limitations of the study and
potential areas for future research. Finally, the contributions of this study are presented
in the fourth section.
6.1 Overview
The main objective of this dissertation was to review the published literature
on Six Sigma and identify opportunities for additional contributions for the academy.
The major problems addressed in this dissertation are:
1. With reference to past academic contributions, what is Six Sigma? Also, what
are the on-going research trends?
2. What are the implications of Six Sigma philosophy and methods for university
education?
126
3. What methods should be used to mine the new databases about project
financial results? Also, what insights can be gained form studying data at a
real company?
This dissertation addressed these aspects of the university-industry relationship
with Six Sigma as follows: For the first question a literature review covering a
fourteen year timeframe was undertaken to describe the trends, sources and findings in
the publications on Six Sigma. Secondly, case base training was examined as a method
to improve Six Sigma education and increase usage on the job among university
student learners. Third, Six Sigma’s acceptance and implementation by a large
number of companies has resulted in the accumulation of a substantial database on
quality improvement projects. The study explored methods to utilize this new data
source to benefit industry management in their decision making process concerning
quality policies.
6.2 Summary of Findings

The literature search resulted in the identification of 201 articles published
between 1990 and 2003. Although this review cannot claim to be exhaustive, it does
provide reasonable insights into the state-of-the-art. As the nature of research on Six
Sigma is difficult to confine to specific disciplines, the relevant material is scattered
across various journals. It is felt that the results have several important implications.
127
To a great extent, the financial impact of Six Sigma on operational
performance is well established. That follows because a defining principle of Six
Sigma is that such justifications are needed for each Six Sigma project. The impacts
on overall company performance as revealed in stock performance are not fully
resolved. Some authors found hints of short-lived abnormal stock performance
associated with the decisions to start Six Sigma programs. Yet, those authors found no
statistical significance in relation to these claims nor evidence of long-term effects.
This suggests a need for additional data collection and analysis to answer the
important question of long-term impacts of decisions to adopt Six Sigma programs.
It is proposed from the study that Six Sigma can be defined as a component
based method involving Define, Measure, Analyze, Improve and Control (DMAIC) or
(DMADV) and two principles. These principles relate both to building and
maintaining management support and to fostering usage of methods among
practitioners who are not experts in statistics. The first principle emphasizes attention
to the bottom line in initiating projects. The second principle emphasized the training
of non-statisticians with minimal theory. Trends in the literature include an increasing
academic participation and a broader focus than solely on manufacturing.
Three main types of contributions of Six Sigma to academia were identified
embodied in the literature: (1) increased emphasis on complete case studies compared
with single sub-method applications, (2) new, relatively specific core and
infrastructure practices, and (3) the development of a large new market of industrial
non-experts who might be interested in practically oriented research and new methods.
128
While over 50% of the articles in the database either explicitly or implicitly
recommend Six Sigma programs, empirical study of the appropriateness of
implementing Six Sigma in specific business contexts has, apparently, not been
investigated. Related, largely unanswered questions include: How can data about any
specific company’s management, training programs, or environment be useful in
decision-making about the adoption of Six Sigma programs?
Focusing on the second principle covered in the definition of Six Sigma which
emphasizes training, an examination of case-based instruction methods was
undertaken. Case-based instruction was introduced into senior and introductory
graduate level engineering courses on statistical process control (SPC) and design of
experiments (DOE). Students seem to identify with the case study "stories" much
better than the traditional lectures. Before implementation of the case study exercises,
end-of quarter evaluations by students for the two classes had been below the average
end-of-quarter class evaluations in the department. After the implementation of the
case study approach and with minimal other changes, the evaluations climbed well
above average. This dramatic turn-around can mainly be attributed to the introduction
of the case based method. Further, presenting applications of course methods in which
literally millions of dollars are saved appears to effectively motivate students
concerned about job security and advancement in the highly competitive industrial
marketplace.
One outcome of the wide spread acceptance of Six Sigma and with its
systematic program is the growing database that now exist within industry and specific
129
companies on individual project improvement activities. The purpose of this study was
to look at the database in a way that could help management better run improvement
projects. Three possible analysis methods investigated for this task were regression,
SPC, and Markov Decision Processes (MDP). The study was based on 39 quality/cost
improvement projects at a medium size U.S. manufacturing company with data
collected over a 30 month period. It was found that by viewing the quality
improvement process in the same light as other processes and applying such
techniques to the accumulating database could provide better insight for operational
managers dealing with defective implementation decisions.
6.3 Limitations and Future Research

One of the defining principles of Six Sigma is the financial justification for
each project. Yet, the literature review found no published work showing statistical
significance in relation to the decisions to start Six Sigma programs and stock
performance. Considering that failure to find a significant effect does not constitute
proof, more work is needed for a thorough evaluation of the bottom line impacts of
Six Sigma.
In the contest of Six Sigma, statements abound that are unsupported by
objective evidence. Examples include self reported profits, the effects of success
factors, and advocacy for Six Sigma in general. For example, as noted above, the
impacts on stock performance investigated by Goh et al. (2003) are not fully resolved.
130
This suggests a need for additional data collection and analysis to answer the
important question of long term impacts of decisions to adopt Six Sigma programs.
Snee (1999 and 2000a) calls for research to help practitioners identify a robust
set of improvement tools to be used in conjunction with the DMAIC process. The
focus in these recommendations is not so much on new techniques as on refined
techniques associated with specific phases. However, new techniques might be
relevant to Six Sigma practitioners who are often not experts in statistics.
Additional modeling techniques to predict and evaluate the bottom-line
impacts of projects are needed. This follows because of the central importance of
profit related justifications in Six Sigma for initiating decisions on projects. Our
research did indicate a correlation with training and the bottom-line impact of projects.
New models that are also easy-to-use could be developed with broader applicability
and improved prediction accuracy.
It might also be useful to extend profit models to investigate the selection of
specific core methods or sub-methods in specific situations. These efforts could be
combined with empirical investigations to permit the development of prescriptive
models to aid practitioners from different disciplines select the most advantageous
techniques. This could build on research related to the most appropriate methods for
training black belts, e.g., in Hoerl (2001a), by associating the methods more
specifically to phases in a project and to situations.
Although the industry has an increased interest in Six Sigma implementation
and many companies have gained the profits and advantages from this disciplined
131
approach, the research of the impacts of Six Sigma implantation and factors
contributed to Six Sigma success remain unclear. Even the existent studies are not
well integrated and the research is mostly anecdotal. Current concepts in the field of
Six Sigma are largely based upon case studies, anecdotal evidence and the
prescriptions of leading “gurus.” Consequently there is little consensus on which
factors are critical to the success of the approach. Most of the articles reported that top
management leadership is the main factor to Six Sigma success [Blakeslee (1999) and
Scalise (2001)]. However, many other factors affecting Six Sigma’s success are
important and need to be better documented.
As part of this work the database was examined at one manufacturing
company. This comprised 39 improvement projects conducted over a 30 month time
frame. This is limited data, collected at one site and no definitive conclusions should
be drawn. The data does provide a start into the use and modeling of Six Sigma,
utilizing the growing database describing the performance of improvement projects
and the methods used that is being generated as a result of it implementation. The
attempt to build a theory of how and why Six Sigma works is aimed at building a
prescriptive model. From this, managers would be able to identify which activities
from which programs are more or less likely to be useful in their situations, as well as
which of their goals would be most affected. An example is the correlation between
training and increased profit seen in this study. With the future success of corporations
riding on the outcome, there is a need for more theory to explain the differences
between successful and unsuccessful efforts.
132
APPENDIX A: THE ARTICLE DATABASE
The descriptors used in the table below are defined in Section 3. The complete
references to the articles are described below in the reference section.
Topics Version 2 Sousa and Voss (2002)

Topics Version 1 Oakland (1989)
Management committed
Speculative in Nature?
Change Management
Structured approach
Research Approach
Team involvement
Customer focused
Project leadership
Adaptable system
Project selection
Industrial Sector
Success Factors
Define DMAIC
Impact factor
Data system
Goals based
Bottom line
Authorship
Right team
Define 3.4
Training
Year
Author(s)
Pe
Abraham et.al. 2001 A 0 0 To Ph To G 1.5 Co N N - - - - - - - - - - - - -
Pe
Ackermann 1993 I 0 0 To Ph To G 0.7 C N N - - - - - - - - - - - - -
Ackermann et.al. 1993 I 0 1 To To M 0.7 C N N - - - - - - - - - - - - -
Ali et.al. 1999 I 0 0 To To M 1.5 C N N - - - - - - - - - - - - -
Antony et.al. 2002 I 0 1 Sy Ph M 0.3 C Y N - - - - - - - - - - - - -
Arvidsson 2003 A 0 0 Sy Ph M 0.3 Su N N - - - - - - - - - - - - -
Pe
Bailey 2001 I 0 0 Sy Ph M 1.5 C N N - - - - - - - - - - - - -
Bartos 1999 I 0 0 Sy Ph M 0.3 TA Y Y - Y - - - - - - - - - - -
Basu 2001 I 0 1 Sy Ph M 0.2 Co N N - - - - - - - - - - - - -
N
Behara et.al. 1995 I A 0 1 Sy Ph M A C N N - - - - - - - - - - - - -
133
N
Benedetto 2003 I 1 0 Sy Ph Se A C N Y Y - Y Y - - - - - - - - -
Berlowitz 2003 A 0 1 To To Se 2.9 C R N Y - - - - Y - - - - - - - -
Binder 1997 I 1 1 Sy Ph M 0.8 Co N N - - - - - - - - - - - - -
N
Bisgaard et.al. 2000 A 0 0 Sy Ph M A TA Y N - - - - - - - - - - - - -
Blakeslee 1999 I 0 0 Sy Ph Se 0.2 C Y Y Y - Y - Y Y - Y - - - - -
N
Blanton 2002 A 0 0 Sy Ph Se A C Y N - - - - - - - - - - - - -
Bossert 2003 I 0 0 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
Breyfogle 2002 I 0 0 Sy Ph Se 0.2 C N N - - - - - - - - - - - - -
Breyfogle et.al. 2003 I 0 0 Sy Ph M 0.7 TA Y N - - - - - - - - - - - - -
N
Breyfogle et.al. 2002 I 1 1 Sy Ph Se A TA Y N - - - - - - - - - - - - -
Breyfogle et.al. Pe
(a) 2001 I 0 0 Sy Ph G 1.5 TA Y N - - - - - - - - - - - - -
Breyfogle et.al.
(b) 2001 I 0 0 To To G 0.2 C Y N - - - - - - - - - - - - -
Broderick et.al. 2002 A 0 0 Sy Ph Se 4.8 C N N - - - - - - - - - - - - -
N
Buck et.al. 2001 I 1 0 Sy Ph Se A C N Y Y - - - - - - - - - - - -
C Su
Buck 1998 I 1 1 Sy Ph Se 1.9 R N N - - - - - - - - - - - - -
N
Buck 2001 I 1 1 Sy Ph Se A C N N - - - - - - - - - - - - -
Buggie 2000 I 0 0 Sy Ph M 0.1 TA Y N - - - - - - - - - - - - -
Card 2000 I 0 0 To To M 0.8 Co N N - - - - - - - - - - - - -
Sy
Caulcutt 2001 I 1 1 To Ph Pr M 0.3 C N Y - - - - Y - - - - - - - -
Chan et.al. 2001 A 0 1 To To M 0.4 C Y N - - - - - - - - - - - - -
Chassin 1998 A 0 1 Sy Ph Se 1.9 R N N - - - - - - - - - - - - -
Chowdhury 2000 I 1 0 Sy Ph G 0.3 TA Y Y - - Y - - - - - - - - - -
N
Clifford 2001 I 0 1 Sy Ph M A C CoN Y Y - - - - - - - - - - - -
Coleman et.al. 2001 I A 0 0 To To M 0.3 TA Y N - - - - - - - - - - - - -
Connolly 2003 I 0 0 Sy Ph M 0.7 C N N - - - - - - - - - - - - -
Conner 2003 I 0 0 Sy Ph M 0.3 C N N - - - - - - - - - - - - -
Cooper 1992 I 0 1 Sy Ph M 0.1 C N N - - - - - - - - - - - - -
Pe
Cooper 2003 I 0 0 Sy Ph Se 0.2 Su N Y - - - - - Y - - - - - - -
Pe
Crom 2000 I 0 0 Sy Ph G 0.2 Co Y Y Y - - - - - - - - - - - -
Pe N
Dasgupta 2003 A 1 1 Sy Ph Se A C Y Y Y - - - - - - - - - - - -
Davies 2001 A 0 0 Sy Ph Se 0.8 R N N - - - - - - - - - - - - -
Davig et.al. 2003 A 0 0 Sy Ph M 0.2 Su N Y Y - - - - - - - - - - - -
De Mast 2003 A 0 0 Sy Ph G 0.2 Co Y N - - - - - - - - - - - - -
De Mast et.al. 2000 A 1 0 Sy Ph G 0.2 Co N N - - - - - - - - - - - - -
134
N
Dedhia 1995 I 0 0 Sy Ph Se A TA Y Y Y - Y - - - - - - - - - -
Pe N
DeFeo 2000 I 0 1 Sy Ph G A Su N N - - - - - - - - - - - - -
Deshpande 1998 A 0 0 To To M 0.2 C Y N - - - - - - - - - - - - -
Deshpande et.al. 1999 I A 1 1 Sy Ph G 0.4 C N N - - - - - - - - - - - - -
N
Does et.al. 2002 A 1 0 Sy Ph Se A Co N N - - - - - - - - - - - - -
Doganaksoy et.al. 2000 I A 0 0 To To M 0.2 C Y N - - - - - - - - - - - - -
Dornheim 2001 I 0 0 To To M 0.3 TA Y N - - - - - - - - - - - - -
N
Douglas 2000 I 0 0 Sy Ph G A C Su N N - - - - - - - - - - - - -
Du et.al. 2000 A 0 1 To Pr To M 0.5 Co Y N - - - - - - - - - - - - -
Duguesaoy et.al. 2002 I 0 0 To To G 0.4 C N N - - - - - - - - - - - - -
Eid et.al. 1997 A 0 0 To To M 0.4 TA Y N - - - - - - - - - - - - -
Farntz 2001 I 1 0 Sy Ph M 0.3 C N N - - - - - - - - - - - - -
Feng et.al. 1997 A 0 0 To To M 0.4 C N N - - - - - - - - - - - - -
N
Ferrin et.al. 2002 I 1 1 To To M A Su N N - - - - - - - - - - - - -
Finn 1999 I 1 0 Sy Ph M 0.3 TA Y N - - - - - - - - - - - - -
Fontenot et.al. 1994 I A 0 1 To To M 0.2 Su N N - - - - - - - - - - - - -
N
Fuller 2000 I 0 0 Sy Ph M A TA Y Y - - - - - - - - - - Y - -
N
Gano 2001 A 0 0 To To G A C N N - - - - - - - - - - - - -
Gautreau et.al. 1997 A 0 0 To To M 0.4 TA Y N - - - - - - - - - - - - -
N
Gill 1990 I 0 1 Sy Ph M A Su N N - - - - - - - - - - - - -
Gnibus 2000 I 0 0 To To G 0.2 C Y N - - - - - - - - - - - - -
N
Goh (a) 2002 A 1 1 Sy Ph G A Co N N - - - - - - - - - - - - -
Pe
Goh (b) 2002 A 1 0 Sy Ph G 0.2 R N Y - - Y - - - Y - - - - - -
Goh 2001 A 0 0 To To M 0.3 TA Y Y Y - - - - - - - - - - - -
N
Goh et.al. (a) 2003 A 1 0 Sy Ph M A R N N - - - - - - - - - - - - -
N
Goh et.al. (b) 2003 A 1 1 To To G A TA Y N - - - - - - - - - - - - -
Gordon 2002 I 0 0 Sy Ph G 0.2 Co Y N - - - - - - - - - - - - -
N
Grandzol et.al. 1998 A 0 0 To Pr M A Su N N - - - - - - - - - - - - -
Greek 2000 I 0 1 Sy Ph M 0.3 Su N Y Y - - - - - - - - - - - -
Gross 2001 I 0 0 Sy Ph G 0.2 TA Y Y Y - - - - - - - - - - - -
Pe
Hahn et.al. 1998 I 0 0 Sy Ph M 0.2 TA N N - - - - - - - - - - - - -
Pe
Hahn et.al. 1999 I 1 1 Sy Ph M 1.2 TA N Y - - - - - - Y - - - - - -
Pe
Hahn 2002 I 0 0 Sy Ph G 1.2 TA Y N - - - - - - - - - - - - -
135
N
Hahn et.al. 2000 I 1 1 Sy Ph G A TA Y Y Y - - - Y - Y - - - Y - -
Hahn et.al. 2001 I 0 0 To To G 0.2 TA Y Y - - Y - - - - - - - - - -
N
Hammer 2002 I 1 0 Sy Ph M A TA N Y - - - - - - Y - - - - - -
Pe
Harrold 1999 I 1 1 Sy Ph M 0.3 C N Y - - - - - Y - - - - - - -
Harrold et.al. 1999 I 1 0 Sy Ph M 0.3 TA Y N - - - - - - - - - - - - -
Pe
Harry 1998 I 1 1 Sy Ph M 0.2 C N Y - - Y - - - - - - - - - -
Sy
Harry (a) 2000 I 0 0 To Ph To G 0.2 Su N Y - - - - Y - - Y - - - - -
Harry (b) 2000 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
Harry (c) 2000 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
Harry (d) 2000 I 0 1 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
Harry (e) 2000 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
Harry (f) 2000 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
Henretta et.al. 2003 I 1 1 Sy Ph Se0.2 C N N - - - - - - - - - - - - -
N
Hild et.al. 2000 I 0 0 Sy Ph M A Co N N - - - - - - - - - - - - -
Pe
Hill 2001 I 0 0 To Ph To M 1.5 C N N - - - - - - - - - - - - -
Hoerl 1998 I 0 0 Sy Ph G 0.2 C N Y - Y - - - - - - - - - - -
Pe
Hoerl (a) 2001 I 0 0 Sy Ph G 1.5 TA N Y - - - - - - - - Y - - - -
Pe
Hoerl (b) 2001 I 0 0 Sy Ph G 1.5 TA Y N - - - - - - - - - - - - -
Horst 1999 I 0 0 Sy Ph Se 0.3 C CoN N - - - - - - - - - - - - -
Howell 2000 I 0 0 Sy Ph M 0.3 Su N N - - - - - - - - - - - - -
Howell 2001 I 0 0 Sy Ph M 0.3 Su N Y Y - - - - - - - - - - - -
Hunter 1999 I 0 0 Sy Ph Se 0.3 C N N - - - - - - - - - - - - -
Hunter 2000 I 0 0 Sy Ph G 0.3 TA Y N - - - - - - - - - - - - -
Hunter et.al. 1999 I 0 0 Sy Ph M 0.3 C N Y Y - - - - - - - - - - - -
Pe
Hutchins 2000 I 0 0 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
Pe N
Ingle et.al. 2001 I A 1 1 Sy Ph M A Co N N - - - - - - - - - - - - -
Pe
Johnson 2002 I 0 1 Sy Ph G 0.3 TA Y N - - - - - - - - - - - - -
Johnson et.al. 2003 I 1 1 Sy Ph M 0.3 Su N N - - - - - - - - - - - - -
Johnstone et.al. 2002 I 0 1 Sy Ph Se 2.9 C N N - - - - - - - - - - - - -
Sy
Johnstone et.al. 2003 I 0 1 To Ph Pr Se 1.2 C N Y Y - - - - - - - - - - - -
Kandebo 1999 I 0 0 Sy Ph M 0.3 TA N N - - - - - - - - - - - - -
Kane 1998 I 0 1 Sy Ph M 0.2 C N N - - - - - - - - - - - - -
N
Kazmer et.al. 2002 A 0 1 To To M A C Y N - - - - - - - - - - - - -
Kazmierczak 2003 A 0 1 To To Se 1.6 R N N - - - - - - - - - - - - -
136
Kendall et.al. 2000 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
Kenett et.al. 2003 I A 0 0 To To M 0.2 C N N - - - - - - - - - - - - -
Knowles et.al. 2003 A 0 0 To To Se 0.2 C N N - - - - - - - - - - - - -
Koch 2002 I 0 1 To To M 0.8 C N N - - - - - - - - - - - - -
Koonce et al. 2003 A 0 0 To To M 0.4 C Y N - - - - - - - - - - - - -
Krouwer 2002 I 0 1 To To Se 0.8 C N N - - - - - - - - - - - - -
Kunes 2002 I 1 0 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
Landin et.al. 2001 A 0 0 Sy Ph G 0.3 Su N Y - - - - - - - - - - Y - -
Sy
Leffew et.al. 2001 I A 1 1 To Ph To M 0.4 C N N - - - - - - - - - - - - -
Linderman et.al. 2003 A 1 1 Sy Ph G 1.5 TA Y Y - - - - - - - - - Y - - -
Sy
Lucas (a) 2002 I 1 1 To Pr To G 0.2 TA Y Y Y - - - - - - - - - - - -
Lucas (b) 2002 I 0 0 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
Mader 2002 I 0 0 Sy Ph M 0.2 TA Y N - - - - - - - - - - - - -
Maguire (a) 1999 I 0 1 To To M 0.2 TA N N - - - - - - - - - - - - -
Magure (b) 1999 I 0 1 Sy Ph M 0.2 C N Y - - - - - - - - - - Y - -
Pe N
Mandal et.al. 1998 A 0 0 To Ph Pr G A Su R N Y Y Y Y - - - - - - - Y - -
Mason et.al. 2000 I 1 0 To To G 0.2 C N Y - - - - - - - - - - - - -
N
McCarthy et.al. 2001 I 1 1 To To M A C N N - - - - - - - - - - - - -
McFadde 1993 A 1 1 Sy Ph M 0.2 TA Y N - - - - - - - - - - - - -
Pe
Montgomery 2000 A 0 0 To Ph Pr G 0.2 TA Y N - - - - - - - - - - - - -
Montgomery 2001 A 0 0 Sy Ph G 0.2 TA Y Y Y - - - - - - - - Y - - -
Pe
Montgomery 2002 A 0 0 To Ph Pr G 0.2 TA Y N - - - - - - - - - - - - -
Montgomery et Pe
al. 2001 I A 0 0 Sy Ph G 1.5 TA Y N - - - - - - - - - - - - -
Mukesh 2003 I 1 1 Sy Ph M 0.4 C N N - - - - - - - - - - - - -
Munro 2000 I 0 1 Sy Ph M 0.2 Co N N - - - - - - - - - - - - -
Murugappan Sy
et.al. 2003 I 0 0 To Ph Pr Se 0.8 C N N - - - - - - - - - - - - -
Nave 2002 I 1 0 Sy Ph G 0.2 Co Y N - - - - - - - - - - - - -
Neuscheler et.al. 2001 I 1 0 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
Nevalainen et.al.
(a) 2000 I 0 0 Sy Ph Se 1.3 TA Y N - - - - - - - - - - - - -
Nevalainen et.al.
(b) 2000 I 1 1 Sy Ph Se1.3 Co N N - - - - - - - - - - - - -
Nielsen et.al. 1999 I 0 0 Sy Ph Se0.3 C N N - - - - - - - - - - - - -
Noble 2001 I 0 1 Sy Ph M 0.4 Su N N - - - - - - - - - - - - -
Olexa 2003 I 0 0 To Pr M 0.3 C N Y - - - - Y - - Y - - - - -
Pearson 2001 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
N
Plotkin et.al. 1999 I 0 1 Sy Ph M A C N N - - - - - - - - - - - - -
N
Pyzdek (a) 2001 I 0 0 Sy Ph M A Co N N - - - - - - - - - - - - -
137
Pe
Pyzdek (b) 2001 I 0 0 To Ph To G 1.5 TA Y N - - - - - - - - - - - - -
N
Ramberg 2000 A 0 1 Sy Ph G A TA Y Y - - - - - - - - Y - - - -
N
Rasis et.al. (a) 2003 I A 1 1 Sy Ph M A C N N - - - - - - - - - - - - -
N
Rasis et.al. (b) 2003 I A 1 0 Sy Ph M A C N N - - - - - - - - - - - - -
N
Rayner 1990 I 0 0 Sy Ph G A C N N - - - - - - - - - - - - -
Ribardo et.al. 2003 I A 0 0 To To M 0.2 C N N - - - - - - - - - - - - -
N
Riley et al. 2002 A 1 0 Sy Ph Se A C N N - - - - - - - - - - - - -
Rowlands et.al. 2003 A 0 0 To To M 0.6 C N N - - - - - - - - - - - - -
N
Sanders et.al. (a) 2000 I 0 0 Sy Ph G A TA Y Y Y - Y - - - - Y - - - Y -
N
Sanders et.al. (b) 2000 I 0 0 Sy Ph Se A C N N - - - - - - - - - - - - -
N
Sanders et.al. 2001 I 0 0 Sy Ph G A TA Y N - - - - - - - - - - - - -
Sarewitz 2000 I 0 0 Sy Ph Se 1.3 TA Y N - - - - - - - - - - - - -
N
Scalise 2001 I 0 0 Sy Ph M A Su N N - - - - - - - - - - - - -
N
Scalise 2003 I 0 0 Sy Ph Se A C N N - - - - - - - - - - - - -
Schmitt 2000 I 0 1 Sy Ph M 0.3 Su N N - - - - - - - - - - - - -
Schmitt 2001 I 0 1 Sy Ph M 0.3 Su N N - - - - - - - - - - - - -
Schmitt 2002 I 0 1 Sy Ph G 0.3 Su N N - - - - - - - - - - - - -
Sigal et al. 2001 A 1 0 Sy Ph Se 4.8 C N N - - - - - - - - - - - - -
Smith 2003 I 0 0 Sy Ph M 0.2 C N Y Y - - - - - - - - - - - -
Pe
Snee 1999 I 1 1 Sy Ph G 0.2 TA Y Y - - - - - Y - Y - - Y - -
Pe N
Snee (a) 2000 I 0 1 Sy Ph G A R Y N - - - - - - - - - - - - -
Pe
Snee (b) 2000 I 0 0 To Ph To G 0.2 TA Y Y - - Y - - - - - - - - - -
Snee (a) 2001 I 0 0 To To G 0.2 TA Y Y Y - - - - - Y - - - - - -
Pe
Snee (b) 2001 I 0 0 To Ph To G 1.5 TA Y N - - - - - - - - - - - - -
Snee 2003 I 1 0 Sy Ph G 0.2 TA Y N - - - - - - - - - - - - -
N
Stamatis 2000 I 0 0 Sy Ph M A C CoY N - - - - - - - - - - - - -
Stein 2001 I 0 0 To To G 0.2 TA Y N - - - - - - - - - - - - -
Studt 2002 I 1 0 Sy Ph G 0.7 TA Y Y - Y - - - - - - - - - - -
Takikamalla 1994 A 0 1 To To M 0.2 TA Y N - - - - - - - - - - - - -
Tang et.al. 1997 A 0 0 To To M 0.2 Co Y N - - - - - - - - - - - - -
Treichler et.al. 2002 I 0 0 To Pr G 0.2 C N Y Y Y Y - Y Y - - - - Y Y -
Trivedi 2002 I 0 0 Sy Ph M 0.4 C N Y - - Y - - - - - - - - - -
Tylutki et.al. 2002 A 0 0 Sy Ph Se 0.2 C N N - - - - - - - - - - - - -
138
Vandenbrande 1998 I 0 0 To To M 0.2 TA Y N - - - - - - - - - - - - -
N
Vaugham 1998 A 0 1 To To G A C N N - - - - - - - - - - - - -
Velocci (a) 1998 I 0 1 Sy Ph M 0.3 Co N N - - - - - - - - - - - - -
Velocci (b) 1998 I 0 0 Sy Ph M 0.3 C N N - - - - - - - - - - - - -
Velocci (c) 1998 I 0 0 Sy Ph M 0.3 C N N - - - - - - - - - - - - -
Velocci 2000 I 0 1 Sy Ph M 0.3 C N N - - - - - - - - - - - - -
Velocci 2002 I 0 0 Sy Ph M 0.3 Su N Y - Y - - Y - - Y - - - Y Y
Voelkel 2002 A 0 0 To To G 0.2 C N Y - - Y - - - - - - - - - -
Walsh et.al. 2000 I 0 0 Sy Ph G 0.3 Su N N - - - - - - - - - - - - -
Pe
Watson 2000 I 0 0 Sy Ph G 0.2 Co N N - - - - - - - - - - - - -
Watson (a) 2002 I 0 0 To To Se 0.2 C CoN N - - - - - - - - - - - - -
Watson (b) 2002 I 1 0 Sy Ph Se 0.5 TA Y N - - - - - - - - - - - - -
Waurayniak 2002 I 0 0 To To M 0.3 Su N Y - - - - Y - - - - - - - -
Pe
Weinstein et.al. 1998 A 0 0 Sy Ph G 0.2 Su N N - - - - - - - - - - - - -
Westgard 2002 I 0 0 To Pr To Se 4.4 TA N N - - - - - - - - - - - - -
Wheeler 2002 I 0 0 Sy Ph M 0.4 Co N Y - - - - Y - - - - - - - -
N
Wiklund et.al. 2002 A 0 0 Sy Ph Se A C N Y Y - Y - Y - - - - - Y - -
Wood 2001 I 0 0 Sy Ph G 0.3 C N Y Y Y - - - - - - - - - - -
Pe N
Wyper et.al. 2000 I 1 1 Sy Ph Se A C N N - - - - - - - - - - - - -
Yeung et.al. 2003 A 0 0 Sy Ph M 1.5 Su N Y Y - - - - - - - - - - - -
Yu et.al. 1994 A 0 0 Sy Ph M 0.4 R N N - - - - - - - - - - - - -
139
APPENDIX B: PROJECT DATABASE
Tabulated data from the 39 quality improvement and cost reduction projects
used in the Meso-analysis study is presented on the next two pages.
140
Exp Exp
Project Savings Time M/I A/P #people EC CH TF PM CE GR
1 $35000 L M A 7 0 1 1 2 1 0
2 $70000 L M A 1 1 1 0 0 0 0
3 $81315 M M A 2 1 1 1 1 0 0
4 $40000 M M A 1 0 0 0 1 0 0
5 $250000 L I P 6 1 1 1 0 2 2
6 $150000 L M P 4 0 1 1 1 0 0
7 $125000 L I P 3 0 1 1 0 0 1
8 $2200000 L M P 9 0 1 0 0 3 0
9 $50000 M M P 5 1 1 1 1 1 1
10 $39195 M M P 1 1 1 0 0 0 0
11 $34500 L M A 1 1 0 0 1 1 1
12 $21000 L M A 1 0 1 0 0 0 0
13 $25000 M M A 1 0 0 0 1 0 0
14 $20000 M M A 1 0 0 0 1 0 0
15 $10000 M M A 1 0 1 0 0 0 0
16 $20000 S M A 1 0 0 0 0 0 0
17 $28000 M I P 1 0 0 0 1 0 0
18 $20000 S M P 5 0 1 1 0 2 0
19 $20000 S M P 1 0 0 0 1 0 0
20 $4350 S M A 1 0 1 0 1 0 0
21 $13750 S M A 1 0 1 0 1 0 0
22 $8500 S M A 1 0 1 0 1 0 0
23 $1600 S M A 1 1 0 0 0 0 0
24 $12500 S M A 1 0 1 0 1 0 0
25 $4000 S M A 1 0 0 0 0 0 0
26 $13000 S M A 1 0 0 0 0 0 0
27 $15000 L I P 1 1 1 0 0 0 0
28 $6000 M I P 1 1 1 0 1 0 0
29 $11500 M I P 2 0 1 1 1 0 0
30 $4500 M I P 1 1 1 0 1 0 0
31 $11000 S M P 5 0 1 1 0 1 0
32 $5400 S M P 5 0 1 1 1 1 0
33 $150000 S I P 4 0 1 0 1 1 1
34 $8600 S I P 2 1 1 0 0 0 0
35 $90000 M M A 5 1 1 1 1 1 1
36 $30000 M M P 7 1 1 1 0 1 0
37 $45000 S M A 3 0 1 0 0 0 1
38 $240000 S I P 3 1 0 0 0 0 0
39 $50000 S I P 4 1 1 0 1 0 0
141
Project DOE SPC DC FT EA OF Time Cost Act Savings Profit
1 0 0 1 2 0 1 13 $48700 $36000 $-12700
2 1 0 0 1 1 1 18 $7590 $0 $-7590
3 0 0 1 1 1 0 25 $35300 $31500 $-3800
4 0 0 0 0 1 0 20 $2900 $0 $-2900
5 2 0 1 7 0 1 16 $325500 $4E+06 $3874500
6 0 0 1 1 1 0 9 $76000 $170000 $94000
7 1 0 0 2 1 0 7 $17725 $130500 $112775
8 4 0 0 7 4 0 30 $220000 $0 $-220000
9 2 2 1 7 2 1 5.5 $31125 $97800 $66675
10 0 0 1 1 1 1 14 $12350 $19575 $7225
11 0 0 1 3 2 0 18 $22800 $13500 $-9300
12 0 0 0 0 1 0 18 $2600 $0 $-2600
13 0 0 0 0 1 0 18 $2000 $0 $-2000
14 0 0 0 0 1 0 20 $7500 $21740 $14240
15 0 0 1 1 1 1 8 $30800 $17200 $-13600
16 0 0 0 0 1 0 9 $2000 $0 $-2000
17 0 0 2 2 1 0 4 $12000 $7000 $-5000
18 2 1 1 6 0 0 1.5 $5300 $23220 $17920
19 0 0 1 1 1 0 3 $1900 $8050 $6150
20 0 0 1 1 0 0 3 $1000 $4025 $3025
21 0 0 1 1 0 0 3 $1000 $4025 $3025
22 0 0 1 1 0 0 3 $1000 $4025 $3025
23 0 0 1 1 1 1 3 $3525 $3125 $-400
24 0 0 1 1 0 0 3 $3000 $8400 $5400
25 0 0 0 0 1 0 18 $1900 $0 $-1900
26 0 0 0 0 1 0 8 $1900 $0 $-1900
27 1 0 1 2 1 0 19 $12125 $14985 $2860
28 0 0 1 1 1 0 2.5 $1700 $6500 $4800
29 0 1 0 1 1 1 8 $12880 $11700 $-1180
30 0 0 1 1 1 0 4.5 $3060 $6300 $3240
31 1 2 1 5 0 0 3 $4250 $10900 $6650
32 0 1 1 3 0 0 1.5 $2400 $5375 $2975
33 2 0 1 5 1 0 6 $38900 $165440 $126540
34 0 0 1 1 1 0 1 $1500 $10750 $9250
35 1 1 1 5 1 0 3 $12640 $66100 $53460
36 0 0 1 2 1 1 10 $18780 $34056 $15276
37 1 0 1 3 1 1 13 $38584 $46300 $7716
38 0 1 1 2 1 0 12 $15690 $236280 $220590
39 0 0 1 1 0 0 1.5 $1275 $11927 $10652
142
LIST OF REFERENCES
Abraham and Mackay (2001) Discussion - Six Sigma Black Belts: What do
they need to know? Journal of Quality Technology, 33 (4): pp. 410-413, Oct.
Ackermann C.S. (1993) Supplier improvement via SPC application

workshops, IEEE Transactions on Semiconductor Manufacturing, 6 (2): pp. 178-
183, May
Ackermann C.S., and Fabia J.M. (1993) Monitoring supplier quality at ppm
levels, IEEE Transactions on Semiconductor Manufacturing, 6 (2): pp.189-195,
May
Ahire S., Landeros R., and Golhar D. (1995) Total quality management: a
literature review and an agenda for future research, Production and Operations
Management, pp. 277-307
Ali O.G., Chen Y.T. (1999) Design quality and robustness with neural
networks, IEEE Transactions on Neural Networks, 10 (6): pp. 1518-1527, Nov.
Allen, T. T. (2003) Design of Experiments and Other Six Sigma Related

Methods Explained and Derived, Thomson Custom Publishing, ISBN: 0759-
318638.
Alloway, J. A. (1993), “Enhancing Statistical Education for Engineers,” in

Proceedings of the Section on Statistical Education, Alexandria, VA: The
American Statistical Association, pp. 182-186.
Antony J. and Coronado R.B. (2002) Design for Six Sigma, Manufacturing
Engineer, 81 (1): pp. 24-26
Arvidsson M., Gremyr I., and Johansson P. (2003) Use and knowledge of
robust design methodology: a survey of Swedish industry, Journal of Engineering
Design, 14(2): pp. 129-143, Jun.
143
Bailey S.P. (2001) Discussion - Six Sigma Black Belts: What do they need to
know? Journal of Quality Technology, 33 (4): pp. 426-431, Oct.
Barton, R. and C. A. Nowack (1998), “A One-Semester, Laboratory-Based,

Quality-Oriented Statistics Curriculum for Engineering Students,” The American
Statistician, Vol. 52, No. 3, pp. 233–237.
Bartos, F.J. (1999) Six sigma for complex systems, Control Engineering,
46(3): pp. 90, Mar.
Basu R. (2001) Six sigma to fit sigma, IIE Solutions, 33 (7): pp. 28-33, Jul.
Behara R.S., Fontenot G.F., and Gresham A. (1995) Customer Satisfaction

Measurement and Analysis Using Six Sigma, International Journal of Quality &
Reliability Management, 12 (3): pp. 9-18
Benedetto A.R. (2003) Adopting manufacturing-based Six Sigma

methodology to the service environment of a radiology film library, Journal of
Healthcare Management 48 (4): pp. 263-280, Jul-Aug.
Berlowitz, D.R. (2003) Striving for Six Sigma I pressure ulcer care, Journal of
American Geriatrics Society, 51(9): pp. 1320-1321, Sept.
Bhote, K. R. (1988), World Class Quality: Design of Experiments Made

Easier, More Cost Effective than SPC, New York: American Management
Association.
Binder R.V. (1997) Can a manufacturing quality model work for software?
IEEE Software, 14 (5): pp. 101, Sep-Oct.
Bisgaard, S. (1991), "Teaching Statistics to Engineers," The American

Statistician, Vol. 45, No., pp. 274-283.
Bisgaard, S. (1998), “Discussion of A One-Semester, Laboratory-Based,

Quality-Oriented Statistics Curriculum for Engineering Students,” The American
Statistician, Vol. 52, No. 3, pp. 238–239.
Bisgaard S. and Freiesleben J. (2000) Quality Quandaries: Economics of Six

Sigma Program, Quality Engineering, 13 (2), pp. 325-331
Blakeslee J.A. (1999) Implementing the Six Sigma Solution – How to achieve
quantum leaps in quality and competitiveness, Quality Progress, 32 (7): pp. 77-85,
Jul.
144
Blanton P. (2002) Quality tools in science education. The Physics Teacher,
40: pp.188-189, Mar.
Bossert J. (2003) Lean and Six Sigma – Synergy made in heaven, Quality
Progress, 36 (7): pp. 31-32 Jul.
Brady, J. and T. Allen (2002) Case Study Based Instruction of SPC and DOE,
The American Statistician, 56, 4, 1-4.
Breyfogle, F.W., (1999) Implementing Six Sigma: Smarter Solutions using

Statistical Methods, Wiley-Interscience.
Breyfogle F.W. (2002) Golf and Six Sigma – Use Six Sigma metrics to drive
proper process behavior, Quality Progress, 35 (11): 83-85, Nov.
Breyfogle F.W., Connolly M. (2003) Six sigma methods to ensure

organizations health, R&D Magazine, 45(4): pp. 28-29, Apr.
Breyfogle F.W., Enck D. (2002) Six sigma goes corporate, Business

Management, pp. 70, May 1
Breyfogle F.W., Enck D., and Meadows B. (2001) Discussion - Six Sigma
Black Belts: What do they need to know? Journal of Quality Technology. 33 (4):
pp. 424-425, Oct.
Breyfogle,F.W., and Meadows,B.,(2001) Bottom-line success with Six Sigma

– Define key process output variables and their effects on the cost of poor quality,
Quality Progress, 34 (5): pp. 101-104, May.
Broderick L.S., Knuteso, H.L., Rankin R.J. et. al. (2002) Use of Six Sigma
methodology to enhance capacity management in an academic center-first year’s
experience, Radiology 225: 1223 Suppl. S Nov.
Buck C., Miller R., and Desmarais J. (2001) Six Sigma – The quest for
quality, Hospitals & Health Networks, 75 (12): pp. 41-48, Dec.
Buck C.R. (1998) Health care through a Six Sigma lens, Milbank Quarterly,
76 (4): pp. 749+
Buck C.R. (2001) What Hospital leaders say about Six Sigma, Hospitals &
Health Networks, 75(12): pp. 43, Dec.
145
Buggie F.D. (2000) Beyond ‘Six Sigma’, Journal of Management
Engineering, 16 (4): pp. 28-31, Jul.-Aug.
Card D.N. (2000) Sorting out Six Sigma and the CMM, IEEE Software, 17
(3): pp. 11-13, May-Jun.
Caulcutt R. (2001) Why is Six Sigma so successful? Journal of Applied

Statistics, 28 (3-5): pp. 301-306, Mar-May.
Chan K.K., and Spedding T.A. (2001) On-line Optimization of Quality in a

Manufacturing System, International Journal of Production Research, 39 (6): pp.
1127-1145. Apr.
Chassin M.R. (1998) Is healthcare ready for Six Sigma quality?, Milbank
Quarterly, 76 (4): pp. 565+
Chowdhury S. (2000) Working toward Six Sigma success, Manufacturing

Engineering, 127(1): pp. 14, July
Clifford L. (2001) Trend spotting – Why you can safely ignore Six Sigma,
Fortune, 143 (2): pp. 140, Jan.
Cobb, G. W. (1992), “Report of the Joint ASA/MAA Committee on

Undergraduate Statistics,” in Proceedings of the Section on Statistical Education,
Alexandria, VA: American Statistical Association, pp. 281-283.
Coleman S.Y., Arunakumar G., Foldvary F., et al. (2001) SPC as a tool for
creating a successful business measurement framework, Journal of Applied
Statistics, 28 (3-4): pp. 325-334, Mar. May
Connolly M. (2003) Six Sigma deployment at DuPont, R&D Magazine, 45

(4): pp. 29 Apr.
Connor G. (2003) Benefiting form Six Sigma, Manufacturing Engineering,

130 (2): pp. 53-59 Feb.
Cook, B.M., (1990) In Search of Six Sigma: 99.9997 per cent Defect-free,
Industry Week, 1 October, pp. 60-65
Cooper D.W., Baabcock J.V., and Dipietro F. (1992) Application of 6 sigma-

statistical quality-control to monitoring the deposition of contaminating particles,
Journal of The IES, 35 (5): pp. 27-32, Sep-Oct.
146
Cooper N. P. and Noonan P. (2003) Do teams and Six Sigma go together?
Quality Progress, 36 (6): pp. 25-28.
Crom S. (2000) Implementing Six Sigma – A cross-cultural perspective,

Quality Progress, 33 (10): pp. 73-75, Oct.
Crosby, P.B. (1980) Quality is Free: The Art of Making Quality Certain, New
York; McGraw-Hill
Czitrom, V. (1999), “One-Factor-at-a-Time Versus Designed Experiments,”

The American Statistician, Vol. 53, No. 2, pp. 126-131.
Dasgupta T. (2003) Using the six-sigma metric to measure and improve the
performance of a supply chain, Total Quality Management & Business Excellence,
14 (3): pp. 355-366, May.
Davies HTO (2001) Exploring the pathology of quality failings: measuring

quality is not the problem – changing it is, Journal of Evaluation in Clinical
Practice, 7 (2): pp. 243-251, May
Davig W., Brown S., Friel T., and Tabibzadeh K. (2003) Quality management
in small manufacturing, Industrial Management & Data Systems, 103 (1-2): pp.
68-77
Dean, J., Bowen, D. (1994). Managing theory and total quality: improving
research and practice through theory development. Academy of Management
Review 19 (3), pp. 392–418.
Dedhia N.S. (1995) Survive Business challenges with the total quality
management approach, Total Quality Management, 6 (3): pp. 265-272, Jul.
DeFeo J.A. (2000) An ROI story, Training & Development, 54 (7): pp. 25+,
Jul.
De Mast J. (2003) Quality improvement form the viewpoint of statistical method,

Quality and Reliability Engineering International, 19: pp. 255-264
De Mast J., Schippers WAJ, Does RJMM, et al. (2000) Steps and strategies in
process improvement, Quality and Reliability Engineering International, 16 (4):
pp. 301-311, Jul-Aug.
Deming, W.E., (1986) Out of the Crisis, Cambridge: MIT, Center for
Advanced Engineering Study.
147
Deshpande P.B., (1998) Emergine technologies and Six Sigma, Hydrocarbon
Processing, 77(4): pp. 55, Apr.
Deshpande P.B., Makker S.L., and Goldstein M. (1999) Boost

competitiveness via Six Sigma, Chemical Engineering Progress, 95 (9): pp. 65-70,
Sep.
Dimock P.V., Technical Editor, (1977) Engineering and Operations in the Bell
System, Bell Telephone Laboratories, Inc.
Does R., van den Heuvel E., de Mast J. and Bisgaard S. (2002) Quality
quandaries: comparing nonmanufacturing with traditional applications of Six
Sigma, Quality Engineering, 15(1): pp. 177-182
Doganaksoy N., Hahn G.J., Keeker W.Q. (2000) Product life data analysis: A
case study, Quality Progress, 33(6): pp. 115, June
Dornheim M.A. (2001) Implement Six Sigma, Aviation Week & Space
Technology, 155(1): pp. 25, July
Douglas P.C., Erwin J. (2000) Six sigma focus on total customer satisfaction,
Journal of Quality & Participation, 23(2): pp. 45-49
Du X.P., and Chen W. (2000) Towards a better understanding of modeling

feasibility robustness in engineering design, Journal of Mechanical Design, 122
(4): pp. 385-394, Dec.
Duguesaoy L., Berger J.L., Prevot P., and Sandoz-Guermond F. (2002)

SIMPA: A training platform in work station including computing tutors, Lecture
Notes in Computer Science, 2363: pp. 507-520
Eid M. S., Moghrabi C. and Eldin H. K. (1997) A Simulation Approach to

Evaluating Quality/Cost Decision Scenarios, Computers & Industrial Engineering,
33 (1-2): pp. 105-108
Even, M. J. (1987), "Why adults learn in different ways", Lifelong Learning

10(8), pp. 22-27.
Feigenbaum, A.V. (1991) Total Quality Control, New York; McGraw_Hill
148
Feng CXJ and Kusiak A. (1997) Robust tolerance design with the integer
programming approach, Journal of Manufacturing Science and Engineering
Transactions of the ASME, 119 (4A): pp. 603-610, Nov.
Ferrin D.M., Muthler D. and Miller M.J. (2002) Six Sigma and simulation, so
what’s the correlation? Proceedings of the 2002 Winter Simulation Conference,
pp. 1439-1443.
Finn G.A. (1999) Six-sigma quality for virtual products, Manufacturing

Engineering, 123 (6): pp. 20, Dec.
Friedman N. and Singer Y. (1999) Efficient Bayesian parameter estimation in

large discrete domains, In Advances in Neural Information Processing Systems 11.
MIT Press, Cambridge, Mass.
Folaron J. (2003) The Evolution of Six Sigma, Six Sigma Forum Magazine,
pp. 38-44, Aug.
Fontenot G., Behara R. and Gresham A. (1994) Six Sigma in customer

satisfaction, Quality Progress, pp. 73-76, Dec.
Frantz K. (2001) Apply quality to motion control, Control Engineering,

48(10): pp. 8-10, Oct.
Fuller H.T. (2000) Observations about the success and evaluation of Six
Sigma at Seagate, Quality Engineering, 12: pp. 311-315
Gano D.L. (2001) Effective problem solving a new way of thinking, Annual
Quality Congress Transactions, pp.110-122.
Garvin D.A. (1988) Managing Quality: the strategic and competitive edge,
New York: Free Press
Gautreau N., Yacout S., and Hall R, (1997) Simulation of Partially Observed
Markov Decision Process and Dynamic Quality Improvement, Computers &
Industrial Engineering, 32 (4): pp. 691-700
Gill M.S. (1990) Stalking Six Sigma, Business Month, pp. 42-46, Jan.
Gnibus R.J. (2000) Six Sigma’s Missing Link – Understanding the quality
tool needed to calculate sigma ratings, Quality Progress, 33 (11): pp. 77+, Nov.
149
Goh T.N. (2001) A pragmatic approach to experimental design in industry,
Journal of Applied Statistics, 28 (3-4): pp. 391-398, Mar-May.
Goh T. N. (2002a) The role of statistical design of experiments in Six Sigma:

perspectives of a practitioner, Quality Engineering, 14: pp. 661-673
Goh T. N. (2002b) A Strategic Assessment of Six Sigma, Quality and

Reliability Engineering International, 18: pp. 403-410
Goh T.N., Xie M. (2003) Statistical control of a Six Sigma process, Quality
Engineering, 15 (5): pp. 587-592
Goh T.N., Low P.C., Tsui K.L. and Xie M. (2003) Impact of Six Sigma
implementation on stock price performance, Total Quality Management &
Business Excellence, 14 (7): pp. 753-763
Gordon D.K. (2002) Quality management systems vs. quality improvement,

Quality Progress, 35(11): pp. 86, Nov.
Grandzol and Gershon (1998) A survey instrument for standardizing TQM

modeling research, International Journal of Quality Science, 3 (1): pp. 80-105
Greek D. (2000) Inefficiency won’t wash, Professional Engineering, pp. 45, Jun
Gross J.M. (2001) A road map to Six Sigma quality, Quality Progress, 34
(11): 24-29, Nov.
Hahn G. J. and Hoerl R.W. (1998) Key challenges for statisticians in business
and industry, Quality Progress, 31 (8): pp. 195-200, Aug.
Hahn G. J., Hill W.J., Hoerl R.W., and Zinkgraf S.A. (1999) The Impact of
Six Sigma Improvement – A Glimpse Into the Future of Statistics, The American
Statistician, 53 (3): pp. 208-215, Aug.
Hahn G.J., Doganaksoy N., and Hoerl R. (2000) The Evolution of Six Sigma,
Quality Engineering, 12 (3): pp. 317-326
Hahn G.J., Doganaksoy N., and Stanard C. (2001) Statistical tools for Six
Sigma – What to emphasize and de-emphasize in training, Quality Progress, 34
(9): pp. 78-82, Sep.
Hahn G.J. (2002) Deming and the proactive statistician, The American
Statistician, 56 (4): 290-298, Nov.
150
Hammer M. (2002) Process management and the future of Six Sigma, IEEE
Engineering Management Review, 30 (4): pp. 56-63
Harrold D. (1999) Designing for Six Sigma capability, Control Engineering,

46(1): pp. 62-70, Jan
Harrold D., Bartos F.J. (1999) Optimize existing processes to achieve Six
Sigma capability, Control Engineering, 46(3): pp. 87-103, Mar.
Harry, M.J. (1994) The Vision of Six Sigma: A Roadmap for Breakthrough,
Phoenix: Sigma Publishing Company
Harry, M.J. (1988) The Nature of Six Sigma Quality, Illinois: Motorola
University Press.
Harry M.J. (1998a) Six Sigma: A Breakthrough Strategy for Profitability,

Quality Progress, pp. 60-62, May
Harry M.J. (1998b) Six sigma article inaccurate – Author’s reply, Quality
Progress, 31 (8): pp. 10, Aug.
Harry M.J. (2000a) A new definition aims to connect quality with financial
performance, Quality Progress, 33 (1) pp. 64-66
Harry M.J. (2000b) Six Sigma leads enterprises to coordinate efforts, Quality
Progress, 33 (3): pp. 70-72, Mar.
Harry M.J. (2000c) Six sigma focuses on improvement rates, Quality

Progress, 33 (6), pp. 76-80, Jun.
Harry M.J. (2000d) Abatement of business risk is key to Six Sigma, Quality
Progress, 33 (7), 72+, Jul.
Harry M.J. (2000e) The quality twilight zone, Quality Progress, 33(2): pp. 68,
Feb.
Harry M.J. (2000f) Quality leads, answers follow, Quality Progress, 33(5): pp.
82, May
Harry MJ. and Crawford D. (2005) Six Sigma – The next generation, Machine
Design, pp. 126-132, Feb. 17
151
Harry, M.J. and Schroeder, R. (2000) Six Sigma: The Breakthrough
Management Strategy Revolutionizing the World’s To Corporations, Currency
Hazelrigg, G. A., (1996) Systems Engineering: an Approach to Information-

Based Design, Prentice Hall
Henretta K., Walker J., and Bellafiore L. (2003) Applying “Six Sigma” to
chromatography – Tutorial: Cutting costs through process improvements, Genetic
Engineering News 23 (1): 54-56 Jan.
Hild C., Sanders D. and Cooper T. (2000-01) Six Sigma* On Continuous

Processes: How and Why It Differs, Quality Engineering, 13(1): pp.1-9
Hill W.J. (2001) Discussion - Six Sigma Black Belts: What do they need to
Hoerl R.W. (1998) Six Sigma and the future of the quality profession, Quality
Progress, 31 (6): pp. 35-42, Jun.
Hoerl R. W. (2001a) Six Sigma Black Belts: What Do They Need to Know?
Journal of Quality Technology, 33 (4): PP. 391-406, Oct.
Hoerl R.W. (2001b) Response - Six Sigma Black Belts: What do they need to
know? Journal of Quality Technology, 33 (4): pp. 432-435, Oct,
Hoerl, R. and Snee, R. D. (2001) Statistical Thinking: Improving Business

Performance, Duxbury Press
Horst R.L. (1999) Safety and Six Sigma, Manufacturing Engineering, 122 (2):
pp. 14, Feb.
Howell (1996), "Introducing Cooperative Learning into a Dynamics Lecture

Class", Journal.of Engineering Education, Jan., pp. 69-72.
Howell D. (2000) The power of six, Professional Engineering, 13 (14), pp. 34-
35, Jul. 19
Howell D. (2001) At sixes and sevens, Professional Engineering, pp. 27, May
Hunter D. (1999) Six Sigma steps, Chemical Week, 161 (33): pp. 3, Sep.
Hunter D. and Schmitt B. (1999),CW Conference – Six sigma: Benefits and

approaches, Chemical Week, 1661 (37): pp. 35-36, Oct. 6
152
Hunter J.S. (1989) A one Point Plot Equivalent to the Shewhart Chart with
Western Electric Rules, Quality Engineering, Vol. 2
Hutchins G. (2000) The branding of Six Sigma, Quality Progress, 33 (9): pp.
120-121, Sep.
Ingle S. and Roe W. (2001) Six sigma black belt implementation, The TQM
Magazine, 13(4): pp. 273-280
Ishikawa, K. (1985) What is Total Quality Control?, New Jersey: Prentice-

Hall
Johnson A. (2002) Six sigma in R&D, Research-Technology Management, 45

(2): pp.12-16, Mar-Apr.
Johnson A. and Swisher. B. (2003) Now Six Sigma improves R&D, Research-
Technology Management 46 (2): pp.12-15 Mar-Apr.
Johnstone P.A. and Dernbach A.H. (2002) Six sigma quality and delivery of
radiation therapy, Cancer Journal, 8 (6): pp. 44, Nov-Dec.
Johnstonem P.A.S, Hendrickson J.A.W., Dernbach A.J., et. al. (2003)

Ancillary services in health care industry: is Six Sigma reasonable?, Quality
Management in Health Care, 12(1): pp. 53, Jan-Mar.
Juran, J.M. and Gryna F. (1980) Quality Planning and Analysis, New York:
McGraw-Hill
Juran, J.M. (1989) Juran on Leadership for Quality, New York: The Free
Press
Kandebo S.W. (1999) Lean, Six Sigma yield dividends for C-130J, Aviation
Week and Space Technology, 151 (2): pp. 59-61, Jul. 21
Kane L.A. (1998) The quest for Six Sigma, Hydrocarbon Processing, 77 (2):
pp. 15, Feb.
Kazmer D., Hatch D. and Zhu L. (2002) Investigation of variation and

uncertainty in Six Sigma, Proceedings of the ASME Design Engineering Technical
Conference, 3: pp.21-29
153
Kazmierczak S.C. (2003) Laboratory quality control: Using patient data to
assess analytical performance, Clin Chem Lab Med, 41(5): pp. 617-627
Kendall J. and Fulenwider D.O. (2000) Six sigma, e-commerce pose new
challenges, Quality Progress, 33 (7): 72+, Jul.
Kenett R.S., Coleman S., and Stewardson D. (2003) Statistical efficiency: The
practical perspective, Quality and Reliability Engineering International, 19: pp.
265-272
Knowles G., Vickers G., and Anthony J. (2003) Implementing evaluation of

the measurement process in an automotive manufacturer: a case study, Quality and
Reliability Engineering International. 19: pp. 397-410
Koch P.N. (2002) Probabilistic design: optimizing for Six Sigma quality,
AIAA-2002-1471
Koonce D., Judd R., Sormaz D, et al. (2003) A hierarchical cost estimation
tool, Computers In Industry, 50 (3): pp. 293-302 Apr.
Krouwer J. (2002) Using a learning curve approach to reduce laboratory

errors, Accred Qual Assur, 7: pp. 461-467
Kunes R. (2002) Six Sigma article is misleading, Quality Progress, 35 (3): pp.
8 Mar.
Lucier G.T. and Seshadri S. (2001) GE Takes Six Sigma Beyond The Bottom
Line, Strategic Finance, May, pp. 40-46
Landin A. and Nilsson C.H. (2001) Do quality systems really make a

difference? Building Research and Information, 29 (1): pp. 12-20
Leffew K.W., Yerrapragada S.S., and Deshpande P.B. (2001) 6 sigma and
solid-state polymerization, Chemical Engineering Communications, 188: 109-114.
Linderman K., Schroeder R.G., Zaheer S. and Choo A.S. (2003) Six Sigma: A
goal-theoretic perspective, Journal of Operations Management, 21, (2), pp. 193-
203.
Lucas J. (2002a) The essential Six Sigma – How successful Six Sigma
implementation can improve the bottom line, Quality Progress, 35 (1), pp. 27-31,
Jan.
154
Lucas J. (2002b) Six Sigma article is misleading – Response, Quality
Progress, 35 (3): pp. 8-8 Mar.
Mader D.P. (2002) Design for Six Sigma, Quality Progress, 35 (7): pp. 82+,
Jul.
Maguire M. (1999a) Six sigma saga, Quality Progress, 32 (10): pp. 6, Oct.
Magure M. (1999b) Cowboy Quality: Mikel Harry’s riding tall in the saddle
as Six Sigma makes its mark, Quality Progress, 32 (10): pp. 27-34, Oct.
Main, J., (1994) Quality Wars: The Triumphs and Defeats of American
Business, New York: The Free Press
Mandal P., Howell A. and Sohal A.S. (1998) A systemic approach to quality
improvements: The interactions between the technical, human and quality sytems,
Total Quality Management, 9 (1): pp. 79-100
Martin J. (1982) A garbage can model of the research process, In J. E.

McGrath (Ed)., Judgment calls in research, Beverly Hills, CA: Sage
Mason R.C. and Young J.C. (2000) Interpretive features of a T(2) chart in
multivariate SPC, Quality Progress, 33(4): pp. 84-89, Apr.
MaCarthy B.M. and Stauffer R. (2001) Enhancing Six Sigma through

simulation with IGRAFX process for Six Sigma, Proceeding of the 2001 Winter
Simulation Conference, pp. 1241-1247
McFadden F.R. (1993) Six sigma quality programs, Quality Progress, pp. 37-
42, Jun.
McKeachie, W.J. (1993), Teaching Tips, D.C. Heath and Co., Lexington, MA.
Montgomery, D.C. (2000). Introduction to Statistical Quality Control, 4th ed.

John Wiley and Sons, New York, NY.
Montgomery D. (2000) The present state of industrial statistics, Quality and

Reliability Engineering International, 16: pp. 253-254
Montgomery D. (2001) Editorial, Beyond Six Sigma, Quality and Reliability

Engineering International, 17(4): iii-iv.
155
Montgomery D. (2002) Changing roles for the Industrial Statistician, Quality
and Reliability Engineering International, 18(5):
Montgomery D.C., Lawson C., Molnau W.E, et al. (2001) Discussion - Six
Sigma Black Belts: What do they need to know? Journal of Quality Technology.
33 (4): pp. 407-409 Oct.
Motyka, M. (2000) Six Sigma, QS-9000 article has one minor flaw, Quality
Progress, vol. 33, no. 8, Aug., p.8
Mukesh D. (2003) Putting Six Sigma processes to work, Chemical

Engineering, 110(12): pp. 62, Nov.
Munro R. (2000) Linking Six Sigma with QS-9000. Quality Progress, pp. 47-
53, May
Munro R. (2000) Six sigma, QS-9000 article has one minor flaw – Response,
Quality Progress, 33 (8): pp. 8, Aug.
Murugappan M. and Keeni G. (2003) Blending CMM and Six Sigma to meet
business goals, IEEE Software, (2): 42+ Mar-Apr.
Nave D. (2002) How to compare Six Sigma, lean and the theory of constraints
– A framework for choosing what’s best for your organization, Quality Progress,
35 (3): pp. 73-78, Mar.
Neuscheler F.D. and Norris R. (2001) Capturing financial benefits from Six
Sigma – five lessons learned will resonate with top management, Quality Progress,
34 (5): pp. 39-44, May.
Nevalainen D.E., Berte L., Kraft C., et. al. (2000a) Evaluating laboratory
performance on quality indicators with the Six Sigma scale, Archives of Pathology
& Laboratory Medicine, 124 (4), pp. 516-519, Apr.
Nevalainen D.E., and Berte L. (2000b) Evaluating laboratory performance

with the Six Sigma scale – Reply, Archives of Pathology & Laboratory Medicine,
124 (12): pp. 1748 Dec.
Nielsen K. and Orshal J. (1999) Companies – Dow accelerates Six Sigma

effort; Reports on ‘social responsibility’, Chemical Week, 161 (37): pp. 9, Oct. 6
Noble T. (2001) Six sigma boosts the bottom line, Chemical Engineering
Progress, 97 (4): pp. 9-11, Apr.
156
Nolan, D. and T. P. Speed (1999), “Teaching Statistics Theory Through
Applications,” The American Statistician, Vol. 53, No. 4, pp. 370-375.
Oakland J.S. (1989) Total Quality Management, Butterworth-Heinemann,

London
Olexa R. (2003a) Driving quality with Six Sigma, Manufacturing

Engineering, 130 (2): 61+, Feb.
Pande P.S., Neuman R.P., and Cavanagu R.R., (2000) The Six Sigma Way,
McGraw-Hill
Pande, P. S., L Holpp (2001) “What is Six Sigma?” McGraw-Hill Trade.
Pearson T.A. (2001) Measure for Six Sigma success, Quality Progress, 34 (2):
pp. 35-40, Feb.
Petruccelli, J. D., Nandram, B., and Chen, M.H. (1995), “Implementation of a

Modular Laboratory and Project-Based Statistics Curriculum,” in Proceedings of
the Section on Statistical Education, Alexandria, VA: American Statistical
Association, pp. 165-170.
Plotkin C.W., Carlson C.S., Gregory F.D., et. al. (1999) Panel: Advisory
board – What are the successful companies doing? Annual Reliability and
Maintainability Symposium 1999 Proceedings, pp. 219-223
Pyzdek, T., (2000) Six Sigma Handbook: A Complete Guide for Greenbelts,
Blackbelts, & Managers at All Levels, New York: McGraw-Hill
Pyzdek T. (2001a) Why Six Sigma is not TQM, Quality Digest, pp. 26, Feb.
Pyzdek T. (2001b) Discussion - Six Sigma Black Belts: What do they need to
Ramberg,(2000) Six Sigma: Fad or Fundamental? Quality Digest, pp. 28-32,

May
Rasis D., Gitlow H.S., and Popovich E. (2003a) “Paper Organizers

International: A Fictitious Six Sigma Green Belt Case Study, I”, Quality
Engineering, 15 (1): pp.127-146
157
Rasis D., Gitlow H.S., and Popovich E. (2003b) “Paper Organizers
International: A Fictitious Six Sigma Green Belt Case Study, II”, Quality
Rayner B.C.P. (1990) Market-driven Quality: IBM’s Six Sigma Crusade,

Electronic Business, 1: pp. 68-74, Oct.
Ribardo C. and Allen T. T. (2003) “An Alternative Desirability Function For

Achieving Six Sigma Quality,” Quality and Reliability Engineering International,
19: pp. 1-14
Riley J.B., Justison G.A., Povrzenic D., et al. (2002) Designing an integrated
extracorporeal therapy service quality system, Therap Apher, 6 (4): 282-287, Aug.
Rowlands H. and Antony F. (2003) Application of design of experiments to a

spot welding process, Assembly Automation, 23(3): pp. 273-279
Sanders D. and Hild C. (2000a) A Discussion of Strategies for Six Sigma

Implementation, Quality Engineering, 12 (3): pp. 303-309
Sanders D. and Hild C. (2000b) Six Sigma on Business Processes: Common

Organizational Issues. Quality Engineering, 12(4): pp. 603-610
Sanders D. and Hild C. (2001) Common myths about Six Sigma, Quality
Sarewitz S.J. (2000) Evaluating laboratory performance with the Six Sigma
scale, Archives of Pathology & Laboratory Medicine, 124 (12): pp. 1748, Dec.
Scalise D. (2001) Six sigma: the west for quality, Hospitals & Health
Networks, 75(12): pp. 41, Dec.
Scalise D. (2003) Six Sigma in action – Case studies in quality put theory into
practice, Hospitals & Health Networks 77 (5): pp. 57+ May
Schmitt B. (2000) CW Conference – Moving ahead with Six Sigma, Chemical

Week, 162 (17), pp. 64+, Apr.
Schmitt B. (2001) Expanding Six Sigma, Chemical Week, Feb. 13: pp. 34
Schmitt B. (2002) A slow spread for Six Sigma, Chemical Week, Feb. 21: pp. 21
158
Shewhart W.A. (1931) Economic Control of Manufactured Product, New
York: D. Van Nostrand, Inc.
Shina, S. G., Six Sigma for Electronics Design and Manufacturing, McGraw-
Hill 2002
Sigal R.C., Dessales-Martin D., Ruelle C., et al. (2001) Implementation of a

PACS using Six Sigma methodology, Radiology, 221: 527-527 Suppl. S Nov.
Small, B.B. Chairman Writing Committee, (1956) Statistical Quality Control

Handbook, Western Electric Company, Mack Printing Company Easton, PA
Smith B. (2003) Lean and Six Sigma – A one-two punch, Quality Progress, 36
(4): pp. 37-41 Apr.
Snee R.D. (1999) Why should statisticians pay attention to Six Sigma?
Quality Progress, 32 (9): pp. 100-103, Sept.
Snee R.D. (2000a) Impact of Six Sigma on quality engineering, Quality

Engineering 12 (3): pp. ix-xiv
Snee R.D. (2000b) Six sigma improves both statistical training and processes,
Quality Progress, 33 (10): pp. 68-72, Oct.
Snee R.D. (2001a) Dealing with the Achilles’ heel of Six Sigma initiatives –
Project selection is key to success, Quality Progress, 34 (3): pp. 66 Mar.
Snee R.D. (2001b) Discussion - Six Sigma Black Belts: What do they need to
know? Journal of Quality Technology. 33 (4): pp. 414-417, Oct.
Snee R.D. (2003) The Six Sigma Sweep, Quality Progress, 36 (9): pp. 76+
Sousa R., and Voss C.A. (2002) Quality management re-visited: a reflective
review and agenda for future research, Journal of Operations Management, 20: pp.
91-109
Stamatis (2000) Who needs Six Sigma, anyway? Quality Digest, pp. 33-38,
May
Stein P. (2001) Measurements for business, Quality Progress, 34(2): pp. 29,
Feb.
159
Studt T. (2002) Implementing Six Sigma in R&D, R&D Magazine, 44 (8): pp.
21-23, Aug.
Takikamalla P. R. (1994) The Confusion Over Six Sigma Quality. Quality

Progress, pp. 83-85, Nov,
Tang L.C., Than S.E., and Ang B.W. (1997) A graphical approach to
obtaining confidence limits of C-pk, Quality and Reliability Engineering
International, 13 (6): pp. 337-346, Nov-Dec.
Treichler D., Carmichael R., Kusamanoff A., et al. (2002) Design for Six
Sigma: 15 lessons learned – Leading corporations find out how to avoid pitfalls,
Quality Progress, 35 (1): 33-42, Jan.
Trivedi Y.B. (2002) Applying Six Sigma, Chemical Engineering Progress, 98

(7): 76-81, Jul.
Tylutki T.P., and Fox D.G. (2002) Mooooving toward Six Sigma, Quality
Progress, 35 (2): 34-41, Feb.
Vandenbrande (1998) How to use FMEA to reduce the size of your quality
toolbox. Quality Progress, pp. 97-100, Nov.
Vaugham T.S. (1998) Defect rate estimation for “Six Sigma” processes,
Production & Inventory Management Journal, 39(4): pp. 5-9, Oct.
Velocci A.L. (1998) Pursit of Six Sigma emerges as industry trend, Aviation
Week and Space Technology, 149 (20): 52+, Nov. 16
Velocci A.L. (1998) High hopes riding on Six Sigma at Raytheon, Aviation
Week and Space Technology, 149 (20): 59, Nov. 16
Velocci A.L. (1998) Six sigma takes a back to “Leaan Electronics” at

Rockwell, Aviation Week and Space Technology, 149 (20): 60, Nov. 16
Velocci A.L. (2000) Raytheon Six Sigma meets initial target, Aviation Week
and Space Technology, 152 (13): pp. 59, Mar. 27
Velocci A.L. (2002) Full potential of Six Sigma eludes most companies,
Aviation Week and Space Technology, 157 (14): 56-60 Sep. 30
160
Voelkel J.G. (2002) Something’s missing – An education in statistical
methods will make employees more valuable to Six Sigma corporations, Quality
Progress, 35 (5): 98-101, May
Walsh K., Fuller J., Wood A., et al. (2000) Six Sigma – Marshaling an attack
on costs, Chemical Week, 162 (9): pp. 25-27, Mar. 1
Wankat, P.C. and F.S. Oreovicz. (1993), Teaching Engineering, McGraw-

Hill, NewYork.
Watson G.H. (2000) Toward a Central Tendency of Six Sigma, Quality

Progress, July pp.16
Watson G. H., (2002a) Selling Six Sigma to Upper Management, Six Sigma
Forum Magazine, 1 (4): pp. 26-37, Aug.
Watson G.H. (2002b) Breakthrough in delivering software quality: Capability

maturity model and Six Sigma, Lecture Notes In Computer Science, 2349: pp. 36-
41.
Waurayniak P. (2002) Statistics improve quality, Manufacturing Engineering,

128(2): pp. 39, Feb.
Weinstein, Petrick, and Saunders (1998) What higher education should be

teaching about quality – but is not, Quality Progress, pp. 91-9, Apr.
Westgard J.O. (2002) Evaluation of cardiac troponin assay systems and

validation of QC in accordance with Six-Sigma principles and ISO guidelines,
Clinical Chemistry, 48 (6): C61 Part 2 Suppl. S Jun.
Wheeler J.M. (2002a) Getting started: Six-sigma control of chemical

operations, Chemical Engineering Progress, 98 (9): pp. 76-81, Jun.
Wheeler J.M. (2002b) Getting started with six-sigma, Chemical Engineering

Progress, 98 (9): pp. 8, Aug.
Wiklund H. and Wiklund P.S. (2002) Widening the Six Sigma concept: An
approach to improve organizational learning, Total Quality Management, 13 (2):
pp. 233-239, Mar.
Wood A. (2001a) Management – Making Six Sigma benefits stick, Chemical

Week, 163 (19): pp. 40, May.
161
Wyper B., and Harrison A. (2000) Deployment of Six Sigma methodology in
human resource function: A case study, Total Quality Management, 11 (4-6):
S720-S727 Sp. Iss. SI Jul.
Yeung ACL, Chan L.Y., and Ledd T.S. (2003) An empirical taxonomy for
quality management systems: a study of the Hong Kong electronics industry,
Journal of Operations Management, 21 (1): pp. 45-62
Young, J., (2001), Driving Performance Results at American Express, Six

Sigma Forum Magazine, November, pp. 19-27
Yu B. and Popplewell K. (1994) Metamodel in Manufacturing: a Review,

International Journal of Production Research, 32: pp. 787-796
Zain Z.M., Dale B.G., and Kehoe D.F. (2001) Total quality management: an
examination of the writings form a UK perspective, The TQM Magazine, 13 (2):
pp. 129-137
Zwass V. (1996) Electronic commerce: structure and issues, International

Journal of Electronic Commerce 1 (1): pp. 3-33
162

Six Sigma and The University: Teaching, Research, and Meso-Analysis

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Six Sigma and The University: Teaching, Research, and Meso-Analysis

Caricato da

Copyright:

Formati disponibili

SIX SIGMA AND THE UNIVERSITY: TEACHING,

RESEARCH, AND MESO-ANALYSIS

Presented in Partial Fulfillment of the Requirements for

The Degree Doctor of Philosophy

In the Graduate School of The Ohio State University

James E. Brady, B.S., M.S., MBA, P.E.

The Ohio State University

Dissertation Committee: Approved by

Dr. Theodore T. Allen, Adviser

Dr. Clark A. Mount-Campbell

Six Sigma was introduced by industry practitioners and consultants as a means

descriptions of research trends with an emphasis on establishing its relationship to

management by data and project based data collection, industry is starting to

of individual quality improvement projects. We propose methods specifically for

form a mid-western manufacturing firm.

Reflecting on my long and continuing education process, I find it difficult to

1974 ……………………………………………… BS Aeronautical Engineering

1974-1976 ……………………………………….. US Air Force

1977 ……………………………………………… MS Mechanical Engineering

1977 -2001………………………………………… Lucent Technologies Bell Labs,

1980 ……………………………………………… MBA Marketing

1997 - Present …………………………………… Ohio State University

2001-Present ……………………………………… LaBarge, Inc

Major Field: Industrial and Systems Engineering

Minor Field: Computational Methods

Chapter 1 Introduction ................................................................................................ 1

Appendix A The Article Database.………………………………………………. 133

Appendix B Project Database……………………………………………………. 140

philosophy to help improve business financial performance. It was developed in

main objectives of this dissertation is to review this information and identify

opportunities for additional contributions for the academy.

partnerships are expected to contribute to the advancement of knowledge and

innovation in the production of competitive products, processes and services and

ultimately contribute to the welfare of society.

1. With reference to past academic contributions, what is Six Sigma? Also,

what are the on-going research trends?

This dissertation addresses these three aspects of the university-industry

establishing its relationship to quality management theory and to business practices in

general and to identify topics for future research.

Secondly, case base training is examined as a method to improve Six Sigma

what people need to know to implement improvement programs effectively. A subset

and techniques of Six Sigma.

data collection, industry is starting to accumulate a large database on quality

improvement programs. Quality/cost improvement project data was collected over a

30 month period at a U.S. based mid-west manufacturing firm. A study was

making applied to improvement projects. Three levels of decision-making are defined

Micro – related to the use of individual statistical methods.

Meso – supervisor level decision-making about method selection and timing.

Macro – dealing with overall quality programs and stock performance.

1.2 Plan of Study

• Chapter 2 is an overview and history of Six Sigma. It describes how

financial accountability for each project is established by the literature

as a key component of Six Sigma philosophy and methods. As a result,

databases about the financial successes or failures of individual projects

have been generated in many companies for the first time.

1990 through 2003. This chapter presents a classification scheme for

the articles, trends, topics covered, contributions, and tabulates

identified success factors for Six Sigma. A definition of Six Sigma is

given based on a synthesis of the literature.

• Chapter 4 examines case based instruction in classes on SPC and DOE

at Ohio State University. The approach is illustrated with an exercise

based on an actual case study of increasing yields for manufacturing

• Chapter 5 describes methods for mining the Six Sigma project