Experiments: Planning, Analysis, and Parameter Design Optimization.(Book review)

Experiments: Planning, Analysis, and Parameter Design Optimization

C. F. Jeff Wu and Michael Hamada

John Wiley & Sons, Inc., New York, 2000, S125.00, ISBN: 0-471-25511-4

There are numerous books on the design of experiments. However, this book by Wu and Hamada clearly stands out from its peers. There are several features that make it a very special book on experiments. The following are a few of them.

First, the book is written by two eminent researchers in the field of design of experiments. They are authorities in the area and therefore the readers can use the methods described in the book with complete confidence. Second, the book gives a comprehensive account of modern experimental design methods, which I have not previously seen in any other book. The presentation of new and important methods makes this book very valuable for practitioners. Third, the book is a nice blend of theory and practice and is written in a style that makes it enjoyable to read. One pleasing aspect of the style is that each chapter starts with one or two real experiments which motivates the reader to read the rest of the chapter. The design and analysis methods are systematically developed and then demonstrated using the motivating examples. Fourth, there is so much depth and breadth in the book that it can be used as a textbook for undergraduate, Master's, or Ph.D. level courses.

The book is organized into 13 chapters. Chapter 1 starts with single-factor experiments. The use of linear regression models for the analysis of experiments is introduced in this chapter. Chapter 2 discusses experiments with more than one factor and covers classic design topics such as randomized block design, Latin squares, balanced incomplete block designs etc. Full factorial experiments at two levels are presented in Chapter 3. The discussion given on normal and half-normal plots is very interesting. The book explains why a half-normal plot is more useful in the detection of significant effects than the usual normal plots. The half-normal plots together with more formal tests such as Lenth's method are used throughout the book. The very important topic of fractional factorial experiments at two levels is discussed in Chapter 4. It introduces the criteria of maximum resolution and minimum aberration. The minimum aberration criterion has emerged as the single most important criterion for optimal design selection. Nevertheless, it is quite surprising that this book is one of the few, and the first applied text, on the market to discuss the minimum aberration criterion. Chapter 5 introduces experiments with three levels. The usual approach to analyzing three-level experiments is to use ANOVA. However, the book explains the limitations of this approach and proposes an alternative analysis strategy. Mixed two-, three-, and four-level designs are discussed in Chapter 6. I am not aware of any other book that discusses the analysis of experiments with four levels. Some of the coding schemes presented in the book seem to be new and original. Another interesting aspect of the book is the use of the effect heredity principle in the analysis. Through the use of such principles one can obtain simple and interpretable models. The chapter also includes the intriguing "sliding-level method" for experimenting with related factors. Chapter 7 discusses the construction of nonregular designs. Although these designs are initially recommended only for screening purposes, many recent developments show their advantages over regular designs. The analysis methods for such designs are presented in Chapter 8. This chapter also introduces some advanced Bayesian methods such as Gibbs sampling, which are very useful to analyze designed experiments with complex aliasing. Chapter 9 discusses the response-surface methodology. Although only one chapter is devoted to this important topic, it is nicely explained with some very interesting examples. Chapters 10 and 11 discusses some topics in robust-parameter design. This technique, introduced by Taguchi, has become the most important tool for quality improvement. These two chapters are more engineering oriented than the rest of the book. Chapter 11 discusses signal-response systems, which is also known as dynamic parameter design. The analysis methods for reliability improvement experiments are presented in Chapter 12. Chapter 13 discusses experiments with non-normal data and explains the use of generalized linear models to analyze such data.

Chapters 1-4, 9, and 10 can be used to teach a Master's-level course. A few topics from Chapters 5 and 8 can also be added to the syllabus. The same chapters but with less emphasis on the mathematics, can also be used to form a senior-level undergraduate course on the design of experiments. Topics from the other chapters can be selected depending on the interest and background of the students to teach a more advanced course. Teaching materials for an introductory course is made available through the author's website.

The book covers almost all of the useful methods in experimental design. A few topics such as split-plot designs, mixture designs, and computer experiments are not discussed which I hope the authors will consider adding to the book in future editions. I have no doubt in saying that this book will become a classic text on experimental design.

It will definitely play a major role in guiding research and development into experimental design for many years to come.

Reviewed by V. Roshan Joseph

School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA

E-mail: roshan@isye.gatech.edu

Contributed by the Book and Software Review Department

Print Get the Mag Weekly Updates