Statistics for Experimenters: Design, Innovation and Discovery

Is it really possible to update a well-known, classic textbook and improve it? Yes, it is not only possible but it has been done. This second edition has retained all of the charm and creativity of the first edition while adding new material as well as improving upon the presentation of some of the original material. This introductory textbook continues to teach the philosophy of design and analysis of experiments as well as the “nuts and bolts” in a way that is accessible to both students and industrial practitioners. The reader generally finds clear and well-motivated examples, excellent discussions of underlying statistical concepts and practical guidelines for experimentation. Students and practitioners alike will find many stimulating problems at the end of each chapter as well as more routine exercises dispersed throughout the chapters. The collected wisdom of many experts in applied statistics is contained in the numerous quotations printed on the front and back covers of the book. I found these quotes to be valuable, entertaining, and thought provoking; indeed I will certainly insist that my students read and understand them in all my future statistics courses. The book gets off to an excellent start in Chapter 1 with a wonderful discussion of the learning process and in particular of the iterative inductive–deductive process. Chapter 2 is a condensed treatment of the statistical basics which I believe is meant more as a review rather than a first exposure. Chapter 3 introduces many of the usual inference topics such as significance tests for comparing means, tests of association for contingency tables, and confidence intervals. However, unlike many traditional textbooks, topics such as blocking, randomization, and reference distributions are treated with appropriate emphasis. In particular, the authors indicate how the use of an external reference distribution eliminates the need for the errors to be statistically independent—a topic which frequently occurs in practice but is rarely treated in textbooks. Completely randomized designs, randomized block designs, and Latin square designs are covered in Chapter 4 as well as balanced incomplete block designs and a preliminary mention of split plot experiments. Chapters 5–8 deal with the popular two-level factorial designs and present the material in a clear and comprehensive manner through the use of several well chosen examples. Many of the familiar examples, illustrations, and tables from the original edition are still used in this edition, but these have been enhanced to include some more up-to-date developments, such as Lenth’s method and Bayes plots for identifying important factors. Indeed, the famous and oftreferred-to table of two-level fractional factorial designs (p. 410 of the original edition) has been updated to reflect the minimal aberration nature of these designs and now appears on page 272. Chapter 9 covers multiple sources of variation in general by furthering the earlier discussion of split plot experiments and introducing the topics of variance component estimation and transmission of error. Least squares estimation and nonlinear models are the main topics in Chapter 10, while response surface methods are discussed in Chapters 11 and 12. The remaining three chapters introduce some methods related to quality improvement and include introductions to robust process, parameter and tolerance design, process control, forecasting, time series, and evolutionary process operation. While my overall impression of this outstanding book is extremely positive, I feel obligated to identify some negative aspects. There is the usual number of typographical errors in a new textbook, and the sentence structure, while grammatically correct, is not always straightforward. Some of the problems at the ends of the chapters seem to rely on information not provided to the reader. But perhaps the authors deliberately made these problems open-ended in order to be thought provoking and realistic. The subject index in the back of the book appears to be somewhat incomplete since it does not include topics such as split plot, stationary, and posterior, all of which appear in the book. There are also a few topics which are missing that I think might have been included, such as generalized resolution, analysis of covariance and mixture experiments, and some other topics which should have been developed more, such as minimum aberration and design optimality. Finally, there is a distinct absence of the use of any computer software packages, especially JMP®, Minitab®, and DesignExpert® which are particularly adept at designing and analyzing experiments. Modern computer software packages are comprehensive, flexible, easy to use and are capable of producing excellent information, output and graphics. In my