This book is a legacy of the late author Dr. Kenneth G. Russell (1950–2019), who passed away just months after its publication. Readers of this book will benefit from his career-long knowledge of the topic and passion for passing on the knowledge to others.
This is the first book focusing on the design of experiments (DOE) for generalized linear models (GLMs). Some of the existing textbooks, monographs, and handbooks on DOE dedicate a chapter on this topic, but none would match the depth and breadth covered in this book.
The intended readership of the book (as indicated in the Preface) is researchers who want to design experiments for collecting data to be analyzed by GLMs. These readers may be professional statisticians (e.g., consultants) or nonstatisticians (e.g., natural/social scientists, engineers). The content is well designed for the intended readership, especially those with minimal knowledge of calculus and matrix algebra. They can follow the examples in the book and accompanying R programs available on the companion website (https://doeforglm.com) to generate an efficient experiment design for their data collection.
Faculty may use this book as the main textbook or a complementary resource for either a semester-long advanced undergraduate course or, more suitably, an introductory graduate course. The course’s prerequisite should be a DOE course, which is offered in many academic institutions with a focus on general linear models such as ANOVA, ANCOVA, and regression analysis. Another prerequisite is at least elementary proficiency in the programming language R so the students can follow computing examples in the book. The book’s R programs are generally self-explanatory and thus useful for distribution to students. The author prioritized clear exposition over efficient computing (e.g., the use of a loop instead of a vectorized function). One missing piece in the book that would be useful for classroom instruction is exercise problems (and the solution manual for instructors). However, this lack of exercise problems would not hinder self-study learners because of the ample examples provided in the book.
This book’s content is well-structured for both self-study and classroom instruction. The book covers core background material in the first three chapters, including GLMs (Chapter 1), numerical optimization and integration minimally needed for optimizing a DOE for GLMs (Chapter 2), and the theory underlying the DOE for GLMs with a focus on the optimality (Chapter 3). The following three chapters focus on DOEs for data from specific distributions such as the binomial (Chapter 4), Poisson (Chapter 5), and other distributions (Chapter 6). The last chapter (Chapter 7) provides a gentle introduction to the Bayesian experimental design, a fast-developing area in the DOE literature. This chapter prepares readers for reading the relevant academic literature. It enables readers to generate Bayesian designs for GLMs using the book’s R programs or the existing R package acebayes. The author notes in the Preface that he intentionally omits some topics that regularly appear in other references. This choice makes the book waste no space for the material that the intended readership already knows or can easily find a reference for.
This book’s style is primarily expository. While the language is concise, the author spares no effort to illustrate important concepts and ensure they are understandable by the intended readership. Especially, the book’s examples are easy to follow (as if the author is speaking directly to you to walk you through each step). The author’s use of language is precise, minimizing the chance of confusing readers. Even though this book is the first edition, I found no typographical errors except for the few known errata (available on the book’s companion website). This book uses terminologies and notations that are standard in the academic literature and in practice (e.g., software documentation) so that readers can use the same language in their academic writing as well as in software implementation. Also, this book cites relevant references for those who want to learn more, instead of providing peripheral details that may hinder the readability. This approach serves well the intended readership who might solely rely on this book to understand the subject in a limited time without external guidance. Thus, I would strongly recommend anyone who collects data for GLMs not hesitate to get this book.
Department of Industrial & Systems Engineering, University of Washington
[email protected]