Statistical Methods for Rates and Proportions (3rd. ed.), by Joseph L. FLEISS, Bruce LEVIN, and Myunghee Cho PAIK, Hoboken, NJ: Wiley, 2003, ISBN 0-471-52629-0, xxvii + 760 pp., $94.95.

I bought the second edition (2E) of this book, which was published in 1981, as a reference book for data analysis. The book had its genesis in 1972. Here it is back in a third edition (3E) that is more than twice the size of the 2E. Two new authors had to be added to accomplish this task, because the original author had already been disabled by an illness that later claimed his life. Whatever his relationship with his two co-authors, I am sure he would be very proud of this impressive new edition.

Much of the methodology that most of us would utilize for handling rates and proportions is computer-intensive and new since 1971. Recognizing the pedantic value of the simplicity of the first two editions, the new authors have chosen to retain much of the content of the 2E's 14 chapters within the 3E's 19 chapters. Because much of the modern methodology, including logistic regression and Poisson regression, generally requires more background than algebra for full comprehension (the only presumption for the 2E), the authors have put the more advanced methods in "*" sections to help retain the original audience. For full appreciation of the text, the authors recommend (Preface, p. xviii) "mathematical preparation equivalent to a first and second course in statistics."

Following the introductory chapter, a new Chapter 2, "Statistical Inference for a Single Proportion," establishes necessary concepts, including exact inference and several alternatives for two-sided confidence intervals. Chapters 3-10 are all holdovers from the 2E. Much of the organization remains the same in Chapters 3-5. The topics covered include significance in a fourfold table, sample sizes, and randomization. Chapter 6, on comparative studies, has a couple of new sections on exact methods. Chapters 7 and 8, on sampling for comparative studies and randomized controlled trials, also retain their original structure. Chapter 9, covering the comparison of proportions from several samples, has new sections on logit models for ordered outcomes and on the effect of randomness, in which Bayesian methods are discussed. Chapter 10, on combining evidence from fourfold tables, has new material on meta-analysis and large-sparse data structures.

Much of the remainder of the book is new. Chapter 11 explores logistic regression and includes polytomous logistic regression. Chapter 12 covers Poisson regression, including a discussion of overdispersion. Chapter 13, on the analysis of data from matched samples, was formerly Chapter 8 in the 2E. It is followed by a chapter on regression methods for matched samples, which includes parametric modeling and conditional logistic regression.

Chapters 15 and 16 are also new. Chapter 15, "Analysis of Correlated Binary Data," provides methods for longitudinal data, and Chapter 16, "Missing Data," considers situations in which the data are incomplete. Chapters 17-19, on misclassification, measurement of interrater agreement, and standardization of rates, are assembled from the last four chapters of the 2E. Each chapter contains some new material.

Upgrading a book originally based on a desk calculator to a book that is appropriate for the modern computing environment obviously would be a challenge for any new authors. The authors have not attempted to complete the entire task. There is no direct application of statistical software, such as SAS or S-PLUS, in this book. Nevertheless, anyone who still uses the 2E would surely want a copy of this new edition. Likewise, persons who regularly encounter this type of data would certainly want this book available as one of their desktop references.