The first, written by an Australian and Canadian team, John Maindonald and John Braun, is probably the more “user-friendly” of the two. There is a comprehensive introduction, a very useful chapter-by-chapter summary, and 12 chapters, supported by an appendix listing S-Plus differences, references, indices of R symbols, functions, and terms. Each chapter has an extensive reading list graded as elementary, basic, and advanced/comprehensive, and there are several end-of-chapter exercises (thoroughly developed solutions are provided in the r-book web site). Many chapters also have a Recap section containing brisk advice—“do … , ensure … , always … , think…”.

The first 2 chapters introduce the Elements of R (these are supplemented by more advanced topics in chapter12) and of Exploratory Data Analysis. The authors make the point that the best modern statistical software combine data analysis with graphics; indeed, that is one of the strengths of the R language. The next 2 chapters examine the concepts of Statistical Modeling and Inference. Again, the end-of-chapter exercises ensure that these concepts are firmly grasped.

The chapter on Regression with a Single Predictor—the most frequently used regression technique—warns the reader that the correlation coefficient (*r*) is a crude and unduly simplistic summary measure, as is *r*^{2} (the explanatory power of the model). The measures that most satisfactorily ensure that the chosen model correctly reflects reality require determination of residuals (the difference between the actual and fitted values of the response variable) and their distribution. R provides 4 diagnostic plots that can be used to detect data items that require further investigation. The techniques outlined in this chapter are then used to develop methods for validating the appropriate statistical model in the chapters on Multiple Linear Regression, Exploiting the Linear Model Framework, and Logistic Regression and Other Generalized Linear Models.

The remaining chapters cover topics such as multilevel models, time series, tree-based classification and regression (usually abbreviated to CART and a very effective diagnostic technique—see, for example, *N Engl J Med* 1982;307:588–96), and multivariate data exploration and discrimination that provide considerable insight into very powerful statistical procedures. The r-book site contains corrections for the numerous printings of the book (these suggest consumer appreciation), changes affecting discussions and code (attributable to newer versions of R), R codes for graphing and calculating, many additional notes, information on the forthcoming second edition, and as noted earlier, solutions for many of the book’s exercises. Access to the DAAG site is unnecessary for newer versions of R, as the data sets and code are now available within the R site itself.

- © 2006 The American Association for Clinical Chemistry