Handbook of Statistics 15: Robust Inference

Regression analysis is perhaps the most widely used of all advanced statistical techniques. The authors of this book, Dennis Cook and Sanford Weisberg, have played a major role in the developments that have taken place in regression analysis over the past 20-25 years. A partial listing of their work is given in the references at the end of this review. Regression diagnostics was the featured topic in the early work of Cook and Weisberg. Cook (1977) introduced Cook's distance for measuring the influence of a single observation in least squares regression. Shortly after that, Cook and Weisberg (1980) extended the concept to multiple observations. A flood of alternative influence measures developed by other researchers soon followed. The book Residuals and Influence in Regression (Cook and Weisberg 1982) summarized the developments up to that time in the area of detecting unusual data and measuring their influence on regression analysis. The second edition of Weisberg's (1985) popular applied regression textbook appeared soon afterward. The following year, Cook (1986) published his first article on local influence, a new paradigm for measuring the influence of one or more observations in regression analysis. In the past decade, the Cook and Weisberg research effort has focused on the use of regression graphics for constructing regression models and detecting deficiencies in both regression models and data. A series of articles on regression graphics (Cook and Weisberg 1989; Cook 1993, 1994; Cook and Weisberg 1997) eventually led to two books on the subject, An Introduction to Regression Graphics (Cook and Weisberg 1994) and Regression Graphics: Ideas for Studying Regressions Through Graphics (Cook 1998). The latest book in the Cook and Weisberg anthology is Applied Regression Including Computing and Graphics. They have blended recent advances in modern graphical regression analysis from the past decade with traditional material on regression to produce an excellent, up-to-date treatment of regression analysis. Typically, new statistical methodology, such as that presented in Applied Regression, takes 10-20 years to make the transition from the research literature into textbooks intended for a wide audience. A major reason for this lag is the lack of available software or hardware for employing new statistical tools. The general situation is improving with statistical languages (such as S-PLUS and Lisp-Stat) and platform-independent programming languages (such as Java). In most cases, though, the user is still required to write his own code or accept the fact that available code for a new procedure or graphical tool will not run on his computer. Cook and Weisberg solved this problem with Arc, a user-friendly graphical regression software package based on Tierney's (1990) Lisp-Stat language that can be deployed under Windows, Macintosh, and Unix operating systems. The Arc software is a free download from their Web site. With the Arc software, the new graphical ideas and tools presented in Applied Regression are immediately accessible to the user. Numerous detailed examples in the use of Arc with real data are given throughout the text. Applied Regression is a "complete package"-a textbook with the latest approaches and tools for regression analysis, easy-to-use software that runs on Windows, Macintosh, and Unix operating systems, and a companion