The conventional statistical approached to Analysis of Variance or ANOVA have been around for years — before the 1960s, and must research has been done on the accuracy of such methods. Recently, a number of researchers have been publishing work on the inefficiencies of ANOVA, and the advantages of what is known as the robust statistical methods. R. R. Wilcox, in the second edition of his text, Introduction to Robust Estimation and Hypothesis Testing, focuses on the new development in the field of robust statistical modeling, and demonstrates how such methods have become practical in light of advancements in the field of computer science and computing processing power in the recent years.
The author walks the reader thru a series of examples and practical methodologies that address the common problems faced today with ANOVA and regression. Technical details are kept to a minimum, with only a couple of “heavy” math chapters, and most of the rest of the book is dedicated to demonstrating the practicality of Robust Estimation when using R or S-PLUS packages. The author has created a number of add-in commands for current R or S-PLUS users, and shows the reader how to use these commands in practice. The package is obviously free to download from the book's web site.
For readers with little statistical knowledge, there are a couple of background chapters that bring you up to speed with statistical modeling including ANONA and its shortcomings. A thorough introduction to Robust Estimation sets the reader on a path to be followed throughout the rest of the text. The author starts with a simple case for Robust Estimation, the One-Sample Case; he goes into detail explaining the flaws in the ANOVA process, and introduces, thru examples, the commands and tools that he has created to use the Robust Methodology. The topics continue with a more complicated case of Two-Groups, and so on.
The problem has been, and to some extent still is, the difficulty of simulating and executing Robust Estimation methods due to the lack of computing power. ANOVA methods, including regression is less accurate and has been documented to fail under certain circumstances, but it takes less computing power. Robust methods require more computing power, but as the author demonstrate throughout the text, are more accurate. A number of R and S-PLUS add-in commands are simulated using medium to large datasets with acceptable computation times and better-than-ANOVA results.
Rand R. Wilcox does a great job at demonstrating the power and the applicability of Robust Estimation. He assumes little or no prior knowledge of the approach and approaches the topic with application and usability in-mind. Only a couple of chapters are dedicated to theory, and can be skipped or skimmed thru by practitioners. Most of the book is dedicated to examples and how-tos that, with the combination of the R/S-PLUS toolkit provided by the author, can become a valuable part of your library.
Art Sedighi received his B.S. degree in Electrical Engineering and will receive his M.S. Degree in Computer Science from Rensselaer in 1998 and 2004 respectively, and is currently pursuing his MS in Bioinformatics from Johns Hopkins University. He is a solutions architect with Platform Computing where he architects enterprise-wide grid solutions for fortune 500 firms in the financial, pharmaceutical and telecommunication industries. His research interests include Grid Computing, Software Engineering and Bioinformatics.