Alternative Methods of Regression Hardcover - 1993 - 1st Edition

This book provides an introduction to five of the most popular alternatives to least-squares regression. Most modern textbooks on regression give brief discussions of a few alternative methods, and specialized books exist on particular methods, but, until now, such a variety of methods have not been presented within a single book. In addition to a review of least-squares regression, Alternative Methods of Regression includes coverage of:

• Least-absolute-deviations regression
• Robust M-regression
• Nonparametric rank-based regression
• Bayesian regression
• Ridge regression

Each method has a chapter devoted to it, describing how regression estimates and tests are calculated and why the method makes sense. Each method is illustrated on a set of simple regression data and a set of multiple regression data. The format of the chapter, with description, justification, and illustration separated into subsections and with many details deferred to notes at the end of the chapter, is intended to allow flexibility of use. Depending on the reader's-current purpose, he or she could pick out the material relating to either the "how" or the "why" of the method. The notes provide additional material for a second or third reading. The descriptions and illustrations are given in sufficient detail to enable the reader to write computer programs implementing the methods. Near the end of each chapter, a section on computation includes a small test case for debugging such a program and also mentions existing programs and packages. Anyone who is interested in regression should learn about the alternatives to least squares. This book could serve as a textbook for a course following a least-squares regression course or for self-study. It is also a good reference for practitioners who may want to supplement a least-squares regression analysis with an alternative analysis.