Nonlinear Regression (Wiley Series in Probability and Statistics) (Paperback)
By George A. F. Seber, C. J. Wild
Paperback: 792 pages
Publisher: Wiley-Interscience (September 19, 2003)
Language: English
ISBN-10: 0471471356
ISBN-13: 978-0471471356
Book Description:
One of the three most excellent books on nonlinear regression.
Chapters 1 and 2 provide good introductory material similar to the other texts. Chapter 1 deals with the models (linear and nonlinear) and Chapter 2 provides the basic estimation techniques. In addition to the standard material on least squares, generalized least squares and maximum likelihood, the authors also cover quasi-likelihood, linear approximations, robust estimation and Bayesian methods. Box - Cox transformations and the issue of variance heterogeneity are also treated in Chapter 2.
As they remark in the preface, they avoid much of the econometric theory and asymptotic theory that is well covered in Gallant's book.
Chapter 3 deals with important practical issues including the convergence properties of the iterative procedures (important for nonlinear models but a non-issue in linear models), ill-conditioning and identifiability (important issues for both linear and nonlinear models).
Chapter 4 deals with curvature issues and covers much of the original work of Bates and Watts with many references to those authors. Oddly though, there is no mention of the Bates and Watts text. Both books were published by Wiley around the same time with Bates and Watts appearing in 1988 and Seber and Wild in 198[local]2[/local]9. Perhaps the Seber and Wild book went to the publisher before the Bates and Watts book came out (their preface has a May 1988 date).
Important and interesting topics covered in this book but not the others include models with time dependent errors, detailed treatment of growth models, compartmental models, multiphase and spline regresions and error-in-variables models. They also devote a whole chapter to software issues (very interesting and practical but probably mostly outdated).
Good for a graduate statistics course or for a research reference source. Has lots of material and references but lacks homework problems.