Book Review

Abstract

New York -London 1982. VIII, 229 pp., e 12,-Residuals are uaed in many procedures designed to detect various types of disagreement between data and an eesumed model. This book is about diagnostics. The major emphaaia is on diagnostic tools for data analyses based on linear models in combination with least squarea methods of egtimation.I n chapter 2 diagnostic methoda using residuala are discussed. Especially the different tpyea of residuals which are important for diagnostica are defined and propertiee of these are proved.Besides these residuals the nuisance of several plota is stated. Also different transformations for the data are inveetigated. All these methods are useful for finding general inadequacies in a model. In chapter 3 the eseeeement of influence is considered. This influence analysis is discussed in details.Here the correspondence to the robustnees theory is to me. In chapter 4 aome alternative approaches to influence are considered. It is looked a t several methode that do not depend directly on the influence curve, but to use cme deletion pertubation schemes. It is considered a method based on the volume of confidence ellipsoids, another is due to ANDBEWB and PBEQIBON and this can be thought of aa a general omnibus diagnostic. The third discussed method beaee on the compari8dn of the predictive diatributions of future observations. In the chapter 6 it is diecutwed how diagnostic for linear least squarea regression might be extended to more complex situations. Here general definitione of reaiduals, a general approach to influence are given and disoueeed.This book gives a comprehensive course in diagnostics of regression. The book is written very clear, many examples help for understanding.H. Liiuter