Prescriptions for working statisticians (Q4040326)

This book is both, an advanced course for students of applied statistics, and a ready handbook for the practioner who will fit functional linear models to data of the real world. Very often the preconditions such as normality, homoscedasticity, and independence of data in regression are not fulfilled, or the data set may be spoiled by outliers. Chapters 1-4 offer `diagnostics' of these delicate problems, e.g. for testing of normality nine different procedures are offered, and their power and usefulness are discussed; here I missed a hint to \textit{O. Vasicek}'s [J. R. Stat. Soc., Ser. B 38, 54-59 (1976; Zbl 0331.62031)] ``test for normality based on entropy'', which is powerful against some special hypotheses. Chapt. 5 is devoted to transformations of data in order to fulfill some preconditions of the tests, and Chapt. 6 to the question how many independent variables to put into a regression model and to the question of multicollinearity. Chapt. 7 is devoted to mixed categorical and measured data and the usefulness of dummy variables. Chapt. 8 gives inferences from cross-classified data. Each chapter is dichotomized in prescriptions (sometimes in cookbookish form) and in a theoretical part (mostly without proofs), necessary to understand the complicated and nested situations of practical data and to avoid misuse. 35 statistical tables of critical values and significance points, which are scattered in the literature, are added. A very useful book for the practionerö