Applied regression analysis. A research tool.

Introductory exposition (textbooks, tutorial papers, etc.) pertaining to statistics (62-01) Research exposition (monographs, survey articles) pertaining to statistics (62-02) General nonlinear regression (62J02) Linear regression; mixed models (62J05) Diagnostics, and linear inference and regression (62J20) Linear inference, regression (62Jxx)

Review text

[For the review of the first edition from 1988 see Zbl 0672.62076.] This book on regression analysis is aimed at graduate students in various scientific disciplines with previous exposure to basic statistical methods. The goal of the authors is to provide an explanation of least squares and related statistical methods by emphasizing regression concepts rather than mathematical proofs. However, proofs are given to help enhance the understanding of essential concepts and to develop facility with matrix algebra. The book consists of nineteen chapters. Essential statistical tables are provided in an appendix along with a list of references, an author index, and a subject index. Chapter 1 is a quick overview of elementary regression results for a single variable linear model although models with several independent variables are also introduced. The least squares criterion is introduced here and a discussion of precision of estimates is provided. Chapter 2 introduces the matrix notation and standard matrix operations. A discussion of eigenvalues, eigenvectors, and the singular value decomposition is included. Chapter 3 presents the multiple regression model in matrix notation. Normal equations and their solutions are developed. Matrix expressions are provided for the variance of linear parametric functions. Chapter 4 considers quadratic functions of a normal random vector \({\mathbf Y}\) and develops important distributional results. These are then used to discuss tests of hypotheses, confidence intervals and joint confidence regions. Chapter 5 demonstrates the application of multiple regression methods to a real data set involving five independent variables. Subset selection ideas are introduced in the context of this example. Chapter 6 is devoted to a discussion of the geometry of least squares. Chapter 7 provides a treatment of model development and the variable selection problem. Methods discussed include all possible regressions, best subsets regression, and stepwise regression including forward selection, backward elimination. Subset selection criteria such as \(R^2\), residual mean square, \(R_{adj}^2\), Mallows' \(C_p\) statistic, AIC and SBC, are explained. Model validation considerations are discussed. Chapter 8 is devoted to polynomial regression, polynomials in several variables, response surfaces, and trigonometric regression. The technique of dummy variables to accommodate categorical or class variables in multiple regression is introduced in Chapter 9. Detailed discussion of various parameterizations associated with the dummy variable technique is given and their interrelationships are discussed. Generalized inverses of matrices are introduced and used to solve normal equations for overparameterized models. The chapter concludes with a discussion of analysis of covariance. Problems that arise in least squares applications are considered in Chapter 10. These include nonnormality, variance heterogeneity, correlated errors, outliers, influential observations, multicollinearity, and model inadequacy. Detection of these problems and corresponding remedial measures are discussed in the following three chapters. Chapter 11 presents methods for detecting violations of regression assumptions. Graphical and numerical methods are described for detecting unusual observations and model inadequacies. Standardized residuals, studentized residuals, recursive residuals, Cook's distance, DFFITS and DFBETAS, and variance inflation factors are some of the numerical measures defined and discussed. Chapter 12 considers the use of transformations of dependent and independent variables to simplify relationships and/or to rectify model violation problems. Weighted least squares and generalized least squares are presented as candidate methods for accommodating heterogeneous variances. Chapter 13 is entirely devoted to a discussion of multicollinearity. Techniques are provided for understanding the relationships among the predictor variables that give rise to the multicollinearity. Principal components analysis and Gabriel's biplots are used to understand the correlated structure of the independent variables. Principal components regression is presented and its limitations are discussed. The techniques are illustrated via a case study example in Chapter 14. Nonlinear regression is the topic of Chapter 15. Many useful nonlinear models are introduced and their applications are discussed. These include exponential growth or exponential decay models, Mitscherlich's growth model, Monomolecular growth model, Inverse polynomial model, Logistic growth model, Gompertz growth model, etc. Segmented regression is also discussed. Fitting procedures for nonlinear least squares problems are explained. Logistic regression is discussed very briefly. Chapter 16 is devoted to case study examples illustrating the techniques developed thus far. Chapter 17 presents methods for analyzing unbalanced data problems. The ``cell mean model is introduced and estimable functions are defined. Various types of sums of squares (type I, type II, type III, and type IV of PROC GLM in the statistical package SAS) are defined and explained. Random and mixed effects models are the topic of Chapter 18. The general mixed linear model is motivated by first considering the one-way random effects model, the split plot model, and the random coefficient regression model. The final chapter is devoted to a case study illustrating issues which arise when analyzing unbalanced data under a model involving more than one random effect. The use of the software package SAS to perform the needed calculations is explained as needed in many of the chapters. There are plenty of problems at the end of each chapter for the reader to test their understanding of the chapter material. This book will serve as a useful reference for students of statistics and practitioners alike. It may also be used as a textbook in an advanced undergraduate or beginning graduate level regression variance course.

Recommendations

Cited in

(53)

Describes a project that uses

Uses Software

This page was built for publication: Applied regression analysis. A research tool.

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4391454)