Linear and generalized linear mixed models and their applications. (Q877444)

This book is an up to date description of linear mixed models, LMM, and generalized linear mixed models, GLMM. The material is complete enough to cover a course in a Ph.D. program in statistics. The contribution of this book is that of pointing and developing the inference and estimation issues for non-Gaussian LMMs. Some recently developed methods, such as mixed models diagnostics, mixed models selection and jackknife methods in the context of mixed models are also presented. The book is divided into four chapters, the first two being devoted to LMMs and the last two to GLMMs. Every chapter is sustained by a final part of exercises and real life data examples. The chapters are also completed with further results and technical notes. Moreover, two appendices on statistics and matrix algebra, and references based on both classical and recent works, are provided. For the first chapter, some notes on computation and software are given. The first chapter of this book presents in the framework of LMMs some introductory examples, a classification of Gaussian linear or non-Gaussian approaches, methods of estimation in Gaussian linear mixed models (maximum likelihood, ML, restricted ML). Also, analysis of variance estimation and minimum norm quadratic unbiased estimation are presented. If in the first part, devoted to LMMs, the author deals with point estimation, in the second part (chapter two), a different type of estimation is involved. Therefore, tests and confidence intervals are derived for both Gaussian LMMs and non-Gaussian LMMs, and matters like prediction, model checking, selection and Bayesian inference are also discussed. In the last part (chapters three and four), the author deals with linear mixed models, starting from the observation that the LMMs from the first part of the book can be used in situations where the observations are continuous, and not from discrete or categorical data. The generalized linear models are first presented as models that include normal, binomial Poisson and multinomial as special cases and also as models with the mean of observations associated with a linear function of some covariates through a link function and with non-constant variance of the observations. The author considers the extension to GLMMs as a solution for the case where the responses are both correlated and, at the same time, discrete or categorical data. Some aspects regarding the likelihood function under GLMMs, focusing on developing Monte Carlo based computational methods, are treated. On the other side, regarding a non-likelihood based inference approach, approximate inference and estimating equations are discussed as alternative methods to avoid the computational difficulties. Moreover, some new techniques on selection model issues are presented.