Introduction

This chapter concentrates on models where the errors have Gaussian distributions. There has been much work in recent years on longitudinal data with non-Gaussian distributions. It also presents the details of maximum likelihood estimation in the balanced compound symmetry case. The mathematical manipulations of matrices that are necessary to obtain likelihoods for longitudinal data problems are developed. Classical longitudinal data analysis is based on balanced designs where every subject is measured at the same time points with no missing observations. If the number of subjects is large relative to the number of observations per subject, multivariate analysis methods which assume a general covariance structure for the observations taken on the same subject can be used. In a typical medical application, subjects are assigned randomly to two or more treatment groups, one of which may be a placebo or control group. The chapter also presents an overview of this book.