Multiple Comparisons for Mixed Models: Concepts, Theoretical Reasons, and Case Studies on Longitudinal and Repeated Measures Data Analysis

The assumptions of constant variance and uncorrelated errors that are made in general linear models often do not hold for longitudinal data, repeated measures, and general mixed models. The traditional multiple comparison (MCP) methods based on the studentized range distribution and multivariate t distribution are exact only under the assumptions of linearity, normality, constant variance, and uncorrelated error. The element of MCPs is often to compare the means of two measures. For the two-sample t test, even a moderate correlation to observations will result in serious bias; this problem definitely carries over to multiplicity-adjusted inferences. The book briefly reviews the background of traditional procedures of MCPs then investigates some MCPs with longitudinal repeated measures. In these cases the sample errors are inflated/deflated because of the correlation among observations, and results in giving incorrect inferences. In this book, we suggest using different ways of dealing with this situation under various mixed models. This book includes full of case studies and gives detailed step by step theoretical explanations.