Comparison Between Treatments in a Repeated‐measurement Design: Anova and Multivariate Methods

This paper investigates several methods available for testing comparisons between treatments in a repeated‐measurement design, within the framework of the general multinormal model. Various ratios are considered, and the weakest assumptions required for the validity of each F ratio are presented. When several F ratios are valid, power considerations are introduced for purposes of choice. Methods enabling the testing of validity assumptions are also presented, together with alternative multivariate statistics which can be used when no F ratio is valid. All procedures presented in the paper are illustrated by a numerical example.