Sample sizes in the multivariate analysis of repeated measurements

Abstract

Determination of sample sizes for comparing two or more treatments in repeated measurements experiments is considered. Multivariate normality of the individual's vector of repeated measures is assumed. Particular emphasis is placed on applications wherein the error variance-covariance matrix is arbitrary positive-definite. Sample size determination is based on power considerations associated with Hotelling's T2 test and the desire to detect a specified difference between any pair of treatment means. Tabulated sample sizes are given in the case of an equal variance-unequal covariance structure. The utility of these sample sizes is also demonstrated for a more general variance-covariance structure. Applications of these sample sizes are illustrated with two examples.