A Comparative Study of Power and Sample Size Calculations for Multivariate General Linear Models

Abstract

Repeated measures and longitudinal studies arise often in social and behavioral science research. During the planning stage of such studies, the calculations of sample size are of particular interest to the investigators and should be an integral part of the research projects. In this article, we consider the power and sample size calculations for normal outcomes within the framework of multivariate general linear models that represent the most fundamental method for the analysis of repeated measures and longitudinal data. Direct extensions of the existing generalized estimating equation and likelihood-based approaches are presented. The major feature of the proposed modification is the accommodation of both fixed and random models. A child development example is provided to illustrate the usefulness of the methods. The adequacies of the sample size formulas are evaluated through Monte Carlo simulation study.