Sample size estimation for randomized pre-post designs

Abstract

Empirical formulae are presented for estimation of sample size in repeated measurement designs where analysis of gain scores or analysis of covariance (ANCOVA) is to be used to control for individual differences. Sample sizes required to achieve desired power are smaller for ANCOVA than for analysis of simple gain scores, and in both cases they depend on the magnitude of the pre-post correlation. An alternative approach to estimation of sample size which obviates the necessity for prior estimates of expected mean difference, experimental error, and pre-post correlation is presented.