Effect of Correlation on Power in Within-Subjects versus between-Subjects Designs

Abstract

It is well known that the two dependent-samples t test (and a within-subjects design in general) is typically more powerful than the two independent samples t test (a between-subjects design), and that this increase in power is in large part a function of the correlation between the two sets of scores (e.g., pre-test and post-test). This paper demonstrates relative power as a function of (1) the correlation between groups with effect size and sample size held constant and (2) sample size with the correlation between groups and effect size held constant.