Parametric and bootstrap approaches to repeated measures designs

Abstract

Recently, a nonparametric technique called bootstrapping has been recommended over the more well-known analysis of variance (ANOVA) for analyzing repeated measures data. Advocates cite as bootstrap's advantages over ANOVA the fact that the former uses distributional information and is free of normal theory assumptions. The present study used a computer simulation to compare the two techniques calculated using data sampled from normal and nonnormal distributions. The parametric test had adequate control of Type I error rates; the nonparametric test had overly liberal Type I error rates and therefore is not recommended. © 1987 Psychonomic Society, Inc.