Efron's Bootstrap with Application to the Repeated Measures Design

Abstract

The traditional univariate ANOVA for the repeated measures or split-plot design, commonly used in the behavioral sciences, requires, in addition to the usual assumptions of error normality and variance homogeneity, that the covariance matrix for the repeated measures have a special form (Type H). Because detection of lack of compliance with these assumptions is problematic, this design is a good candidate for alternative analysis. This paper illustrates an application of Efron's bootstrap to the repeated measures design. While the bootstrap approach does not require parametric assumptions, it does utilize distributional information in the sample. By appropriately resampling from the data collected in a study, the bootstrap may determine quite accurate sampling distributions for estimators, effects, or contrasts of interest.