Some Results on a Wilcoxon--Mann--Whitney Approach to Interactions in a Two-Way ANOVA Design.

Abstract

There are now several ways of characterizing an interaction in a two-way ANOVA design. For four independent random variables  (let  and . One approach is based on , which represents a simple generalization of the Wilcoxon—Mann—Whitney method. Recently, two methods for making inferences about  were derived. One goal in this paper is to report simulation results indicating that both methods can be unsatisfactory when there is heteroscedasticity. The main goal is to describe an alternative approach that performs much better in simulations.