Chapter 9: Robust Step-Down Tests for Multivariate Group Differences

Multivariate analysis of variance (MANOVA) is probably the most well-known technique for testing group differences on several response variables. However, a significant omnibus test is usually only the first step in the analysis; researchers typically wish to determine which dependent variable(s) is (are) responsible for rejection of the global hypothesis. One common approach is to conduct a univariate test on each dependent variable, applying a Bonferroni correction to control the overall rate of Type I errors to the nominal level of significance (Keselman, Wilcox, & Lix, 1998). However this approach does not account for the correlation between the measurements on successive dependent variables and can therefore result in low power to detect group differences.