CHOOSING A TEST STATISTIC IN MULTIVARIATE-ANALYSIS OF VARIANCE - COMMENT

Questions C. L. Olson's (see record 1976-25211-001) claim that the Pillai-Bartlett statistic (V) is superior to the Wilks λ (W) and the Hotelling-Lawley trace (T) for general use in multivariate ANOVA because of much greater robustness against unequal covariance matrices. It is shown by a sampling of studies from the literature that the example Olson used to demonstrate superiority had extreme subgroup variance differences, which occur very infrequently. For subgroup variance differences much more likely to occur, it is shown that the actual Type I error rates for V, T, and W are very similar. (12 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved).