Limiting follow-up hypotheses to be tested can reduce problems relating to the control of Type I and Type II errors in multivariate analysis of variance (MANOVA). Such limitations can also improve the interpretability of results. The importance of sample size, shape of population distribution, within-group correlations and heterogeneity of variances are demonstrated. The protected greatest characteristic root (GCR) procedure is shown to work well for small, group size, N (≤ 10). The unprotected GCR is shown to work well for larger N.
Ramsey, Philip H.; Ramsey, Patricia P.; Hachimine, Priscila; and Andiloro, Nancy
"Robustness, Power and Interpretability of Pairwise Tests of Discriminant Functions in MANOVA,"
Journal of Modern Applied Statistical Methods: Vol. 10
, Article 2.
Available at: http://digitalcommons.wayne.edu/jmasm/vol10/iss2/2