| Scope and method of study. The purpose of this study was to simulate conditions of dependence of observations for the Pillai-Bartlett, Coombs-Algina, and Johansen multivariate analysis of variance procedures. One thousand runs each of 960 different conditions involving different values for the number of groups, number of dependent variables, treatment group size, effect size, type of noncentrality, size of subgroups and degree of correlation within subgroups were each replicated five times. Rejection rates for each test and each condition were analyzed using a split-plot ANOVA.;Findings and conclusions. There were seven statistically significant five-way interactions between the variables analyzed, and many other significant lower-order effects subsumed by them, including all eight main effects. As expected, increases in the correlation of elements within subgroups, the size of subgroups, the number of groups, the number of dependent variables, and the effect size increased rejection rates. Differences between the statistical tests were statistically significant but were not deemed practically significant. Increases to the treatment group size decreased rejection rates. Differences in noncentrality structure were statistically significant but had little practical effect. |
