GLOBAL MEASURES OF ASSOCIATION IN MULTIVARIATE ANALYSIS OF VARIANCE

Four possible global measures of association for estimating strength of association in MANOVA are presented, One of them, Wilks's generalization of eta squared, is rejected because of undesirable properties. Of the other three, one in- volves Wilks's Lamda, with the remaining two involving the trace statistic V. It is now that an unbiased estimate can be obtained for V, and hence that the global measure involving this unbiased estimate might be the best. Finally, an algebraic demonstration showed that the estimates of association obtained from the three measures will generally differ by only a few percent or less.