COMBINING INDEPENDENT TESTS IN MULTIVARIATE LINEAR-MODELS

A class of independent multivariate linear models is considered, havin g a common parameter matrix Theta in their means, but having different covariance matrices. For testing H-0:Theta=0, some test procedures ar e derived, which combine the information from the different models. In the context of the interblock analysis of block designs, simulated po wers of some combined tests are reported in the bivariate case, for co mbining the intra- and interblock tests based on the Wilk's Lambda cri terion. The numerical results indicate that the combined tests offer s ubstantial improvement in power compared to the Wilk's Lambda test bas ed only on intrablock information. (C) 1994 Academic Press, Inc.