摘要：In this article we examined the performance of a mixed model Kenward-Roger’s adjusted F-test based on the correct covariance structure and the multivariate extension of the modified Brown-Forsythe method in a mixed repeated measures design. These two procedures were compared with regard to their power and robustness when multisample sphericity and multivariate normality assumptions are violated separately and jointly. Monte Carlo comparison shows that, overall, both methods do a reasonable job of controlling the rates of error for both normal data, as well as certain types of non-normal data. With respect to power, the results indicate that the mixed model analyses using the true structure model was generally more powerful than the modified Brown-Forsythe procedure.
