Abstract
A major problem in conducting repeated measure analysis is the sphericity assumption. The consequences of applying ANOVA or MANOVA when this assumption is violated lead to inappropriate use of covariance structure. In the case of ANOVA, degree of freedom adjustment methods are usually used. While in multivariate methods, profile analysis (Repeated MANOVA) in General Linear mixed model is usually applied. Empirical Comparison of the covariance structure of both methods could provide useful information for researchers considering the application of either method. In this study data from weekly weight of albino rats were used to determine the most appropriate covariance structure for both univariate and multivariate repeated measure designs. The covariance structures used are Unstructured (UN), Unstructured correlation matrix (UNC), Compound Symmetry Correlation matrix (CSC), Diagonal (DI) and Factor Analytic first-order (FA (1)). The Goodness of fit criteria used to evaluate the performance of each covariance structures are the Burnham-Handerson (AICC) and Schwartz's Bayes (SBC). The data set used was found to violate the assumption of sphericity. The results from AICC and SBC criteria showed that UN is most appropriate covariance structure for the data set studied.

Key words: Repeated measure, Sphericity test, covariance structures, Goodness of fit criteria, univariate and multivariate analyses.