This paper examines three statistical analysis procedures appropriate to repeated measures data -- classical mixed model analysis of variance, multivariate analysis of repeated measures, and analysis of covariance structures. These procedures differ in their assumptions concerning the covariances between the latent random variables in the model underlying the repeated measures. Simulated data were employed to investigate the effect that violating the assumptions of each model had on the following: Type I error rates; power of the test; degree of bias in the parameter estimates; and the relative efficiency of the estimates. The data indicate that with respect to Type I error rates and estimates of repeated measures effects, violations of assumptions for all three procedures produce similar results. Depending on the size of the repeated measures effects, there are differences between procedures concerning power. The estimates produced by analysis of covariance structures techniques tend to have smaller standard errors.
