Restricting the Covariance Structure

Since the variances of the two measurement periods are nearly equal, a covariance structure that assumes the variances are equal may work as well as the unstructured covariance. By assuming the variances to be equal, the model would use one less covariance parameter, and in general you want to use the simplest model that fits the data well. Also note that the test of the fixed effect [TIME=1] in such a model, like that for the unstructured covariance model, is equivalent to a paired t-test.

To see if restricting the variances to be equal across the measurement periods makes sense, you will fit a model with a compound symmetry covariance structure and compare it to the unstructured covariance model. The compound symmetry structure requires constant variation and constant covariation. In this case, since there are only two time periods, the assumption of equality of covariances is trivial.

Next