A diagnostic data and model checking methodology is developed and illustrated for validating the "circularity conditions" that are required for the univariate analysis of data from a repeated measures experiment. This condition mandates that for a t degree freedom family of comparisons among the repeated measures; all orthogonal contrasts contained in the t-dimensional vector space have identical variances. Just as the numerator of a within-subjects F ratio can be decomposed into orthogonal components, so can the denominator or "error term". Simply put, circularity is met when the orthogonal components (contrast variances) comprising the error term are homogeneous. It is shown how the recently developed AVE variance component diagnostic methodology can be successfully employed to detect potential sources of departures from circularity. The diagnostic elements are seen to be sample variances and covariances arising from an orthonormal transformation of the data. Weighted averages of sets of these diagnostic elements are the estimated variance components of the transformed model. The diagnostic methodology consists essentially in searching the diagnostic elements for sources of instability in the estimated variance components.

The distribution theory for the diagnostic elements is provided thereby formalizing the diagnostic process. The effectiveness of the methodology is illustrated with a variety of numerical examples including data from actual repeated measures experiments. Sample computer programs are provided for some standard applications.