The symmetry of symptom patterns in pre-post treatment designs

This article proposes a new nonparametric method for statistical evaluation of clinical pre-post treatment designs. In clinical research, models of marginal symmetry typically are estimated from log-linear models of axial and quasi-symmetry. As such, they provide overall goodness-of-fit information concerning change in probabilities of categories of one variable that was observed twice. This paper proposes the following three extensions: (1) using models of marginal symmetry for changes in patterns of two or more variables, and (2) following up global marginal symmetry tests using Lehmacher's sign tests. (3) To protect the experiment-wise alpha, a modified Bonferroni-Holm procedure is proposed. The new approach allows researchers to make statements about treatment effects at the level of single symptoms. Examples illustrate application of all three symmetry models and the follow-up test using data from pharmaco-psychiatry. The discussion relates Lehmacher's tests to two-sample Configural Frequency Analysis of multi-discrimination types. Strategies of statistical significance testing are presented and the importance of the proposed methodological approach for psychiatric research is discussed.