Studying Change in Configural Frequency Analysis Through Different Base Models

In configural frequency analysis, a type or an antitype only appears when the observed data pulls far enough away from what the null hypothesis would predict. That dependence matters more than it first seems. Whatever a type is taken to mean rests entirely on how the null model was written. Change the assumptions behind the null, and the same cell in a table can tell a different story.

Mark Stemmler and Alexander von Eye work through three base models aimed at measuring change. Two of them already had a place in the literature: directed configural frequency analysis (DCFA) and prediction configural frequency analysis (PCFA). The third is their own addition, which they call configural frequency analysis of change, or Change-CFA for short.

The Change-CFA model draws on marginal homogeneity in contingency tables. Its probability model assumes the pre-treatment and post-treatment marginals are equal, and that assumption does the heavy lifting. When a type or antitype surfaces under this setup, it reads as a movement from the pretest state to the posttest state rather than as a static association.

To see how the three approaches behave in practice, the authors run all of them against a single dataset drawn from psychopharmacological treatment. Comparing the output across models makes the point they are after: CFA bends to fit a range of hypotheses rather than locking the analyst into one fixed question about change.