Studying Psychotherapy Effects Through Two-Sample CFA and Markov-Chain Analysis

This paper puts forward a way of examining what psychotherapy actually does by pairing configural frequency analysis (CFA) with the tools of time-series analysis. The idea is to combine two methods that are rarely used together and see whether the combination can say something useful about change over the course of treatment.

The starting point is the set of behavior patterns a client shows in each therapy session. Those patterns are multivariate and messy. Using Two-Sample CFA, they get condensed down to a single zero-or-one judgment per session, so that what remains is a binary sequence running across the whole course of therapy.

Once you have that sequence, the next assumption is a first-order Markov dependence: each state depends on the one immediately before it. From there, two properties of Markov processes come into play, intercalary independence and the truncation property. These make it possible to test for therapy effects without relying on parametric assumptions, which matters when you are working with a single case rather than a large sample.

Krauth also lays out how results from separate single-case studies can be pooled or set against one another. A single study on its own carries limited weight; being able to combine independent cases, or compare them directly, gives the approach more reach.

To show the method in action, the procedure is applied to an empirical data set drawn from therapy research, walking through what a single-case analysis looks like in practice.

The work appears in Volume 47 (2005), Issue 3/4, authored by Joachim Krauth of the Department of Experimental Psychology at the University of Düsseldorf, and published by Pabst Science Publishers.