Mixed Rasch Models for Analyzing the Stability of Response Styles Across Time: An Illustration with the Beck Depression Inventory (BDI-II)

Abstract

Questionnaires for clinical studies are often evaluated in cross-sectional settings and on the basis of classical test theory. Some of them, like the BDI-II which is one of the most widely used self-report instruments for assessing depression severity, are considered to have very good psychometric properties. However, these properties are rarely evaluated in longitudinal designs, and even less with models of item response theory (IRT). In addition, analyses of self-report questionnaires with IRT models provided evidence of two major response styles: the tendency to prefer extreme response categories, and the tendency to prefer the middle categories. Rasch models, in particular their extension to the so-called mixed Rasch model, are well suited to address these questions. They allow one to determine latent classes with different response styles and to analyze qualitative aspects of change such as the consistency of response styles across time. In this chapter first, an introduction to response styles and an overview of the mixed Rasch model, especially in the context of measuring change, are given and second, a practical example is elaborated using a sample of in-patients from a psychosomatic clinic that were assessed with the BDI-II at the beginning and at the end of in-patient treatment. The presence of two response styles is confirmed for the admission data, whereas for the discharge data the Rasch model seems sufficient. A combined analysis of both time points reveals three classes, one of which is a low symptom class and the other two reflect, again, the two response styles; these two classes remain quite stable over time.