Some Latent Trait and Latent Class Analyses of the Beck-Depression-Inventory (BDI)

For assessing the intensity of depression in both psychiatrically diagnosed and normal populations, the Beck Depression Inventory (BDI) is one of the most widely used instruments. The BDI was introduced in 1961 and revised in 1978. It is a self-report questionnaire and consists of 21 items derived by clinical observations about symptoms displayed frequently by depressed patients and infrequently by nondepressed psychiatric patients. The items were chosen to assess the intensity of depression and were not selected to reflect a particular theory of depression. Each item contains four statements reflecting increasing levels of intensity of a depressive symptom. The items are scored from 0 to 3 and are summed for an overall depression score.Evaluation of the psychometric properties of the BDI and tests for the presence of one or more underlying dimensions were mostly performed within the framework of classical test theory, correlational analysis and factor analysis. The psychometric properties are considered to be good, and Hautzinger (1991) confirmed the main results for a German translation of the BDI. A recent overview by Richter, Wemer, and Bastine (1994) discusses further research on these topics. Factor analyses of the BDI have resulted in several different solutions (Richter, 1991). However, a general factor has been found in many studies (and for the German translation (Hautzinger, 1991», and, therefore, the BDI is assumed to represent one underlying general syndrome of depression that can be decomposed into three highly intercorrelated factors.Studies using probabilistic models to test unidimensionality are scarce. Clark et al. applied the Bimbaum model, and Bouman & Kok (1987) performed a dichotomous Rasch analysis of the BDI. They revealed that the BDI does not, in terms of the RaschModel, constitute a unidimensional scale for the measurement of depression in clinical samples. Nonetheless, the authors proposed three subscales of the BDI that were found to be homogeneous (»Guilt & Failure«, »Mood & Inhibition« and »Somatic« (cr. Table I» and intercorrelated modestly.In this paper, we give a short summary of our previous research concerning the replication of homogeneity in a German sample and the extension to polytomous Rasch analysis and latent class analysis (LCA) of the Bouman-Kok subscales. Then, we show results of the application of LCA to analyze second order relationships of class-membership on these three subscales. The contingency of the syndromes (identified by second order LeA) with gender and time of assessment is analyzed as well. Finally, we propose several other subscales derived from the BDI (the main symptoms of depression according to DSM-III-R and three subscales composed with regard to clinical importance and similarity) and discuss their homogeneity