Application of modern psychometric theory in psychiatric research.

This paper describes the general statistical theory of item-response modeling as developed in the fields of statistics and education. Generalization of these procedures for application in the analysis of psychiatric rating scales is the focus of this paper. Questions of unidimensionality vs multidimensionality and choice of distributional transform (logistic vs normal) are both discussed and statistically examined using data on the Beck Depression Inventory (BDI). Application of these models to the BDI revealed two subscales that maximally differentiate high and low levels of depression in psychiatric and medically ill patients respectively. There was considerable but not complete overlap between the two subscales. These statistical models are found to have desirable properties when used to analyze psychiatric rating scales and provide a refinement over existing techniques of classical test theory and factor analysis.