An Application of the Item-Response Model to Psychiatric Epidemiology

This article presents the item-response model and an application of the model to variables representing symptoms of anxiety and depression. The model is presented as having features in common with a multivariate logistic regression where the independent variable is continuous, real-valued, and unobservable. Some connections to dichotomous variable factor analysis and the latent class model are also discussed. In the application, data are used from the National Institute of Mental Health Epidemiologic Catchment Area Survey. The model is most successful with symptoms of anxiety, where the variables are small in number and satisfy the assumption of unidimensionality. It is less successful with the symptoms of major depressive disorder, where the unidimensionality does not seem to hold.