Predicting the Use of Outpatient Mental Health Services: Do Modeling Approaches Make a Difference?

Abstract

Studies attempting to project the impact of providing health coverage to the uninsured population have demonstrated considerable variation in the estimated costs of mental health care. Different modeling approaches to project health care use and costs have been shown to address some data characteristics well, but not all of them. Using data from Health Care for Communities, a recent national household survey, this paper attempts to estimate and predict the use of mental health outpatient services if insurance coverage were extended to the uninsured. The study employs two-part models, with the second part based on an ordinary least squares (OLS) approach and a generalized linear model (GLM), and a zero-inflated negative binomial model (ZINB). Estimates and predictions are not sensitive to the modeling approaches chosen, although the ZINB model outperforms the two-part models in terms of out-of-sample prediction.