Modelling bivariate count distributions with finite mixture models: application to health care demand of married couples

Abstract

Count measures of health care utilization are often correlated with other measures of utilization. In addition, utilization measures display a high proportion of zero observations. This article attempts to accommodate both data features in one model, with an application to medical care usage of husbands and wives. A bivariate count representation is used to model dependence between husbands' and wives' utilizations, and a finite mixture specification accommodates the problem of excess zeros. Results show that married couples are characterized by two distinct subpopulations according to the intensity of utilization of the wife.