Analysis of multinomial counts with joint zero-inflation, with an application to health economics

Zero-inflated regression models for count data are often used in health economics to analyze demand for medical care. Indeed, excess of zeros often affects health-care utilization data. Much of the recent econometric literature on the topic has focused on univariate health-care utilization measures, such as the number of doctor visits. However, health service utilization is usually measured by a number of different counts (e.g., numbers of visits to different health-care providers). In this case, zero-inflation may jointly affect several of the utilization measures. In this paper, a zero-inflated regression model for multinomial counts with joint zero-inflation is proposed. Maximum likelihood estimators in this model are constructed and their properties are investigated, both theoretically and numerically. We apply the proposed model to an analysis of health-care utilization. •Propose a new zero-inflated regression model, for multinomial count data with zero-inflation.•Establish rigorously the asymptotics of the maximum likelihood estimator in this model.•Assess the numerical properties of the proposed estimator via extensive simulations.•Illustrate the proposed model and methodology on a data set arising from the field of health economics.