Specification and Simulated Likelihood Estimation of a Non-normal Outcome Model with Selection: Application to Health Care Utilization

We develop a model specification and estimation framework that is applicable to many microeconometric models which fall into a treatment-outcome framework. We apply our methodology to examine the causal effect of man-aged care on the utilization of health care services. Specifically, we jointly model multinomial choice of insurance plans (treatment) and counts and binary choices of utilization (outcome) using a latent factor structure, enabling a distinction between selection on unobservables and observables. We apply maximum simulated likelihood techniques to estimate the parameters of our model and find that there are significant unobserved self-selection effects and that these effects substantially change the effects of insurance on utilization.