Testing exogeneity in the bivariate probit model: Monte Carlo evidence and an application to health economics

Many economic applications involve the modeling of a binary variable as simultaneously determined with one of its dycotomous regressors. In this paper we deal with a prominent health economics case study, that of cesarean section delivery utilization across public and private hospitals. Estimating the probability of cesarean section in a univariate framework neglecting the potential endogeneity of the hospital type dummy might lead to invalid inference. Since little is known about the exact sampling properties of alternative statistics for testing exogeneity of a dycotomous regressor in probit models, we conduct an extensive Monte Carlo experiment. Equipped with the simulation results we apply a comprehensive battery of tests to an Italian sample of women and find clear evidence against exogeneity of the hospital type dummy. We speculate on the economic implications of these results and discuss the misleading interpretation arising from the adoption of either univariate probit model or seemingly unrelated bivariate probit model.