Estimating the binary endogenous effect of insurance on doctor visits by copula-based regression additive models

The paper estimates the causal effect of having health insurance on healthcare utilization, while accounting for potential endogeneity bias. The topic has important policy implications, because health insurance reforms implemented in the USA in recent decades have focused on extending coverage to the previously uninsured. Consequently, understanding the effects of those reforms requires an accurate estimate of the causal effect of insurance on utilization. However, obtaining such an estimate is complicated by the discreteness inherent in common measures of healthcare usage. The paper presents a flexible estimation approach, based on copula functions, that consistently estimates the coefficient of a binary endogenous regressor in count data settings. The relevant numerical computations can be easily carried out by using the freely available GJRM R package. The empirical results find significant evidence of favourable selection into insurance. Ignoring such selection, insurance appears to increase doctor visit usage by 62% but, adjusting for it, the effect increases to 134%