Partial identification of treatment effect in binary outcome models: An application to health insurance and dental service utilisation

Hosted by Pravin Trivedi

Recent developments in the literature of partial identification have significant implications for the econometric estimation of important policy effects. In empirical economics, it is often of interest to estimate the effect of a binary policy treatment variable on a binary outcome variable where both may be driven by common observable and unobservable factors. A common approach is to assume a parametric model, such as a bivariate probit, together with the use of instrumental variables to achieve point identification.

Partial identification analysis of such problems allows for less restrictive assumptions for the underlying data generating process (DGP) in empirical applications, and the estimated bounds offer more robust measures for policy impacts. This paper applies the partial analysis approach to a health economics application. We estimate the bounds for average treatment effect (ATE) of private health insurance status on dental service utilisation, using data from the Australian National Health Survey.

Four sets of bounds from the literature under varying DGP assumptions and their 95% confidence regions are estimated. The resulted ATE confidence bounds are much wider than the confidence intervals using a conventional bivariate probit. We found that two of the bounds have reasonably narrow widths to be informative. We also estimate bounds for different sub-populations with varying widths. Performances of global parametric, local parametric, and smoothed and raw non-parametric estimators for bounds are also studied using generated data.