Aperçu

Résumé (français)

Veuillez noter que les résumés n'apparaissent que dans la langue de la publication et peuvent ne pas avoir de traduction.

Résumé (anglais)

Predicting health care utilization is the foundation of many health economics analyses, such as calculating risk-adjustment capitation payments or measuring equity in health care utilization. The most common econometric models of physician utilization are parametric count data models, since the most common metric of physician utilization is the number of physician visits. This paper makes two distinct contributions to the literature analyzing GP utilization: (i) it is the first to use a nonparametric kernel conditional density estimator to model GP utilization and compare the predicted utilization with that from a latent class negative binomial model; and (ii) it uses panel data to control for the potential endogeneity between self-reported health status and the number of GP visits. The goodness-of-fit results show the kernel conditional density estimator provides a better fit to the observed distribution of GP visits than the latent class negative binomial model. There are some meaningful differences in how the predicted conditional mean number of GP visits changes with a change in an individual’s characteristics, called the incremental effect (IE), between the kernel conditional density estimator and the latent class negative binomial model. The most notable differences are observed in the right tail of the distribution where the IEs from the latent class negative binomial model are up to 190 times the magnitude of the IEs from the kernel conditional density estimator.