Estimating the Effect of Insurance on Doctor Visits by Copula-Based Regression Additive Models

Regression is one of the core statistical methods and is used in a wide variety of empirical applications. It typically involves one response variable and a set of covariates. However, the importance of modelling simultaneously two or more responses conditional on some covariates has been increasingly recognised. Starting from a case study in the field of insurance, in this talk I will discuss a model developed to help answer some real questions, and make links to the more general modelling framework that I have been co-developing for the past 10 years. The statistical framework builds upon copulae, a rich variety of distributions and smoothing splines, and has so far found use in many practical situations in the fields of medicine, political and social science, microeconomics and epidemiology, to name but a few. The modelling framework has been implemented in the R package GJRM (Generalised Joint Regression Modelling) which has been created to facilitate the use of such models in industry and academia and to enhance reproducible research, two aspects often neglected in scientific research. The core algorithm of GJRM is based on a carefully designed and very generic penalised likelihood-based estimation approach. The framework is illustrated on a case study which investigates the effect of insurance status on doctor visits using the US Medical Expenditure Panel Survey. The method finds statistically significant evidence that insurance is endogenous with respect to usage of doctor services. When endogeneity is taken into account, the effect of insurance is larger than when endogeneity is ignored.

Giampiero Marra, Statistical Science, University College London

Rosalba Radice, Cass Business School, University of London

David Zimmer, Economics Department, Western Kentucky