Moral Hazard and Demand for Physician Services: An Estimation Using the Rubin Causal Model Framework
Posted: 12 Jun 2007
Date Written: June 2007
During the last twenty years a large empirical literature has been devoted to the effect of health insurance on medical care utilization. Developments in contract theory have shown that if part of the risk is private information, agents will self select on the basis of this private information and a positive correlation between risk and coverage should be observed (adverse selection or selection effect). On the other hand comprehensive coverage provides no incentive to reduce risk and if part of that risk is endogenous agents with more comprehensive coverage will produce less effort to reduce it than those with incomplete coverage (moral hazard or incentive effect). Thus opposite causality generates the same correlation between coverage and risk. Empirically it is quite tricky to distinguish between adverse selection and moral hazard. Traditionally when no experimental data are available the estimation strategy is to find instrumental variables for the insurance decision.
We propose here a new identification strategy: we try to disentangle adverse selection and moral hazard using the fact that in France some employees are provided mandatory additional insurance by their employer. It relies on the hypothesis that, when mandatory, the employer supplemental insurance is exogenous to the employees' medical care utilization conditionally on a number of individual characteristics (such as age, sex, education, industry). The empirical framework is based on the model set up by Angrist, Imbens and Rubin (1996) : the Rubin Causal Model (RCM). In their terminology, the treatment variable is whether one has additional insurance or not. This supplementary coverage can either be provided by the employer, on a voluntary or mandatory basis depending on the employer, or it can be contracted individually. The exclusion condition is the fact this supplementary insurance provided by some employers is mandatory. Our two main contributions to research are the use of this new intrumental variable and the fact we put it in a framework in which the treatment is heterogeneous across individuals. Moreover, in the first part of our work, the approach is non parametrical and thus, we believe, rather robust.
We estimate the Local AverageTreatment Effect (LATE) for the non treated (people who don't have supplemental insurance), and focus on annual number of visits to the physician. Our dataset is the Enqute Dicennale Santi 2002-2003, a large survey conducted every ten years in France, with 17 000 households interviewed. The main results are the following. We cannot reject the hypothesis of no adverse selection. We find an incentive effect, although imprecisely estimated : comprehensive coverage would increase by 20 percentage points the probability that 'compliers' would visit a doctor at least once a year (without any supplemental insurance, this probability is around 70%). Our results suggest that people less covered use less medical services mainly because they contribute more to the payment of these services. Eventually we discuss the validity of our exclusion hypothesis by testing it on hospital services, for which part of supplemental insurance is more limited.
Suggested Citation: Suggested Citation