HETEROGENEOUS EFFECT OF COINSURANCE ON HEALTHCARE COSTS

Originally entered health economics literature as a concept related to moral hazard, coinsurance rate has been shown to impact the intensity of medical treatment.  Accordingly, various forms of copayments and coinsurance have been widely used in the U.S. since 1970s and currently are given a special emphasis within Medicare’s part D reform. While empirical evidence generally demonstrates negative effect of coinsurance rate on health care expenditure for the whole pool of consumers, such estimations may conceal the issues of price inelastic demand for certain sub-groups. Therefore, disentangling variation in consumer response to coinsurance rate may be viewed as an integral part of making the judgments about the efficiency of health insurance policies.

The novelty of the present paper is a combination of matching estimators and finite mixture models, which account for unobservable consumer heterogeneity and nonlinear effects of coinsurance rate on health care expenditure. The second contribution is an application of panel data finite mixture generalized linear models. Finally, the paper is the first assessment of a natural experiment in terms of heterogeneous effect of coinsurance rate.

We focus on the most recent rise in nominal coinsurance rate for a number of Japanese consumers (April 2003). The analysis exploits a unique combination the 2000-2008 longitudinal data of The Japanese Panel Survey of Consumers (the longest household survey, with a wide range of questions on health, health care expenditure, and lifestyle of Japanese females aged 24-49) and the novel Japan Household Panel Survey, which differentiates between health care expenditure covered and non-covered by health insurance.

The results demonstrate that both loglinear model and generalized linear models with Weibull distribution family provide adequate fit of health care expenditure. The estimations with panel data finite mixture models reveal that the population of Japanese females decomposes into groups with high and low health care expenditure. Group membership is explained by lifestyle variables: number of female friends (a proxy for latent health status) and hours overtime work (a proxy for time cost). The groups constitute 38% and 62% of the whole age-demographic cohort in the loglinear model and 29% and 71% in the generalized linear model with Weibull distribution family. These values are generally close to the relative shares of adult U.S. consumers with high and low number of outpatient visits (23% and 73%, Deb and Trivedi 2002) or Medicare elderly  consumers with high and low health care expenditure (25% and 75%, Munkin and Trivedi 2010). 

Overall, the effect of nominal coinsurance rate is negative and significant. While marginal effect is stronger in the component with low health care expenditure, conditional average treatment effect in difference-in-difference estimations is larger in absolute terms in the component with high health care expenditure. The values of conditional average treatment effect coefficients (compared to coefficients obtained in mean unconditional and conditional comparison) reveal the presence of nonlinear effects in each component of the model.