Demand Elasticities and Service Selection Incentives among Competing Private Health Plans
Ellis and McGuire (2008) were the first to show that service-level selection incentives, quantified by the derivative of plan profits under a capitated system with respect to a change in the degree of tightness of rationing specific services, are proportional to the product of three terms:
ηs The demand elasticity for service s
CV(xse) The coefficient of variation in the expected level of spending on service s across individuals (the predictability of the service)
ρ(xse,X) The correlation of expected spending on service s with total spending (the predictiveness of the service, i.e., how well it predicts high total spending)
Alternatively, using a quantity-based rationing framework instead of a shadow price framework, Layton et al (2015) capture demand elasticity by ǝπ/ǝxse, where π is the probability of choosing a health plan. In this paper, we propose to estimate both price (ηs) and quantity-based (ǝπ/ǝxse) metrics empirically and use them to calculate selection indices for disaggregated services at three levels of fineness.
We build upon results in Martins et al (2015) that estimates service level demand elasticities using the Truven Analytics MarketScan claims and encounter data for a large sample of employers over a seven year period. That paper estimates short-run elasticities based on spending responsiveness within a year to changes in cost sharing induced by deductibles and stoplosses, as well as long run service elasticities by comparing people across plans and years. Time paths interacted with the size of the deductibles are used to instrument within-year price variation while person-level fixed effects and employer level instruments are used to control for endogeneity of cost sharing across years and plans. The effect of quantity rationing on health plan choice probabilities for different consumers is also estimated and compared to demand elasticities for each service.
Prior work has shown that health care services vary in their product of predictability and predictiveness by a factor of more than 100, and that plan types are correlated with selection on the identified services. This paper adds in variation in service-level demand elasticities to these calculations. It also explores the extent to which diagnosis-based risk adjustment, mixed payment, and reinsurance succeed at reducing estimated service selection effects.