Selection and Causal Effects in Voluntary Programs: Bundled Payments in Medicare
Governments and firms frequently introduce new contracts on an optional basis, aiming to improve agent performance and outcomes. It is well established that recovering the average treatment effect by evaluating such opt-in programs is challenging due to selection bias and treatment effect heterogeneity. However, a countervailing benefit is that principals can design the program to attract participants whose behavior they would most like to change, usually by offering financial incentives. We quantify both these potential confounders in the context of a large voluntary payment reform introduced by Medicare to reduce spending. Idiosyncratic program rules generated plausibly exogenous variation in expected gains across hospitals, which we exploit for identification using an instrumental variable approach. The IV estimate of the causal effect on spending for complier hospitals is much larger than the difference in differences estimate. We estimate marginal treatment effects to recover the average treatment effect across all hospitals, revealing that the difference between the IV and DD estimates is largely due to selection bias, not treatment effect heterogeneity. The program design was successful in attracting hospitals who otherwise would have done worse, but this feature also led naive evaluations to understate its effectiveness.