Working Paper: NBER ID: w31256
Authors: Atul Gupta; Joseph R. Martinez; Amol S. Navathe
Abstract: Voluntary participation is a central feature of reforms being tested across US healthcare. Allowing choice can enhance effects if participants sort on unobserved treatment gains. However, selection may also bias program evaluation, misleading policymakers. We study this trade-off in the case of a national reform to reduce spending on surgeries. We exploit variation due to idiosyncratic program rules to instrument for participation. We detect considerable treatment effect heterogeneity but no evidence for sorting on treatment gains. In contrast, there is substantial selection on untreated outcomes in an unexpected direction, leading the difference in differences estimator to understate the causal effect.
Keywords: voluntary programs; bundled payments; Medicare; healthcare policy; selection bias
JEL Codes: H00; I10
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Reduction in post-acute care (I11) | Reduction in episode spending (H61) |
| Selection bias (C24) | Underestimation of average treatment effect (ATE) (C22) |
| Participation in the BPCI program (I18) | Adverse selection (D82) |
| Naive DD estimate (C51) | Underestimation of ATE (C51) |
| Financial gains (G19) | Participation in the BPCI program (I18) |
| Intrinsic motivations and administrative considerations (M54) | Participation in the BPCI program (I18) |
| Participation in the BPCI program (I18) | Reduction in episode spending (H61) |