Estimating Hospital Quality with Quasi-Experimental Data

Abstract

Non-random sorting can bias observational measures of institutional quality and distort quality-based polices. I develop alternative quasi-experimental approaches to quality estimation that accommodate nonlinear causal effects, institutional specialization, and unobserved selection-on-gains. I use this framework to compute empirical Bayes posteriors of the quality of 4,821 U.S. hospitals, combining estimates from ambulance referral quasi-experiments with predictions from observational risk-adjustment models. Higher-spending, higher-volume, and privately-owned hospitals are of higher quality, and most healthcare markets exhibit positive Roy selection-on-gains. I then simulate Medicare reimbursement and consumer guidance programs based on different hospital quality measures. Higher-spending providers tend to see moderately larger performance-linked subsidies when quality posteriors replace conventional rankings, while teaching hospitals are reimbursed relatively less. Admissions policy simulations highlight limitations of consumer guidance programs in settings with unobserved Roy selection: redirecting patients to top-ranked hospitals may worsen expected survival when based on observational rankings, while quasi-experimental rankings appear to generate modest gains.