Estimating Hospital Quality with Quasi-experimental Data

Non-random sorting can bias outcome-based measures of institutional quality and distort the growing set of quality-based policies intended to incentivize institutions and inform consumers. I develop an alternative framework for quasi-experimental quality estimation that accommodates selection on unobservables, nonlinear causal effects, institutional comparative advantage, and Roy selection. I then use this approach to compute empirical Bayes quality posteriors from the 30-day mortality rates of a large set of U.S. hospitals. These posteriors optimally combine estimates from quasi-experimental ambulance company assignment and predictions from observational risk-adjustment models (RAMs) along the lines of a conventional bias-variance tradeoff.

I find that higher-spending, higher-volume, and privately-owned hospitals have better quality posteriors, and that most emergency healthcare markets exhibit positive selection-on-gains with patients being admitted to more appropriate hospitals on average. I then quantify the effects of this non-random selection by simulating Medicare reimbursement and consumer guidance policies that use quality posteriors instead of RAMs. The types of hospitals subsidized by performance-linked payment schemes (e.g. Value-based Purchasing) are largely unchanged when quasi-experimental data is incorporated, but existing transfers are magnified. My admission policy simulations, however, highlight the limitations of consumer guidance programs in settings with significant selection on match-specific quality.