Read on Twitter
As we look to understand the coronavirus pandemic, it seems increasingly important to evaluate hospital performance. But, estimating hospital quality is very difficult. Together with @KelliMarquardt, we created Stata code that can help. 1/7 https://ideas.repec.org/c/boc/bocode/s458713.html
The Stata code is based on a paper in Econometrica ( @econometricsoc) with John Geweke and Robert Town that addresses the difficult problem of estimating hospital quality when selection matters and is based on unobservables. 2/7 https://doi.org/10.1111/1468-0262.00444
Why might selection matter? Sicker patients may choose higher quality hospitals. If we just looked at covid-19 death rates by hospital, this will then pick up both the quality of care and unobserved patient severity, resulting in biased estimates. 3/7
The paper (and code) combine choice and mortality models to separate out the parts of mortality that are due to selection and quality. Identification requires an exclusion restriction. We used distance; others use ambulance referrals. What might be good ones for covid-19? 4/7
The paper uses Bayesian estimation methods. Why Bayesian? The methods are helpful in extracting a limited signal from a lot of noise. *Think shrinkage estimator.* They efficiently combine aspects of fixed-effects and random-effects estimators. 5/7
This sounds like it might be really hard to implement! That’s where our (beta) Stata code comes in. You need to format the data in the right away. But it’s about as difficult as what you’d do in estimating a multinomial choice model in Stata. 6/7
We’re here to help! If you’d like to use the code (for anything from covid-19 care quality to school outcomes), download it: “ssc install ggt,” read the help file, and then contact us if you have questions. 7/7
