|
Abstract:
|
“Report cards” on service provider performance are becoming common in multiple sectors including education, health care, and the justice system. Typical report cards, however, often do not account for massive variability in casemix. Specifically, doctors differ greatly on the mix of patients they treat and police officers work in different environments facing variation in suspicious activity. Fair and effective report cards and risk management systems must address the casemix problem so that observed differences in outcomes cannot be blamed on confounding. I will present a statistical method for customizing performance benchmarks by blending three methodologies: propensity score weighting, doubly robust estimation, and false discovery rates. I will demonstrate the methodology in a variety of contexts. I flag providers writing excessive opioid prescriptions, hospitals with unusually high mortality rates, communities with deep dissatisfaction with police services, and specific police officers with inexplicably high rates of stopping black pedestrians. The proposed benchmarking methodology has broad applications for improving system performance by identifying underperforming and problematic system components.
|
