Research

Working Papers

Fickle Discrimination: Evidence from Emergency Triage (Job Market Paper)

Presentations: NBER Economics of Health (scheduled), SITE Economics of Health Inequality, Princeton University (SWEET), University of Pittsburgh, ESA World Meeting  

Abstract: Discrimination is usually attributed to decision-makers' stable beliefs and preferences. I show that persistent and consequential discrimination can also be driven by a more fickle process: transitory and irrelevant features of the decision context sharply increase or decrease disparities produced by the same decision-maker from one moment to the next. In 1.9 million emergency department triage decisions, non-white patients are 6 percentage points (21%) less likely than observably identical white patients to receive top priority from the same nurse in the same shift. Exploiting quasi-random patient arrivals and nurse assignment, I find that the racial triage gap widens by 27% immediately after encounters with familiar, lower-risk frequent emergency department users, most of whom are non-white. This increase masks opposite responses by frequent-user race: the gap widens after non-white frequent users (+43%) but narrows after white ones (-20%). The racial triage gap moves in the same directions with longer-run exposure, reversing against white patients after enough exposure to white frequent users. Time pressure leads nurses to collect less patient information and respond nearly twice as strongly to frequent-user encounters. These results cannot be explained by standard statistical learning but point to nurses assessing patients by coarse racial similarity to prior lower-risk cases, particularly when rushed. This fickle discrimination pushes triage away from a race-blind machine-learning benchmark and carries into downstream care, including hospital admission. Removing frequent-user and time-pressure effects would shrink the overall racial triage gap by at least 28%. These findings suggest disparities are more malleable than previously believed and respond to how institutions allocate cases and time.

Gender Bias and Differential Belief Updating

Abstract: I study whether people respond differently to identically informative signals about men and women. To ensure signals convey identical information, I develop a novel experimental design that fully fixes both prior beliefs and signal distributions across gender. In an experiment with 3,204 subjects, individuals respond dramatically to the gender of the signal-sender. Subjects are more likely to violate Bayes' Rule, failing to update or updating in the wrong direction, after learning a woman succeeded or a man failed. They update more rationally after learning a man succeeded or a woman failed. Additional treatments explore how to mitigate these gendered mistakes.

Publications

Please Take a Minute: How Prosocial Choices Change with Deliberation 

(with Judd B. Kessler, Hannu Kivimaki, and Muriel Niederle)

Accepted, American Economic Journal: Microeconomics

Measuring Discrimination with Experiments 

(with Corinne Low)

Handbook of Experimental Methods in the Social Sciences, edited by Alex Rees-Jones, Edward Elgar Publishing Ltd, 2026. 

A Conditional Cash Transfer and Women's Empowerment: Does Bolsa Familia Influence Intimate Partner Violence? 

(with Elizaveta Perova and Sarah Anne Reynolds)

Social Science & Medicine, 238 (112462), 2019.  

Selected Work in Progress

The Glass Cliff: Leadership Selection During Crises 

Reducing Avoidable Emergency Department Use: A Field Experiment