American Economic Review: Insights
ISSN 2640-205X (Print) | ISSN 2640-2068 (Online)
Building Nondiscriminatory Algorithms in Selected Data
American Economic Review: Insights
(pp. 231–49)
Abstract
We develop new quasi-experimental tools to understand algorithmic discrimination and build nondiscriminatory algorithms when the outcome of interest is only selectively observed. We first show that algorithmic discrimination arises when the available algorithmic inputs are systematically different for individuals with the same objective potential outcomes. We then show how algorithmic discrimination can be eliminated by measuring and purging these conditional disparities. Leveraging the quasi-random assignment of bail judges in New York City, we find that our new algorithms not only eliminate algorithmic discrimination but also generate more accurate predictions by correcting for the selective observability of misconduct outcomes.Citation
Arnold, David, Will Dobbie, and Peter Hull. 2025. "Building Nondiscriminatory Algorithms in Selected Data." American Economic Review: Insights 7 (2): 231–49. DOI: 10.1257/aeri.20240249Additional Materials
JEL Classification
- C51 Model Construction and Estimation
- J15 Economics of Minorities, Races, Indigenous Peoples, and Immigrants; Non-labor Discrimination
- K41 Litigation Process
- K42 Illegal Behavior and the Enforcement of Law