Public Finance: Measuring Racial Discrimination in Pretrial Bail Decisions
Peter Hull, University of Chicago
Abstract:
Large racial disparities exist in many settings, including criminal justice and healthcare delivery systems, but it is unclear whether these disparities reflect omitted variables or racial discrimination. We develop a quasi-experimental approach to measure and characterize discrimination in the context of pretrial bail decisions. Race-specific average treatment effects (of pretrial release on subsequent misconduct) are used to purge omitted variables bias from conventional benchmarking regressions, isolating legally-unwarranted release rate disparities among black and white defendants with the same potential for pretrial misconduct. Leveraging the quasi-random assignment of bail judges in New York City, we find that approximately 70 percent of the average release rate disparity between observationally-equivalent white and black defendants is due to discrimination. Nearly all judges discriminate against black defendants, with higher levels of discrimination among judges who are older, more stringent, and less exposed to black cases. To explore the form that this discrimination takes, we develop and estimate a hierarchical marginal treatment effects model in which judges can differ in their rankings of defendant misconduct risk (in violation of the conventional monotonicity assumption). We find evidence of both racial bias and statistical discrimination, with the latter arising from both lower mean risk and more precise risk signals for white defendants.
joint with David Arnold and Will Dobbie
Large racial disparities exist in many settings, including criminal justice and healthcare delivery systems, but it is unclear whether these disparities reflect omitted variables or racial discrimination. We develop a quasi-experimental approach to measure and characterize discrimination in the context of pretrial bail decisions. Race-specific average treatment effects (of pretrial release on subsequent misconduct) are used to purge omitted variables bias from conventional benchmarking regressions, isolating legally-unwarranted release rate disparities among black and white defendants with the same potential for pretrial misconduct. Leveraging the quasi-random assignment of bail judges in New York City, we find that approximately 70 percent of the average release rate disparity between observationally-equivalent white and black defendants is due to discrimination. Nearly all judges discriminate against black defendants, with higher levels of discrimination among judges who are older, more stringent, and less exposed to black cases. To explore the form that this discrimination takes, we develop and estimate a hierarchical marginal treatment effects model in which judges can differ in their rankings of defendant misconduct risk (in violation of the conventional monotonicity assumption). We find evidence of both racial bias and statistical discrimination, with the latter arising from both lower mean risk and more precise risk signals for white defendants.
joint with David Arnold and Will Dobbie
| Building: | Lorch Hall |
|---|---|
| Event Type: | Workshop / Seminar |
| Tags: | Economics, seminar |
| Source: | Happening @ Michigan from Department of Economics, Public Finance, Department of Economics Seminars |
