Working Papers
No Money Bail, No Problems? Trade-offs in a Pretrial Automatic Release Program
Related Materials: Slides (July 2022), 10-minute presentation (Sep 2022), Probable Causation Interview (Dec 2022)
Media Coverage: Phenomenal World
Many jurisdictions across the United States are implementing bail reform programs to reduce the use of money bail. Bail reform opponents argue that money bail is critical for averting pretrial misconduct, while proponents counter that the effects are small and not worth the consequent costs of pretrial detention. I examine this detention-misconduct trade-off using a program in Kentucky that automatically released people arrested for low-level offenses – people who usually would have had financial conditions for release from jail. I find that the program reduced total annual time in pretrial detention by over 25 person-years with no detectable effect on pretrial rearrest. Meanwhile, the program increased the number of annual court non-appearances by about 364. This trade-off is desirable if 1 court non-appearance is less costly than 26 days in detention.
The Hidden Effects of Algorithmic Recommendations
Updated July 2023, subsumes and replaces "If You Give a Judge a Risk Score" (2019)
Related Materials: Slides (May 2023), 10-minute presentation (Sep 2022)
Media Coverage: WIRED, Axios, Probable Causation
Algorithms inform human decisions in many high-stakes settings. They provide decision-makers with predictions concerning the probability of an event. However, there is typically an additional step involved: decision-makers are recommended particular decisions based on the predictions. I isolate the causal effects of these algorithmic recommendations by leveraging a setting in which the recommendations given to bail judges changed, but the algorithm’s predictions given to judges did not. Recommendations significantly impacted decisions: lenient recommendations increased lenient bail decisions by over 50% for marginal cases. I explore possible mechanisms behind this effect and provide evidence that recommendations can change the costs of errors to decision-makers. Judges may be more lenient when their choices are consistent with recommendations because the recommendation can shield them from political backlash. Finally, I show that variation in adherence to recommendations complicates how algorithm-based systems affect racial disparities. Judges are more likely to deviate from lenient recommendations for Black defendants than for white defendants with identical algorithmic risk scores.
After The Burning: The Economic Effects of the 1921 Tulsa Race Massacre
(with Jeremy Cook, James Feigenbaum, Laura-Thorne Kincaide, Jason Long, and Nathan Nunn)
Related Materials: NBER Digest Article (Sep 2021)
Media Coverage: The New York Times, CNN, TIME Magazine, CNBC, Black Enterprise, The Root, Business Insider
The 1921 Tulsa Race Massacre resulted in the looting, burning, and leveling of 35 square blocks of a once-thriving Black neighborhood. Not only did this lead to severe economic loss, but the massacre also sent a warning to Black individuals across the country that similar events were possible in their communities. We examine the economic consequences of the massacre for Black populations in Tulsa and across the United States. We find that for the Black population of Tulsa, in the two decades that followed, the massacre led to declines in home ownership and occupational status. Outside of Tulsa, we find that the massacre also reduced home ownership. These effects were strongest in communities that were more exposed to newspaper coverage of the massacre or communities that, like Tulsa, had high levels of racial segregation. Examining effects after 1940, we find that the direct negative effects of the massacre on the home ownership of Black Tulsans, as well as the spillover effects working through newspaper coverage, persist and actually widen in the second half of the 20th Century.
Resting Papers
Uncorking Expert Reviews with Social Media: A Case Study Served with Wine
(with Peter Pedroni and Stephen Sheppard)