Pre-Trial Release Decisions and Machine Crime Forecasting
Rearrest Rate of Released Defendants moved from 18.2% to 13.7% — a +-4.5 percentage-point (-25% relative) change (sample: 750,000 cases in New York City).
Study details
- Challenge
- Judges decide whether to release defendants before trial, but struggle to predict failure-to-appear or rearrest rates, resulting in either excessive incarceration or high crime rates.
- Intervention
- Evaluated a machine learning model designed to predict defendant flight risk and crime probabilities, compared against the historical release rates and subsequent rearrests under human judges.
- Outcome
- Rearrest Rate of Released Defendants: 18.2% (control) → 13.7% (treatment); +-4.5 pts, -25% relative
- Sample
- 750,000 cases in New York City
- Key takeaway
- Algorithmic decision rules can reduce crime by 24.8% without changing the jail population size, or reduce jail populations by 41.9% with no increase in crime, proving that human judges are heavily influenced by salient, noisy details.
- Biases leveraged
- Algorithm AversionSalience Bias
- Source
- Kleinberg, J., Lakkaraju, H., Leskovec, J., Ludwig, J., & Mullainathan, S. (2018). Human decisions and machine predictions. Quarterly Journal of Economics, 133(1), 237-293. ↗ source
Cite this page
Behavioral Economics Lab. "Pre-Trial Release Decisions and Machine Crime Forecasting." Behavioral Economics Lab, https://behavioraleconomicslab.com/findings/judicial-bail-algorithms.
Primary source: Kleinberg, J., Lakkaraju, H., Leskovec, J., Ludwig, J., & Mullainathan, S. (2018). Human decisions and machine predictions. Quarterly Journal of Economics, 133(1), 237-293.