Working Paper: CEPR ID: DP13981
Authors: Esther Duflo; Abhijit Banerjee; Daniel Keniston
Abstract: Should police activity should be narrowly focused and high force, or widely dispersedbut of moderate intensity? Critics of intense “hot spot” policing argueit primarily displaces, not reduces, crime. But if learning about enforcementtakes time, the police may take advantage of this period to intervene intensivelyin the most productive location. We propose a multi-armed bandit modelof criminal learning and structurally estimate its parameters using data froma randomized controlled experiment on an anti-drunken driving campaign inRajasthan, India. In each police station, sobriety checkpoints were eitherrotated among 3 locations or fixed in the best location, and the intensity of thecrackdown was cross-randomized. Rotating checkpoints reduced night accidentsby 17%, and night deaths by 25%, while fixed checkpoints had no significanteffects. In structural estimation, we show clear evidence of driver learning andstrategic responses. We use these parameters to simulate environment-specificoptimal enforcement policies.
Keywords: learning models; choice modeling; information acquisition; illegal behavior; law enforcement; crime prevention
JEL Codes: No JEL codes provided
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| rotating sobriety checkpoints (R48) | night accidents (R41) |
| rotating sobriety checkpoints (R48) | night deaths (Y70) |
| fixed checkpoints (C62) | night accidents (R41) |
| fixed checkpoints (C62) | night deaths (Y70) |
| rotating sobriety checkpoints (R48) | lasting change in driver behavior (R48) |