The Efficient Deployment of Police Resources: Theory and New Evidence from a Randomized Drunk Driving Crackdown in India

How to cite

Keniston, Daniel; Banerjee, Abhijit; Chattopadhyay, Raghabendra; Duflo, Esther; Singh, Nina. 2017. "The Efficient Deployment of Police Resources: Theory and New Evidence from a Randomized Drunk Driving Crackdown in India." AEA RCT Registry. AEARCTR-0002343. https://doi.org/10.1257/rct.2343-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Daniel Keniston* , Abhijit Banerjee , Raghabendra Chattopadhyay , Esther Duflo , Nina Singh

* Primary investigator

Registry ID
AEARCTR-0002343
DOI
https://doi.org/10.1257/rct.2343-1.0
First registered
Last updated
Status
Completed
Trial dates
2009-01-01 – 2012-08-31
Countries
United States of America (Massachusetts)
Registry record
http://www.socialscienceregistry.org/trials/2343

Abstract

A central question in law and economics is whether, with limited resources, police activity should be narrowly focused and high force, or widely-dispersed but of moderate intensity. Critics of intense “hot spot” enforcement argue that this approach will only lead to crime shifting to other locations or other times. But if law breakers take time to learn that enforcement has begun, the police may take advantage of this period to intervene intensively in the most productive location. We propose a model where criminals progressively learn about policing, and structurally estimate its parameters using a randomized controlled experiment on an anti-drunk driving campaign with the police department in Rajasthan, India. In each station, sobriety checkpoints were either randomly rotated on 3 main routes or fixed in the best route, and the intensity of the crackdown was cross-randomized. We find clear evidence of driver learning about the crackdown and strategically responding to it, causing rotating checkpoints to dominate the fixed location approach. Our structural estimates allow us to design the optimal policy: we find that 50% of checks should occur on roads with, ex-ante, less criminal activity. We estimate that crackdowns in rotating locations reduced night accidents in the area covered by a particular police station by 17%, and night deaths by 25% over a two month crackdown and 6 weeks following it.

Intervention

In partnership with the Rajasthan Police, the PIs conducted a randomized control trial to evaluate an anti-drunk driving program using sobriety checkpoints in Rajasthan, India. In September 2010, the PIs began by randomly assigning police stations to a control or treatment groups. Treatment assignment was stratified by district, whether a station was on a national highway, and total vehicle accidents from 2008 to 2009.

The analysis suggest that the intervention did decrease accidents and deaths, and that these impacts are largely driven by the results at police stations assigned to random checkpoints. Stations with rotating checkpoints saw 29% fewer accidents and 30% fewer deaths, statistically significant at the 10% and 5% level. These effects persist 90 days post-intervention.

Experimental design

The treatment varied along four dimensions: location, frequency, duration, and personnel. For location, police stations were randomly assigned to hold all sobriety check points at a fixed location for the duration of the program, or to alternate checkpoints between three different locations. The fixed checkpoint was held at the single best location to prevent accidents from drunk driving, while the rotating checkpoints alternated between the best, second best, and third best locations to prevent such accidents. These locations were identified as such by local police chiefs prior to treatment assignment. For frequency, police stations were randomly assigned to hold checkpoints once, twice, or three times a week. Control police stations held no checkpoints. For duration, the date of the last checkpoint was randomly assigned; durations ranged from two to three months. For personnel, checkpoints were randomly assigned to be manned by police officers from the local station, or district reserve officers on "line duty", perceived to be a punishment assignment that officers may be motivated to escape by good performance.

The intervention was rolled out in two phases. The first phase served as a pilot and took place at 40 police stations. All treatment stations held checkpoints twice a week; otherwise, the pilot phase was identical to the second phase. The PIs used a combination of administrative data, on road accidents and deaths, and data collected by hired surveyors on the number of vehicles that passed by and were stopped at checkpoints.

Design details

Randomization method
Randomization was performed by computer. Treatment status was randomly assigned and stratified by district, whether the station was located on a national highway, and total accidents from 2008 to 2010.
Randomization unit
Police stations
Clusters
First phase: 40 police stations Second phase: 183 police stations
Observations
1580 checkpoints
Arms
control group - 76 police stations; rotating location, once a week, local police checkpoints - 10 stations; rotating location, once a week, non-local police checkpoints - 8 stations; rotating location, twice a week, local police checkpoints - 16 stations; rotating location, twice a week, non-local police checkpoints - 16 stations; rotating location, three times a week, local police checkpoints - 10 stations; rotating location, three times a week, non-local police checkpoints - 10 stations; fixed location, once a week, local police checkpoints - 14 stations; fixed location, twice a week, local police checkpoints - 13 stations; fixed location, three times a week, local police checkpoints - 11 stations; fixed location, once a week, non-local police checkpoints - 9 stations; fixed location, twice a week, non-local police checkpoints - 7 stations; fixed location, three times a week, non-local police checkpoints - 9 stations
Primary outcomes
Accidents, deaths, number of vehicles passing checkpoints, number of vehicles stopped at checkpoints

Cited by

No citation data available for this registration yet.

Keywords