Drunk Drivers: Beware the Economists

Every year, traffic accidents cause 50 million injuries and 1.35 million deaths globally. A staggering 90% of these deaths happen in developing countries. Though it is difficult to isolate accidents caused by drunk driving, they are estimated to represent 30% of all accidents.

In countries like India with high frequencies of murder, kidnapping, rape, and theft, drunk driving is often insufficiently monitored and penalized. Police occasionally conduct breath checks at a fixed location on routes considered prone to crime. However, the duration, frequency, and intensity of these checks is very low due to budget constraints and a shortage of police officers.

Given these resource constraints, how can the police curtail drunk driving? An experiment in India (PDF)—conducted by U.S. economists in partnership with Indian police—found that the police can utilize game theory to deter drunk driving and thereby reduce car accidents. The game theory approach begins with the drunk driver’s belief that a particular route either leads to no police encounter or to a stop at a checkpoint. While not every checkpoint is placed by police, the researchers assume, by design, that if a drunk driver is caught at checkpoint, or someone they know is caught and informs them, they will believe that taking that route will lead to a police encounter. With police-imposed breath checkpoints at a fixed location, the drunk driver has three strategies: stay at home temporarily until the removal of the checkpoint, use alternative routes, or generally refrain from drunk driving.

Motivated to help police obtain a strategic advantage, Abhijit Banerjee, Esther Duflo, Daniel Keniston, and Nina Singh conducted randomized control trials with police stations in the Indian state of Rajasthan that tested the effect of unpredictable crackdowns on drunk driving. The trials occurred in two phases—September to October 2010 and September to November 2011—and covered 11 districts and 183 police stations spanning 125,000 square kilometres and 30 million people.

While the control police stations had no intervention, the treatment stations adopted various permutations of checkpoint policies: the location, frequency, and duration of breath check programs altered between different stations. Two enforcement agencies were used to implement the program: the Police Lines Force (considered a punishment posting) and the Rajasthan local police force. If they carried out the intervention, officers in the Police Lines Force were promised the incentive of being allowed to transfer out of the posting. The Rajasthan police were not provided additional incentives.

Drunk drivers were aware of budget and police personnel constraints and hence knew that the checks would not last forever. However, as the crackdowns were not officially announced it was difficult for drivers to know when they would be penalized. The amount of time drunk drivers had to pursue the strategy of staying home became uncertain. When the frequency and location of checks were randomized as part of the experiment, it became increasingly difficult for drunk drivers to accurately update their priors as to when and where they would encounter a checkpoint. The earlier strategy of using alternative routes was no longer optimal.

The intervention appeared to have worked: nighttime accidents and deaths fell by 17% and 25%, respectively. More importantly, the effect lasted not only during the intervention period but up to three months after the program ended. The Police Lines Force teams were more efficient in terms of arriving at the checkpoints and conducting the checks for the assigned duration. The results show that even when resources are scarce, organizations can still improve outcomes through strategic design by creating the right incentives for personnel engagement.

---

Banerjee, Abhijit, Esther Duflo, Daniel Keniston, and Nina Singh. 2019. “The Efficient Deployment of Police Resources: Theory and New Evidence from a Randomized Drunk Driving Crackdown in India.” National Bureau of Economic Research Working Paper 26224. https://doi.org/10.3386/w26224.