VoxEU Column Frontiers of economic research

Subjective beliefs, deterrence, and the propensity to drive while intoxicated

Driving while intoxicated is a serious problem in the US. What policymakers disagree about is how best to discourage drunk driving. This column argues that the perceived risk for detection has a deterrent effect on drunk driving. Harsher sanctions do not convey the desired effect if the perceived risk for detection is low. The best policy thus should increase the probability of detection or manipulate peoples’ beliefs for such a risk.

Driving while intoxicated (DWI) is a serious problem in the US, having caused more than 10,000 deaths in 2010 (US National Centre for Statistics and Analysis 2012). Studies have documented that 2% of weekend night time drivers on US roads had illegal blood alcohol concentrations (Lacey et al. 2009), and more than 110 million alcohol-impaired driving episodes occur in US annually (Bergen et al. 2011). Legally intoxicated drivers are 13 times more likely to cause fatal crashes than sober drivers (Levitt and Porter 2001). However, there is a lack of consensus whether in order to deter drunk driving, policymakers should increase punishment severity or increase the probability of detection and conviction.

In our recent paper (Chen and Sloan 2014), we investigated two related issues:

  • First, whether subjective probabilities of being pulled over and being involved in accidents when driving under the influence of alcohol would affect individuals’ drinking and driving choices;
  • Second, how changes in penalty severity following a conviction of driving while intoxicated affect individuals’ probability of driving after having had too much drink by experiments that randomised the severity of punishments.

Our data come from the Survey of Alcohol and Driving (SAD), a two-wave longitudinal survey conducted by Battelle Memorial Institute in eight geographically dispersed US cities across four states during 2010-12.1 To deal with the possibility that risk perceptions are associated with unobserved preferences for drinking and driving and thus bias the estimation results, we used an instrumental variable approach and controlled for a set of proxies for latent preferences for risky behaviours, including risk tolerance, time preference, alcohol addiction level, impulsivity, cognitive ability, and risky behaviours that may directly capture unobserved preferences for risks.


  • We found that, on average, a 10% increase in an individual’s perceived probability of being pulled over if driving after having had too much to drink would lead to a 0.16 reduction in the number of drinking and driving episodes in a year.

This reduction is relative to a mean number of 1.05 drunk driving episodes per year in the sample. A 10 percentage point increase in the perceived increases in odds of having an accident when driving intoxicated compared with driving sober would decrease the number of drunk driving episodes in a year by 0.10.

In addition, we found substantial differences in responsiveness to a change in beliefs about the adverse consequences of drinking and driving.

  • In particular, only persons who are alcohol-addicted, lack self-control over drinking, and are more impulsive reduce their number of drinking and driving episodes when perceived risks of being pulled over and having accidents increase.

By contrast, non-alcohol addicted, well self-controlled, and less impulsive persons are not responsive to changes in these subjective beliefs. An explanation for this pattern relates to differences in costs of being pulled over by police and having an accident across groups. Our data show that persons who are alcohol addicted and lack self-control over alcohol consumption report higher subjective probabilities of being convicted for drunk driving conditional on being pulled over. Moreover, alcohol-addicted and more impulsive persons were substantially more likely to report that being arrested for drunk driving would almost ruin their lives than their non-addicted and less impulsive counterparts were. A plausible conjecture is that given past encounters with the law, these individuals are more sensitive to an additional arrest for driving while intoxicated, as criminal penalty severity tends to increase with past criminal records. For example, punishments for drunk driving recidivism are harsher than for first offenses.  

  • In the second part of our study, we found no deterrent effects of harsher sanction for drunk driving on individuals’ drinking and driving behaviours.

The survey randomly assigned criminal penalties and conditional on a criminal penalty selected for the respondent, elicited his/her probability of drinking and driving during the following year. The severity of sanctions here refers to the amount of fine and the length of sentence, license suspension, SCRAM use, and mandatory alcohol counselling. Non-response to harsher punishment severity for driving while intoxicated offenses may reflect that severity is insufficiently high or that people are irrational. But we found that such non-response more likely reflects the fact that the probability of being detected or punished for driving while intoxicated is very small so that increased sanctions generate no statistically significant effects on deterrence. According to our data, the objective probability of a drunk driving conviction conditional on drinking and driving is only 0.006. In general, an arrest on a charge of drunk driving is relatively unlikely unless an accident actually happens (Benson 2000). Drunk drivers are more likely to be escorted or driven home than arrested by police.

Concluding Remarks

Although our results suggest deterrent effects of risk perceptions of being pulled over and being involved in an accident on individuals’ drinking and driving behaviours, the perceived risk of being pulled over is substantially higher than the corresponding actual risk. Therefore, programmes designed to manipulate beliefs of being pulled over would be an inappropriate way for reducing driving under the influence of alcohol. However, being informed of the actual increase in risk of having an accident may be a more promising approach. Ninety-three percent of respondents reported an increase in the accident probability of lower than 100% when driving intoxicated compared with driving while sober. Levitt and Porter (2001) concluded that drivers with alcohol involvement, but not necessarily legally drunk, pose a risk of seven times greater than sober drinkers to cause a fatal crash; drivers above the blood-alcohol limit of 0.10 are 13 times more likely to be the cause of fatal crashes. These results suggest that information campaigns may seek to influence drivers’ perceptions of accident risks to reduce drunk driving violations.2 

Harsher sanctions for drinking and driving do not deter drinking and driving when drivers’ perceived risk of being penalised for this behaviour is low. Therefore, rather than increasing sanction severity, increasing the probability of detection and/or conviction may be more effective in reducing drinking and driving.


Bergen, G, R A Shults, and R A Rudd (2011), “Vital signs: alcohol-impaired driving among adults--United States, 2010”, Morbidity and Mortality Weekly Report 60, no. 39: 1351-1356.

Benson, B L, B D Mast, and D W Rasmussen (2000), “Can police deter drunk driving?” Applied Economics 32, no. 3: 357-366.

Chen, Y, and F Sloan (2014), “Subjective Beliefs, Deterrence, and the Propensity to Drive While Intoxicated”, NBER working paper No. w20680.

Lacey, J H, T Kelley-Baker, D Furr-Holden, R B Voas, E Romano, P Torres, A S Tippetts, A Ramirez, K Brainard, and A Berning (2009), “National Roadside Survey of Alcohol and Drug Use by Drivers: Alcohol Results”, No. HS-811 248.

Levitt, S D, and J Porter (2001), “How dangerous are drinking drivers?” Journal of Political Economy 109, no. 6: 1198-1237.

National Center for Statistics and Analysis (2012), “Alcohol-impaired driving”, In Traffic Safety Facts: 2010 Data, Washington, DC: US Department of Transportation, National Highway Traffic Safety Administration.


[1] The eight cities were: Raleigh, North Carolina (NC); Hickory, NC; Seattle, Washington (WA); Yakima, WA; Philadelphia, Pennsylvania (PA); Wilkes-Barre, PA; Milwaukee, Wisconsin (WI); and La Crosse, WI. The states and cities were selected to yield a broad geographic spread and variety of drinking and driving prevalence, DWI prevention laws and demographic composition.

[2] Although one caveat here is that since we are identifying across-individual variation in perceived risks of being involved in accidents, it is not absolutely clear that the within-person change in this information would cause the same behavior change in drinking and driving.

2,310 Reads