Life and Death in the Fast Lane: Police Enforcement and Traffic Fatalities

Abstract

Simultaneity complicates the estimation of the causal effect of police on crime. We overcome this obstacle by focusing on a mass layoff of Oregon State Police in February of 2003. Due solely to budget cuts, 35 percent of the roadway troopers were laid off, which dramatically reduced citations. The subsequent decrease in enforcement is associated with a significant increase in injuries and fatalities. The effects are similar using control groups chosen either geographically or through data-driven methods. Our estimates suggest that a highway fatality can be prevented with $309,000 of expenditures on state police. (JEL H76, K42, R41)

The causal effect of wrong-hand drive vehicles on road safety

“…I employ the synthetic control method developed by Abadie and Gardeazabal (2003) and Abadie et al (2010Abadie et al ( , 2015 in a rather new fashion; OLS and WLS difference-in-differences estimates as proposed by DeAngelo and Hansen (2014) validate the results as I will show later on. The basic idea of the synthetic control method is to construct a synthetic counterfactual for a treated unit by matching on non-treatment period trends of the outcome variable of interest.…”

“…As a downside, however, this procedure, may result in a violated common trend assumption. Therefore, I conduct a second set of estimations combining difference-in-differences with the synthetic control group as proposed by DeAngelo and Hansen (2014). The basic idea is to use donor pool weights derived by the synthetic control group as observation weights in a WLS difference-in-differences regression.…”

Punishment and Deterrence: Evidence from Drunk Driving

“…The main results are based on a local-linear regression discontinuity design with a rectangular kernel, while the sensitivity of the results are tested (…nding little to no major di¤erences) using local linear models with other kernels or higher order polynomials. 14 An indicator for either a DUI or aggravated DUI indicates respectively whether the BAC falls above the .08 or .15 thresholds. In the regression models the BAC variable is rescaled around the relevant threshold, either .08 or .15, X i is a vector of controls, and y i is a measure of recidivism.…”

The US Crime Puzzle: A Comparative Perspective on US Crime & Punishment

“…All of the cited cross-country studies control for the level of development, usually operationalized as GDP per capita. 31 In fact, the impact of development is so fundamental that presumably most other theories implicitly hold the level of development constant, and I control for it in all of the regressions. 32 The level of development a¤ects the opportunity set of potential criminals, and the institutional capacity of public law enforcement.…”

“…In that view, societies that support the poor with generous welfare spending, and that support employees with protective labor regulation, are also likely to employ only moderate punishment. 33 I control for income inequality using the Gini coe¢ cient, and for labor regulation 31 Some authors, such as Neapolitan (2001), use instead the Human Development Index, which combines GDP per capita, life expectancy, and educational achievement; I used it in some regressions with identical results. Other authors, such as Soares (2004), separately include the level of education.…”