Estimates of the prevalence and risk of drinking-and-driving are a high-priority need for researchers and policymakers. Levitt and Porter (Journal of Political Economy, 2001, 109(6), 1198-1237) demonstrate how these can be recovered using publicly available information in the Fatality Analysis Reporting System (FARS). Although robust to systematic misreporting and sample selection and far cheaper to implement than surveys, their methodological innovations are largely ignored. We believe this arises partly from difficulty in replicating their results. This article identifies the underlying causes of replication failure and offers practical guidance for future implementation that takes advantage of the current structure of the FARS data.
