Untitled

Abstract

We construct and document a novel integrated dataset combining fatal traffic crashes in Western US states from the Fatality Analysis Reporting System (FARS) with OpenStreetMap road network attributes and marijuana legalization policy timing. The resulting dataset of approximately 140,000 crashes spanning 2001–2019 (of which 96% have valid geocoding) enables unprecedented granularity in studying the geography of impaired driving. The continuous annual coverage—including the critical 2012–2015 period when Colorado, Washington, Oregon, and Alaska legalized recreational marijuana—supports event study designs for crash counts and alcohol involvement that were previously infeasible. (THC detection requires text-based matching available only from 2018 onward.) We document three key patterns: (1) among fatal crashes with any drug record in 2018–2019, the share with THC detected is approximately 19% in legalized states versus approximately 10% in comparison states; (2) THC detection rates show visible discontinuities at several state borders, with patterns varying across border pairs (motivating spatial RDD designs); (3) alcohol involvement exhibits a secular decline from approximately 40% in the early 2000s to under 30% in recent years. Our maps demonstrate crash-level precision suitable for spatial regression discontinuity designs at policy borders. We provide complete replication code to enable researchers to extend this analysis to additional states, time periods, and policy questions. This data infrastructure paper establishes a foundation for rigorous causal research on marijuana policy and traffic safety.