Legal Status vs.\ Physical Access: Testing the Cannabis-Alcohol Substitution Hypothesis at State Borders

Abstract

Does access to legal cannabis reduce alcohol involvement among fatal traffic crashes through substance substitution? Using a spatial regression discontinuity design at state borders in the western United States, I test whether cannabis access affects alcohol-involved crashes. The standard RDD yields a null result (9.2 pp, SE = 5.9, p = 0.127). Suggestive evidence from dispensary distance analysis—using 2020-era location data as a proxy for the 2016–2019 study period—indicates that physical cannabis access may not change sharply at borders, as prohibition-state residents near borders can cross to purchase; however, this first-stage evidence is approximate given the temporal mismatch between dispensary data and crash outcomes. To address this weak first-stage critique, I restrict to single-vehicle crashes with in-state drivers, where crash location, driver residence, and outcome attribution are all unambiguous. This specification produces a null result: $-5.2$ pp (SE = 11.4, p = 0.649). Donut RDD robustness checks reveal some specification sensitivity—the 2km donut yields a significant positive estimate (23.7 pp, SE = 8.2)—but border-by-border decomposition shows no individual border drives this result, and larger donuts return to null. Strikingly, cross-border analysis reveals that prohibition-state residents who crash in legal states have lower alcohol involvement (21.6%) than those who crash at home (31.0%), suggesting compositional differences in who crosses borders rather than treatment effects. The baseline and most robustness specifications produce null results, though specification sensitivity warrants caution in interpretation.