Recreational Marijuana Legalization and Business Formation: Evidence from Staggered State Adoption

Abstract

Does recreational marijuana legalization affect entrepreneurial activity? I exploit the staggered opening of recreational marijuana retail sales across 21 U.S.\ states from 2014 to 2024 to estimate the causal effect on new business formation using the Census Bureau's Business Formation Statistics. Using the heterogeneity-robust estimator as the primary specification, I find a Callaway-Sant'Anna overall ATT of $-0.028$ log points (SE $= 0.029$, 95% CI: $[-0.085, 0.029]$), indicating a modest but imprecisely estimated decline in business applications per capita. Conventional TWFE estimates are larger ($-0.068$ log points, 95% CI: $[-0.148, 0.012]$), consistent with heterogeneity bias under staggered adoption. Event-study estimates reveal no evidence of differential pre-trends and suggest effects emerge gradually, reaching $-0.15$ to $-0.20$ log points by 6–7 years after retail opening. A descriptive analysis of actual business formations (BF8Q) shows a positive association, though this cannot be interpreted causally because BF8Q is forward-looking and mechanically spans post-treatment periods for pre-treatment cohorts. To address potential spillover contamination of the control group, I show that results are robust when restricting controls to "interior" states that do not border any treated state ($\text{ATT} = -0.042$, SE $= 0.034$). Results are further robust to randomization inference ($p = 0.093$), pairs cluster bootstrap ($p = 0.064$), medical-marijuana-only controls, and excluding COVID-era observations.