High on Employment? A Spatial Difference-in-Discontinuities Analysis of Marijuana Legalization and Industry-Specific Labor Market Effects

Abstract

Does recreational marijuana legalization affect local labor markets? I exploit the retail openings in Colorado and Washington (2014) using a spatial difference-in-discontinuities (DiDisc) design that compares counties on opposite sides of state borders before and after legalization. Using Census Quarterly Workforce Indicators at the county-quarter-industry level, I find no significant aggregate effect on new hire earnings within 100km of treated borders ($\hat{\tau} = -3.1%$, SE = 6.2%, 95% CI: $[-15.1%, 9.0%]$). Crucially, temporal placebo tests validate the design: all eight pre-treatment discontinuity changes are centered at zero, supporting the identifying assumption. Industry heterogeneity analysis with Benjamini-Hochberg FDR correction reveals one significant result: the information sector shows a $-13.0%$ decline in earnings (FDR-adjusted $q = 0.03$), while tourism-exposed accommodation and food services shows a marginally significant $+5.5%$ increase. Contrary to theoretical predictions, agriculture and retail show null effects. The spatial design addresses concerns about policy endogeneity that plague state-level difference-in-differences, while administrative data captures the exact population—new hires—affected by labor market changes. Results suggest marijuana legalization has minimal direct effects on labor market equilibria at state borders, with heterogeneous industry-specific effects that partially support theoretical predictions about safety-sensitive and tourism-exposed sectors.