Legal Weed, Self-Made? Does Recreational Marijuana Access at Age 21 Shift Workers Into Self-Employment? A Difference-in-Discontinuities Analysis

Abstract

We investigate whether legal recreational marijuana access at age 21 shifts workers from traditional employment into self-employment. The theoretical mechanism is straightforward: while Colorado legalized recreational marijuana for adults 21 and older in 2014, employers retain the right to terminate workers for off-duty marijuana use (upheld in Coats v. Dish Network, 2015). This creates a trade-off where workers who wish to consume marijuana legally may prefer self-employment, where no employer can enforce drug-testing policies. Using American Community Survey data from 2015-2022 and a difference-in-discontinuities design comparing Colorado to six control states where recreational marijuana remained illegal, we find no statistically robust evidence that marijuana legalization affects self-employment. Point estimates are positive—1.05 pp for overall self-employment, 0.97 pp for incorporated self-employment—but with only 7 state clusters (1 treated, 6 controls), conventional standard errors are unreliable. Wild cluster bootstrap inference yields p-values of 0.42 and 0.26, respectively, far above conventional significance thresholds. Permutation inference produces similar null results ($p = 0.86$). Our methodological contribution is demonstrating how point estimates that appear "highly significant" under conventional clustering ($p < 0.001$) can be entirely attributable to noise under appropriate inference methods. The positive point estimates are consistent with theoretical predictions but cannot be distinguished from chance. Future research with more treated states or alternative identification strategies is needed to test this mechanism credibly. \bigskip Keywords: Marijuana legalization, self-employment, regression discontinuity, labor supply, Colorado Amendment 64, inference with few clusters, wild cluster bootstrap