PORTFOLIO UPDATED — SEPTEMBER 23, 2026

§3.4 — Research

Medical Marijuana Laws & Crime

Replicating Gavrilova, Kamada & Zoutman (2019) on medical marijuana laws and violent crime — a real DDD re-estimation, checked against the published paper's own tables.

real DDD re-estimation · mmlBorder -131 (t=-3.25) vs. the paper's -107.984 (se=20.969) — same sign, same significance

group project with two classmates (Eamon Coffey, Will Kearney) · not a solo project

Research

1. the paper we replicated

Is Legal Pot Crippling Mexican Drug Trafficking Organisations?

Gavrilova, Kamada & Zoutman (2019), The Economic Journal, Vol. 129, Issue 617, pp. 375–407 · doi:10.1111/ecoj.12521

›the paper's theory & findings

Theory: legal marijuana grown in MML states displaces illegal Mexican imports, cutting into drug-trafficking-organization (DTO) profits and reducing the incentive for the violence that protects that trade. Tested via a difference-in-difference-in-difference (DDD): before/after MML, MML/non-MML states, near/far from the Mexican border.

  • 12.5% drop in violent crime in counties near the Mexican border after medical marijuana law (MML) adoption
  • 40.6% drop specifically in drug-law-related homicides in those counties
  • effect concentrated within ~350km of the border; no comparable effect in distant inland counties
  • inland MML states also reduce crime in nearby border states (spillover), consistent with reduced demand for DTO-supplied marijuana

2. real data & the exact regression we ran

1994–2012 · county-year panel, 59,601 control-variable observations.

›data sources & the real Stata command

Crime: FBI Uniform Crime Reports (violent crime rates) + Supplementary Homicide Reports (incident-level, including drug- and gang-related circumstances). MML dates: state legislative records + ProCon.org (MML adoption dates and provisions). Controls: U.S. Census Bureau, Bureau of Labor Statistics, Bureau of Economic Analysis.

reghdfe violent_rate mmlInland mmlBorder portionhispanic portionmale portion20_24 povertyrate incomepercapita unemployment, absorb(geofips border#year) vce(cluster statefips)

3. our replication vs. the paper's own table

mmlBorder (our replication)-131 (t=-3.25)**

n=12,033

MML × Mexico border (paper, col. 3)-107.984 (se=20.969)***
mmlInland (our replication)15.2 (t=0.72) — n.s.
MML inland (paper, across columns)1.169 to 4.111 — n.s.

conclusion

Both mmlInland and mmlBorder come out with the right sign and land close to the paper's own published magnitude — mmlBorder significant and negative in both, mmlInland small and statistically indistinguishable from zero in both. The standard errors don't match exactly, most likely from small discrepancies in county coverage or missing covariates that shift the effective number of clusters. Overall: a real, independent replication that reproduces the paper's central result, not just a citation of it.

← back to Research