Recreational Marijuana Dispensaries and Fatal Car Crashes
When a marijuana dispensary opens nearby, do the roads get more dangerous?
Car crashes are the second leading cause of death for Americans ages 1 to 54, and they sit at the center of the debate over legal marijuana. Experiments show that marijuana impairs driving. But if people swap alcohol for marijuana, legal access could make roads safer. Which force wins is an empirical question.
Past studies have lacked the precision to settle it. Most compare whole states before and after legalization, and only a handful of states had legal sales for more than a few years. I compare zip codes within states instead.
In Six Years, Dispensaries Reached 818 Zip Codes
Zip codes that have had a licensed recreational marijuana dispensary, by month of the first opening, in the five states studied
I built a dataset of license records covering nearly every recreational dispensary in California, Colorado, Massachusetts, Oregon, and Washington, and matched it to the federal registry of every fatal crash in the United States, located by latitude and longitude. The result follows 3,357 zip codes month by month from 2005 through 2019: 604,260 zip-code months and 72,039 fatal crashes.
The staggered rollout makes the comparison possible. I compare how fatal crashes change in a zip code when its first dispensary opens with how they change in other zip codes in the same state, in the same month. Anything that hits a whole state at once, including legalization itself, other drug and traffic laws, and the economy, drops out of the comparison.
Fatal Crashes Rise After a Dispensary Opens
Before opening, zip codes that got dispensaries show no clear trend relative to the others
more fatal crashes in a zip code once a recreational dispensary is open. The 95% confidence interval runs from 1.7% to 10.9%. Across the treated zip codes, that adds up to roughly 265 additional fatal crashes over the study period.
The Estimate Holds Once Statewide Shocks Are Controlled
Accounting for statewide shocks matters most; local demographics and business activity barely move it
Is It Impairment, or Just Traffic?
A new store of any kind brings cars. If dispensaries raise crashes only because they draw traffic to a new shopping area, restricting them would not save many lives: some other business would bring the same cars. If the cause is impaired driving, policy has much more to work with. Four tests point to impairment.
First, a new pharmacy does not have this effect. Retail pharmacies are also state licensed, also open in commercial areas, and also attract shoppers. When one opens, fatal crashes do not rise meaningfully.
A New Pharmacy Does Not Raise Fatal Crashes
Effect of each kind of opening on fatal crashes, in the four states with pharmacy license data
Second, the effect is concentrated at night. In Google’s popular-times data, foot traffic at dispensaries peaks in the middle of the day. The increase in fatal crashes is concentrated in the evening and overnight, when most stores have closed: daytime crashes rise a statistically insignificant 2.9%, nighttime crashes 9.0%.
The Increase Comes at Night
Effect of an open dispensary on fatal crashes, by the time of day the crash occurred
Third, the first dispensary matters; additional ones show no clear effect. The first store in a zip code creates legal access where there was none. Later stores bring more commercial traffic but partly split existing customers. If traffic were the cause, every opening should count.
The First Dispensary Raises Crashes; Later Ones Show No Clear Effect
Effect on fatal crashes of the first dispensary in a zip code and of additional ones
Fourth, drunk-driving crashes rise too. Extra traffic to a store should not change how many crashes involve alcohol. Yet in the four states with usable alcohol data, fatal crashes in which police reported a drinking driver rise 11.3% after a dispensary opens. In a linear version of the model, these crashes account for about 26% of the total effect. Alcohol is a possible complement to marijuana, so this too fits impairment. Alcohol involvement is reported inconsistently, so this estimate deserves more caution than the others.
How Big Is 6.3%?
It falls between the effects of texting bans (4%) and mandatory seat belt laws (8%), in the opposite direction.
Comparable in Size to Major Traffic Safety Policies
Estimated percent change in traffic fatalities
Dispensary-level sales records from Washington State allow a rough translation into dollars. In Washington, opening a dispensary brings about $293,000 in monthly marijuana sales (in 2010 dollars) to a zip code. Set against the rise in crashes across all five states, that implies one additional fatal crash for roughly every $21.2 million of marijuana sold, close to the ratio for alcohol. Sales per zip code with a dispensary differ across states, so this figure is more tentative than the others.
Dispensaries have many effects, good and bad, and this paper measures one cost. Because the evidence points toward impaired driving rather than traffic, policies aimed at impaired driving may do more to reduce it than policies aimed at traffic.
Data and methods
Data. Dispensary license records for CA, CO, MA, OR, and WA, from public records and Freedom of Information requests, located by zip code. Fatal crashes from the National Highway Traffic Safety Administration’s Fatality Analysis Reporting System, 2005 to 2019, located by latitude and longitude. Zip-code demographics from the Census and American Community Survey, business activity from Data Axle, and county unemployment from the Bureau of Labor Statistics. Of 3,357 zip codes, 818 opened a dispensary during the study period and 2,539 never did.
Model. Poisson difference-in-differences of monthly fatal-crash counts on an indicator for at least one open recreational dispensary, with zip-code and state-by-month fixed effects, demographic and business controls, and a population offset. Standard errors are clustered by zip code. As in the paper, coefficients are reported as approximate percent changes, and the intervals drawn here are the estimate plus or minus 1.96 standard errors.
Robustness. An ordinary least squares version gives 0.0138 more fatal crashes per zip-code month. Following de Chaisemartin and D’Haultfœuille, 12 of the 28,114 underlying comparisons carry negative weights, all from one zip code; dropping it leaves the estimate at 6.2%.
Limitations. Zip codes reduce aggregation bias but do not eliminate it, and confounding trends at the zip-code level cannot be fully ruled out. Spillovers into neighboring zip codes would bias the main estimate toward zero; the distance model, which allows for them, also finds an increase.