Lies, Damns Lies and Statistics

The marijuana lobby’s media spokesmen Westword and Merry Jane were quick to laud trauma surgeon Jayson Aydelotte’s August 2017 report from the University of Texas – Austin. The report in the American Journal of Public Health studied fatal vehicle crashes in several states. They found that in Colorado and Washington – states where selling weed online for recreational use is allowed – fatal crashes were not statistically different from similar states without recreational marijuana.

Although the statistical tools used were proper, readers should avoid inferring too much from the report, since it did not study the impact of marijuana use on driving safety. Instead, it looked at all collisions, not just collisions caused by impaired drivers. A multitude of other studies have already convincingly shown that marijuana can and does cause driving impairment, crashes, and death.

“Total fatal crashes” is a very blunt tool to measure the impact of marijuana legalization on traffic fatalities. Marijuana involvement in fatal crashes is a small portion of the total, and marijuana-only impaired drivers is an even smaller portion. For example, in Colorado the percentage of drivers in fatal crashes testing positive for cannabinoids rose from 9% in 2009 to 20% in 2016 according to the same FARS database used in the Texas report. In 2016, 36% of the cannabinoid-positive drivers tested positive for only cannabinoids. By measuring total fatal crashes, even a doubling in traffic fatalities caused by increased marijuana consumption would be diluted by all the other causes of traffic fatalities. 

And by at least one measure, that doubling may actually have occurred in Colorado. FARS data show that deaths where the driver tested positive for cannabinoids more than doubled from 55 deaths in 2013 to 123 deaths in 2016. Be aware that this is an inferential, rather than a conclusive indictment of marijuana’s impact on road safety. FARS was never designed to adequately capture drugged driving information. The National Highway Traffic Safety Administration that administers FARS cautions against many drugged driving inferences based on FARS data. Nevertheless, FARS data are routinely used to assess drugged driving in Colorado and Washington because neither state measures and reports drugged driving data. 

Aydelotte followed the same mistake made in an earlier publication in the same journal by different authors who measured total fatalities as a function of medical marijuana laws. Just because a statistical study can be done, doesn’t mean that it makes any sense.

Aydelotte’s study captured FARS data from 2009 through 2015, and chose 2012 as the cutoff date to define pre/post legalization. But the cutoff date is arbitrary and not particularly useful for two reasons. First, most changes in marijuana use in Colorado began in 2009 and 2010 when commercialization was enabled by the issuance of the Ogden memo. Second, although 2012 was the date of Colorado’s legalization vote, legal availability did not begin until 2014, a point the authors acknowledged.

Aydelotte et al. found that since legalization, the fatality rate change per year rose in Colorado and Washington by 0.3 fatalities per Billion Vehicle Miles Traveled (BVMT), whereas the rate change per year dropped in the comparison states by 0.8, a difference of 1.1 fatalities/BVMT. Since the comparison states weren’t identical to Colorado and Washington in many respects, the authors applied 9 adjustment factors to their raw data that lowered the net difference by over 80% to 0.2 fatalities/BVMT.

After factoring in their adjustments, the authors estimated Colorado and Washington had 77 excess crash fatalities over nearly 38 million person-years of exposure. They commented, “We do not view that as a clinically significant effect, but others might disagree.”

Presumably the 77 “excess crash fatalities” would disagree, had they survived.


DUID Articles