Perhaps you've heard of a recently released study on the effect of medical marijuana on traffic fatalities.  The study by Mark Anderson and Daniel Rees is presented as a discussion paper posted on the IZA website.  It is not peer reviewed.  What is astounding is that the authors find that Medical Marijuana Laws (MML) reduce alcohol related fatalities, and in fact reduces overall deaths by 9%.  This startling finding, they explain, is the result of marijuana replacing alcohol use among young, high risk drivers.  They argue that either marijuana is safer, or is used in safer contexts than alcohol, and therefore this replacement is protective.

There are very few studies on the effect of MML, either on general consumption or on traffic consequences.  Indeed this may be the first that has attempted to document a traffic safety effect.  It uses some sophisticated analyses from state-level data sets to make the case that MML is protective.  However, in spite of its novelty, and respected data sources, caution needs to be practiced before accepting their results. Peer reviewed articles often have undergone substantial revisions; other experts use fresh eyes to examine the methods and interpretation of results during the peer review process.  This paper has not undergone this, and indeed a careful look at the paper indicates that some of the findings may be spurious.

Before I launch into a critique, however, it is only fair to acknowledge that had they found that MML yields increased traffic fatalities, I'd probably read their paper with a much less critical eye.  But indeed, it does not take long to find serious questions about the analysis methodology; serious concerns that actually could entirely account for their findings.

Really, there are two concerns: state selection and lack of controls.  The first reflects their initial effort to demonstrate that MML predict increase marijuana use.  They do this by selecting only three MML states and then compare their change in marijuana use rates to a set of neighboring states.  They find that for MT and RI, MLL appears to predict increases.  But the authors do not explain how they choose only MT, RI and VT for this analysis.  It is not at all clear that these three states are representative of all 16 states and the District of Columbia with MML.  But at least for this part of their analyses use comparison states. 

For their main analysis of fatal crashes, there are no non-MML states included.  Instead they look for before-after MML changes in crashes.  They attempt to control for other reasons for possible crash decreases, such as the adoption of zero-tolerance DUI laws, but statistical controls can never be trusted for accounting for all of the confounding factors that can be at play.  So, it's highly likely that other trends can account for the effects they found. Indeed, because states always moved from no MML to having MML in their sample, MML is confounded with the passage of time.  So any gains in DUI prevention above and beyond the handful included in their statistical model get credited to MML.  That seems very unlikely to be a fair assessment of what is going on.  Indeed an indication of this problem  comes from their own results; without any control variables the authors find that MML accounts for a 22% decline in fatalities, but with their handful of covariates included, that effect drops down to a 9.7%  decline.  Presumably with a complete set of controls (something that is usually impossible to achieve), MML could easily drop to zero or below.

So this means... The jury is very much still out on the effect of MML laws on traffic safety. We know from other research (see Grotenhermen et al., 2007 for a review) that marijuana use poses substantial risk for road safety.  While it is reasonable to assume that alcohol, which is often served at locations that require vehicles to access, could cause far more carnage on the road, the belief that marijuana is measurably replacing alcohol to such an extent that it could produce detectable reductions seems far fetched.  Of course, if replicated using methods that address the general trend downward in alcohol related crashes, it could point to the need to rethink marijuana policy.  But that rethinking can wait for now.

UPDATE:

It appears that an independent review of this study by Kevin Sabet, Ph.D. (an adviser to ONDCP) and CADCA comes to very similar conclusions as I have, giving me more confidence in my assessment that the study is deeply flawed.

References:

Anderson, D. M., & Daniel Rees. (2011). Medical Marijuana Laws, Traffic Fatalities, and Alcohol Consumption (Discussion Paper No. 6112) (p. 46). IZA. Retrieved from http://www.iza.org/en/webcontent/publications/papers/viewAbstract?dp_id=6112

Grotenhermen, F., Leson, G., Berghaus, G., Drummer, O. H., Krüger, H.-P., Longo, M., Moskowitz, H., et al. (2007). Developing limits for driving under cannabis. Addiction, 102(12), 1910-1917.