Anderson et al. report an analysis of the association between state medical marijuana laws and suicide rates in the United States over the period 1990 to 2007.1 They found that medical marijuana legalization was associated with reduction in suicide risk for men, with a 10.9% decline in risk for men aged 20 to 29 and 30 to 39 years. Unfortunately, there are several reasons to believe that this work produced a biased estimate of this association.

The approach used by Anderson et al. is powerful because it controls for unobserved time-invariant factors through methods known as “fixed-effects” or “differences in differences regression.”2However, confounding can still occur as a result of factors that change over time and are correlated with both state policy and suicide risk. In our own recent work, we replicated and extended these results using more recent data (1990–2010) and a more extensive set of covariates.3 We identified several time-varying factors that are important to consider when analyzing the impact of medical marijuana policy on suicide and other public health outcomes. For example, Anderson et al. did not control for race/ethnicity, and state that legalized medical marijuana during this period tended to have growing minority populations.3 This is important because Blacks and Hispanics have lower suicide risk than Whites.4 Another confounding factor is tobacco control policy—other work from our group suggests that state cigarette excise taxes and smoke-free air policies are associated with lower suicide rates,5 a result that is supported by a growing body of literature exploring both the epidemiology and biological plausibility of a link between smoking and suicide.6–9 As states legalized medical marijuana, they also tended to adopt stricter tobacco control policies.3 After adjusting for these factors, as well as other demographic and state variables, we determined that the relative risk ratio describing the association between medical marijuana legalization and suicide among men was 0.996 (95% confidence interval = 0.951, 1.043; P = .87). There were no significant or even suggestive protective associations for either gender or for any of the gender-by-age groups examined by Anderson et al.3

We conclude that adoption of medical marijuana policies between 1990 and 2010 was correlated with important, time-varying demographic and political characteristics of states. Lack of control for these factors biased the estimate from Anderson et al. toward a protective association. Future research should address these factors and additionally explore other potential confounders.