The Journal of the American Academy of Child & Adolescent Psychiatry (JAACAP) just published a paper that found that new marijuana legalization laws correlated with increases in suicide among youth. Although the authors likely did their math correctly, they may have chosen a modeling strategy that produced inaccurate results. The results also don’t explain how marijuana laws are connected to these suicide rate increases among youth when youth marijuana use did not increase after marijuana markets were established at the state level.
In the new publication, “Association Between Marijuana Laws and Suicide Among 12- to 25-Year-Olds in the United States From 2000 to 2019”, Christopher J. Hammond, J. Madison Hyer, Anne E. Boustead, Mary A. Fristad, Danielle L. Steelesmith, Guy N. Brock, Deborah S. Hasin, and Cynthia A. Fontanella compared suicide mortality data to state-level marijuana legalization efforts, such as permitting regulated medical and recreational markets. The authors found that recreational marijuana laws were associated with a 9% increase in suicide rates among all youth ages 14 to 16. They also found that medical and recreational laws were associated with 10% and 16% increases in suicide rates among female youth ages 12 to 25, respectively.
My preprint study with Robert Capodilupo, Michael Schemenaur, and Jeffrey A. Singer found the opposite results among similar age groups—marijuana laws were associated with a decline in suicide rates. So why the difference?
The difference is due to technical concerns of statistical analysis. Concisely, the authors chose an inappropriate model given the data distribution. The methods in the Hammond study exaggerate the effect of marijuana laws on suicide rates in states with relatively few suicides. States that had not yet or never did legalize marijuana were more likely to have near-zero suicide rates, because suicide rates have increased across all states over time. Because of the model the authors selected, the JAACAP study shows that states with regulated marijuana markets have higher suicide rates, regardless of whether suicides have risen at faster rates in those states after their marijuana laws went into effect.
The new JAACAP study employed a negative binomial regression, which assumes that the plurality of states had zero suicides to begin with and then examines changes in the number of suicides in each state. In reality, suicides occur in every state, and the average state-level suicide rate is approximately 14 deaths per 100,000 population. Therefore, state-level suicide rates don’t follow a negative binomial distribution.
For example, consider the distributions of state-level suicide rates among males ages 15 to 19 in Figure 1 below, which we analyzed in our study. The raw data (on the left) almost follow a normal distribution in which observations are closely grouped around a median value. A negative binomial distribution, however, assumes the modal observation is a zero value. Many statistical models exist to help researchers match their approach to the data and, in theory, the distribution curve associated with any model should closely fit the shape of the data. In Figure 1, we superimpose the data shape implied by the two models over the data as a red line. In our preferred model (on the right), we have taken a natural log of the data and then treated it as a normal distribution. The data fits the normal distribution shape implied by the red line much better than the model used by the Hammond team. This means that a basic linear regression on log-transformed data is more appropriate for this type of analysis, lending greater credibility to our analysis and conclusions.
After transforming the data and controlling for various sociodemographic factors, such as personal income and unemployment, we found that suicide rates on average generally drop among female and male youth after states allow both medical and recreational marijuana access, although these results weren’t statistically significant for all of our robustness checks. Therefore, suicide rates tend to drop with greater marijuana access, but we can’t conclude that the policies themselves were the cause of the drop. However, suicide rates do predictively drop about 5.4% for males ages 30 to 39 following medical marijuana access, which drove reductions in the total male suicide rate (table of our full results). To be clear, our results are all associations and should not necessarily be interpreted as causal, but the data show that suicide rates don’t increase after state laws allow for greater marijuana access.
Actual marijuana use by teens hasn’t changed with legalization, another strike against the Hammond study’s conclusion. If marijuana legalization led to changes in suicide rates among youth, you’d expect it to be because youth are using marijuana more often. However, our analysis also shows that rates of marijuana use among youth remained stable after states adopted recreational marijuana markets (Figure 2).
Inappropriate model selection is why conflicting results appear throughout the public health literature. In the case of suicide rates, our study rigorously shows that they tend to drop among youth following both recreational and medical marijuana access, which supports the original findings by Anderson et al. (2014) in the American Journal of Public Health.