State-Level Relationships Cannot Tell Us Anything About Individuals

Anderson et al.’s¹ finding that states with legalized medical marijuana had lower suicide rates among young men calls to mind the work of the famous French sociologist, Emile Durkheim. In Le Suicide,² published in 1897, Durkheim observed that suicide rates were lower in regions with a higher proportion of Catholics and concluded that social controls within the Catholic religion reduced the likelihood of Catholics taking their own lives. Durkheim’s individual-level interpretation of his region-level data subsequently became a textbook example of the “ecological fallacy”^3,4: the assumption that associations observed at a higher level of aggregation (e.g., state level) are mirrored at the individual level.

Anderson et al. provide another example of the ecological fallacy when they suggest their findings are “consistent with the hypothesis that legalizing medical marijuana leads to increased marijuana use, which in turn helps individuals cope with stressful life events.”^1(pp2369) In fact, their state-level data can tell us nothing about the relationship between individual-level medical marijuana use and suicide risk. The state-level relationship observed does not mean that using medical marijuana protects against suicide, and if the state-level correlation had been in the opposite direction, it would not mean that using medical marijuana causes suicide either.

As nonintuitive as it sounds, relationships at different levels of aggregation are not necessarily parallel and indeed can go in opposite directions. For example, geographic areas with higher levels of smoking and radon exposure have lower cancer rates,^5,6 but substantial data indicate smoking and radon put individuals at risk for cancer. We published a paper in this journal describing the ecological fallacy in the context of health care quality measures, illustrating how relationships between health care processes and outcomes can differ for groups and individuals.⁷ In Durkheim’s case, it might have been that Catholics living in Protestant majority provinces drove the observed higher suicide rates observed there. In Anderson et al.’s case, the reduced suicide rate in states with medical marijuana could have been driven by young men who did not smoke medical marijuana.

The ecological fallacy is now well known in epidemiology and sociology, but is a persistent and unfortunately common problem in other areas of science. Often we have individual-level questions but only aggregate data, so we are tempted to make a seemingly logical, but erroneous, leap. Awareness of the ecological fallacy should give us pause in these situations.