A Massive Study Correlating Cannabis Use with Clinical Depression and Bipolar Disorder Illustrates Why Longitudinal Studies Aren't a Silver Bullet for Establishing Causality
The key passages from Susan Pinker’s July 6, 2023 Wall Street Journal article, “Cannabis Is Linked to Mental Illness” are:
Now a new longitudinal study has examined the medical records of all citizens of Denmark over the age of 16, some 6.5 million people in all, for patterns of diagnosis, hospitalization and treatment for substance use between 1995 and 2021. In the paper, published in the journal JAMA Psychiatry in May, Dr. Oskar Hougaard Jefsen of Aarhus University and colleagues showed that people who had previously been diagnosed with cannabis use disorder were almost twice as likely to be diagnosed later with clinical depression.
and
Though the association was strong, the authors note that they can’t say for certain whether chronic and heavy cannabis use induces psychosis, or whether people prone to mental illness are more likely to be heavy users. It makes sense that people who feel the symptoms of incapacitating depression or mania, or who sense apparitions or voices only they can hear, might try to self-medicate with cannabis.
Having a psychiatric disorder diagnosed after someone is determined to be addicted to marijuana doesn’t mean the marijuana use caused it. The beginnings of the psychiatric disorder were likely there before the diagnosis of it, and those beginnings could have made marijuana quite attractive as self-medication. I also wonder if people who are connected enough to the mental health system to get an addiction diagnosis aren’t more likely to get diagnosed for psychiatric disorders they have.
How then, can we establish causality? In her article, Susan Pinker writes:
Without a randomized controlled trial, which would be unethical in the extreme, it’s hard to untangle these strands definitively.
I don’t think that is quite true. First of all, a randomized controlled trial wouldn’t be unethical if the intervention was something that lowered marijuana use rather than raised it. One simply needs to get a brilliant idea of how to get a randomly chosen half of one’s sample to reduce their marijuana use.
Also, in this case, different states legalizing marijuana at different times provides a reasonably good natural experiment for differences-in-differences estimation that assumes the trend lines in each state for depression and bipolar disorder would have had an unchanging slope if it weren’t for marijuana legalization at different dates. Which states legalize isn’t random, but the known dates of legalization can help identify whether marijuana legalization had an effect on those mental health trends.
Further Reading: “An Example of Needing to Worry about Reverse Causality: Satisfaction with Aging and Objective Aging Outcomes” addresses a similar issue of reverse causality in a longitudinal study (that shows what happens to someone over time).