WEEDMAPS, an app that allows cannabis users to find sellers and review their wares, advertises its services through “Weedfacts”—marijuana-promoting factoids on bus stops and billboards in Washington, DC, where recreational use has been decriminalised. The signs proclaim that the legalisation of cannabis improves property values and decreases teenage use and crime. Harmless advertising perhaps. But some of its signs also allege that cannabis helps protect against opioid addiction, which is killing nearly 50,000 Americans a year. “Cannabis has been shown to reduce opioid deaths 25%”, says one poster. If that claim seems too good to be true, that is probably because it is.

The idea that cannabis legalisation—for both medical and recreational purposes—could seriously make a dent in America’s opioid crisis is common. This month, the state health commissioner of New York issued emergency regulations allowing anyone with an opioid addiction to obtain medical marijuana, calling it “a critical step in combating the deadly opioid epidemic affecting people across the state”. Illinois has introduced a similar programme. Yet the evidence behind the theory that legalising marijuana can help combat opioid addiction is thin. And the consequences of bad policy could be severe.

The chief evidence for the theory is a paper published in the Journal of the American Medical Association (JAMA). Analysing the relationship between opioid overdose deaths and medical marijuana laws from 1999 to 2010, the authors found a strong negative association between the two. “States with medical cannabis laws had a 24.8% lower mean annual opioid overdose mortality rate compared with states without medical cannabis laws,” they wrote. They took pains to describe this as correlational and not causal. But their findings were not interpreted so carefully by cannabis advocates and some policymakers.

A recent paper by Chelsea Shover, an epidemiologist and a postdoctoral fellow at Stanford University, showed that this was a mistake. By simply extending the study period up to 2017—the time period during which the opioid crisis accelerated—she shows that the association between opioid deaths and medical marijuana legalisation flipped from strongly negative to strongly positive. Rather than the 25 percentage-point reduction in the opioid overdose mortality rate cited by cannabis promoters, the study shows a 23 percentage-point increase between 1999 and 2017. “One way to interpret that would be causally: To say that cannabis was saving people then, and is killing people now. But we don’t believe that,” says Ms Shover. “That’s one reason why it’s fraught to make policy decisions based on this data.”

The flawed interpretations of the original paper fell afoul of what statisticians call the ecological fallacy, when an inference is made about an individual based on statistical data for a group. Poorer states may be more likely to vote for Republicans, but that does not mean that poorer people are more likely to vote for the party, for example. There are very few individual-level studies, either observational or clinical, that would allow researchers to support the claim that access to marijuana limits addiction to opioids or successfully treats those recovering from addictions. The only individual-level analysis to date, a nationally representative observation study of 57,000 Americans, found that medical cannabis users were more likely to misuse prescription painkillers. This does not mean that marijuana-decriminalisation laws worsen opioid addiction—only that a lot more research is needed before policymakers make up their minds.

There is danger in haste. Opioid addiction is a chronic disorder characterised by repeated relapses. It is exceptionally hard to kick. People who receive medically-assisted treatment—generally methadone, buprenorphine or naltrexone—are much less likely to relapse than those who try mere abstention. But only one in five Americans with an opioid use disorder—the medical term for addiction—are currently receiving treatment. If people were to errantly substitute cannabis for one of the proven treatments, it could cause greater rates of relapse and more overdose deaths. “The suggestion that patients should self-substitute a drug (ie, cannabis) that has not been subjected to a single clinical trial for opioid addiction is irresponsible,” wrote Keith Humphreys and Richard Saitz, two experts on addiction in JAMA.

This does not mean that decriminalising marijuana is a bad idea. Cannabis is probably not the panacea advocates claim, but it does not have the deleterious effects of other banned drugs like heroin or cocaine. There is solid evidence for the medical benefits of CBD, a non-psychoactive cannabis compound, for some forms of epilepsy. And enough patients report that cannabis relieves their pain for such claims to be taken seriously. But combating a crisis as urgent and widespread as the opioid epidemic requires responses that are proven to be effective. These include drastically increasing medically-assisted treatment, increasing the distribution of naloxone (which reverses fatal overdoses), and reducing the prescription of pain killers. Until these responses are prioritised, it is hard to justify turning to speculative—and possibly spurious—options.