In what looked like a return to the dark days of Reefer Madness – a 1936 film that portrayed cannabis as a drug that could drive people to madness and murder – a recent article in the Daily Telegraph claimed that cannabis users are “twice as likely to be violent”.

The claim comes from a systematic review by King’s College London that combined 63 studies covering more than 265,000 people. The researchers did, indeed, find a link between cannabis use and violence, but the headline overstates what the numbers can tell us.

There are three reasons to be cautious: other factors that could explain the link (known as confounders), the way the studies were designed, and potential biases in how the data was gathered.

Poverty, alcohol, other drugs and a difficult upbringing are all linked to violence in their own right. Cannabis is also illegal in many places, and buying it can bring people into contact with people and places where violence is more common. Some of these factors are more common among cannabis users, so these confounders could partly explain the link.

Ice cream sales rise and fall alongside drownings, yet nobody thinks ice cream causes drowning. Hot weather makes people eat more ice cream and swim more, linking the two. Cannabis and violence may be connected in a similarly roundabout way.

The review did try to account for other factors. But its authors generally favoured the versions of studies that adjusted for the fewest of these factors, making it harder to rule out alternative explanations.

Study design

Of the 63 studies in the review, 46 were cross-sectional, meaning they measured cannabis use and violence at one point in time. This makes it difficult to tell whether cannabis use came before the violence, followed it, or simply occurred alongside other circumstances that contributed to it. These studies found the strongest link between cannabis use and violence.

Studies that follow people over time – called longitudinal studies – can help establish which came first. These studies found a much weaker link between cannabis use and violence than the cross-sectional studies.

In the one-off studies, the odds of violence were more than twice as high among cannabis users. The Telegraph presented these figures as meaning cannabis users were up to twice as likely to be violent. In the longitudinal studies, cannabis users had 17% higher odds of violence than non-users.

It should also be noted that twice the odds is not the same as twice as likely. Likelihood looks at your chance out of the entire group. Odds compares the people an event happens to against those it doesn’t.

The two are close when violence is rare, but they drift apart when it is common.

For example, if one in 100 people experience something, doubling the likelihood gives you two in 100 (2%). Doubling the odds gives you two in 101 (1.98%) – a tiny distinction.

The gap widens a lot when an event is common. Imagine something that already happens to 50 out of 100 people. Twice as likely doubles that 50% chance to 100% (it happens to everyone). Twice the odds only increases that 50% chance to 67%.

One of the review’s authors told the Telegraph that a study of this size lets researchers “speak with more confidence about the risks”. That is true only in a narrow sense. Imagine measuring a room with a slightly stretched tape measure: a thousand repeat measurements might give a consistent answer, but the wrong one.

A large sample reduces random error – the chance that researchers happen to study an unusual group of people. It cannot fix systematic bias, where the original studies measured the wrong things or measured them badly.

Bias can creep in at every stage. Studies disagree on what counts as cannabis use (once, in the past month or daily?) and violence (a shove, a fight or a criminal conviction?). The review groups together domestic incidents, sexual assault, fights and robbery, which may have very different causes.

Results across the 63 studies also varied widely. Three analyses of the same US national drug survey, all included in the review, produced odds ratios ranging from about 1.3 to nearly 10. This may reflect differences in how cannabis use and violence were defined and measured, as well as in who was surveyed.

Also, some studies may be missing because research showing little or no link is less likely to be published. This could make the link between cannabis and violence look stronger than it is. The review found signs of this problem, and its attempts to account for it could not rule it out.

Finally, the review’s supplementary material rates the certainty of the evidence as very low. In other words, the authors have very little confidence in the size of their estimate, though they consider the association itself likely to be real. This important qualification does not appear in the Telegraph report.

None of this means cannabis and violence are unrelated, and the authors themselves acknowledge many of these limitations. But the headline figure leaves readers with a number that looks much more certain than the evidence warrants.