What does 'correlation is not causation' mean?

What does "correlation is not causation" mean?

Correlation means two things tend to happen together or change together. Causation means one thing directly produces a change in the other. The phrase "correlation is not causation" is a warning: observing a relationship between two variables does not, by itself, prove that one causes the other.

The distinction is foundational in science, but it is especially relevant when reading wellness and health claims. A study might find that people who meditate also report better sleep. That is a correlation. It does not prove meditation caused the better sleep. People who meditate may also exercise more, drink less alcohol, or have quieter evenings, any of which could explain the link.

A concrete example: ice cream and drowning

The classic illustration uses two things that clearly have no direct causal relationship. Ice cream sales and drowning incidents both rise in summer. They are strongly correlated, but buying ice cream does not cause drowning. A third factor, hot weather, drives both: more people buy ice cream and more people swim when it is warm.

This structure, two outcomes sharing a common driver, appears constantly in wellness research. A study might link a supplement with better mood, but people who take supplements may also eat better, sleep more, or have higher incomes. Those factors, not the supplement, could be doing the work.

Why the confusion happens

The human brain is pattern-seeking. When two things occur together, assuming one caused the other feels natural and satisfying. Researchers call this the correlation-causation fallacy, and it is not just a problem for casual readers. A systematic review of top-tier biomedical journals found that all 21 neuroscience papers examined, which linked central nervous system regeneration to behavioural outcomes, were subject to possible ecological fallacy¹. That fallacy involves drawing conclusions about individuals from group-level averages, a distinct but adjacent mistake that shows how easily even expert researchers slip between association and causation.

Misinformation also exploits this tendency. In a randomised controlled trial of 377 participants, a logic-focused message without a modus tollens argument reduced belief in false claims linking COVID-19 vaccines to cancer more effectively than fact-focused messages that simply presented medical findings². Understanding the logical structure of the error, not just receiving correct facts, helped people resist the misinformation.

Correlation, causation, and the insulin example

Even when researchers run experiments rather than observe correlations, causation can be more complicated than it appears. A study examining insulin and fasting glucose introduced a distinction between "driver" and "navigator" causation³. A driver is necessary to reach a destination but does not decide where that destination is. A navigator decides the destination and path but cannot drive the system there.

Using five different approaches, including systematic reviews and experiments in rats, the researchers concluded that insulin action hastens the return to a steady state after a glucose load, but does not determine the steady-state glucose level itself³. In their framing, insulin is a driver, not a navigator. The finding suggests that the current line of clinical action in type 2 diabetes may have limited success because it is based on a misinterpretation of the glucose-insulin relationship³. This analysis is not a clinical trial, so it does not change treatment guidance. But it illustrates a deeper point: even interventional studies can misread what a causal relationship actually does.

How to spot the fallacy in wellness claims

When you encounter a headline such as "people who do X have lower risk of Y", run through a short checklist.

First, ask what kind of study produced the claim. Observational research can identify correlations but cannot, on its own, establish causation. Randomised controlled trials are better suited to causal questions, though they have their own limits.

Second, ask whether a third factor could explain the link. Income, education, existing health behaviours, and access to healthcare are common confounders in wellness research.

Third, ask about direction. If a study links poor sleep with stress, did poor sleep increase stress, or did stress worsen sleep? Reverse causation is common and often untested.

Fourth, ask whether the size of the effect is meaningful. A statistically significant correlation can be so small that it has no practical importance.

For more on why single studies often overstate wellness findings, see our piece on why so many wellness claims rest on a single small study. If you want to understand why published research sometimes fails to replicate, read what the replication crisis in psychology actually is.

What to do this week

Pick one wellness claim you have seen recently, in a headline, a podcast, or a social media post. Write down the claim, then ask the four questions above. If the source is an observational study, a survey, or an expert opinion piece, treat the claim as an association, not a proven cause. If the source is a randomised trial, check whether it measured the outcome directly and whether the effect size was reported.

This habit takes a few minutes and does not require statistical training. It is the difference between absorbing a claim and understanding what kind of evidence stands behind it.