What is the difference between correlation and causation?

Correlation is a statistical relationship: as one variable changes, another tends to change in a predictable direction. Causation is a claim about mechanism, that changing one produces a change in the other. Correlation is evidence consistent with causation and is not sufficient to establish it, because several other explanations produce the same statistical pattern. Confounding is the most common: a third variable causes both, as when ice cream sales correlate with drowning deaths because summer causes both. Reverse causation is the second, where the arrow points the other way. And coincidence is real, since comparing enough variable pairs guarantees striking correlations with no relationship at all.

What does establish causation. Randomized controlled experiments are the strongest tool, because randomly assigning the treatment breaks any link between the treatment and confounding variables, so a difference in outcomes is attributable to the intervention. Where experiments are impossible or unethical, researchers use natural experiments, instrumental variables, difference-in-differences, and regression discontinuity designs, all of which attempt to approximate random assignment from observational data. Observational evidence can support causal conclusions when it is strong, consistent, dose-dependent, temporally correct, and mechanistically plausible, which is essentially the framework that established smoking as a cause of lung cancer without a randomized trial.

Where this matters practically. Health and nutrition headlines are frequently built on observational studies reporting correlations, and the population that eats more of some food also differs in income, exercise, and healthcare access in ways that are difficult to fully adjust for. When you read that something is linked or associated with an outcome, that language is usually signaling correlation rather than causation, and the useful questions are whether the study was randomized, how large the effect is in absolute rather than relative terms, and whether it has replicated.