Can correlation ever prove causation on its own?

No. Correlation from observational data alone can never prove causation, because it cannot rule out reverse causation or a confounding variable. It can, however, be combined with a well-established causal mechanism, controlled experiments, or statistical techniques that explicitly control for known confounders to build a credible causal case.

What is a confounding variable?

A confounding variable is a third factor that influences both variables being studied, creating an association between them that has no direct causal link. The classic textbook example is ice cream sales and drowning deaths, which correlate strongly because hot summer weather independently increases both — not because ice cream causes drownings.

Why do randomized controlled experiments help establish causation?

Random assignment of subjects to treatment and control groups spreads both known and unknown confounding factors roughly evenly across the groups purely by chance. If the outcome still differs meaningfully between groups afterward, confounding has been statistically ruled out as the explanation, leaving the treatment itself as the remaining explanation.

Is a negative correlation less meaningful than a positive one?

No — the sign only indicates direction (variables moving oppositely vs. together), not strength or causal validity. A correlation of −0.85 is just as strong an association as +0.85; both require the same scrutiny before any causal claim is drawn from them.

What does "correlation does not imply causation" actually mean in practice?

It means observing that two variables move together is not, by itself, sufficient justification to conclude that changing one will change the other. Before acting on a correlation — redesigning a process, changing a policy, adjusting a control input — an engineer should look for a plausible mechanism, rule out reverse causation and known confounders, and ideally test the relationship with a controlled experiment before treating it as causal.

← STEM Learning Studio

Concept Explainer · STEM

Correlation vs. Causation

Two variables can move together perfectly and still have nothing to do with each other — a third, hidden variable is often driving both.

Correlation is a purely statistical statement: two variables tend to rise and fall together (positive correlation) or move in opposite directions (negative correlation), measured by a correlation coefficient r between −1 and +1. Causationis a much stronger claim: changing one variable directly produces a change in the other. Correlation is necessary for a causal relationship to show up in data, but it is nowhere near sufficient — data alone, without an experimental design or a mechanism, cannot distinguish "A causes B" from "B causes A," from "C causes both A and B," from pure coincidence.