Correlation Does Not Imply Causation

Correlation alone does not identify a causal effect. Associations can reflect causation, reverse direction, shared causes, selection, measurement, or chance, and causal conclusions require an appropriate design and assumptions.

An association can be consistent with a causal effect, reverse direction, shared causes, selection, measurement processes, or chance. The slogan means that association alone does not identify an intervention effect; it does not mean associations are useless evidence. Define what changing one variable would do to another before asking which design can support that claim.

Causal inference uses a design and assumptions that connect observed comparisons to counterfactual outcomes. Randomization can help establish exchangeability for assigned treatments, while observational approaches require other defensible conditions. Merely listing possible confounders, finding a plausible mechanism, or controlling for many variables does not automatically establish causation.

When to use it

When causal claims are being made from correlational evidence; when the four alternative explanations need systematic checking; when business or policy decisions are being based on observed correlations; when the mechanism behind a correlation needs specification before action.

How it can help

Define the intervention and comparison, investigate alternative causal structures, and evaluate whether the evidence identifies the effect. Treat adjustment and plausible stories as parts of an argument rather than automatic proof.