Correlation vs Causation
That two things move together does not mean one drives the other; mistaking association for cause is the most common inferential error.
Association Is Not Mechanism
A correlation is a statistical association: two quantities tend to move together. A causal relationship means changing one changes the other. Correlation can arise from causation in either direction, from a common cause driving both, or from pure coincidence. The data alone rarely tell you which.
Why Correlations Deceive
- Confounding: a hidden third variable drives both, creating a spurious link.
- Reverse causation: the effect is mistaken for the cause.
- Selection: the way data were gathered manufactures the association.
- Coincidence: with enough variables, some will correlate by chance.
The Confounding Problem
Ice-cream sales correlate with drownings, but neither causes the other; hot weather drives both. This is confounding, and it is everywhere. Observing that two things co-occur, even strongly and repeatedly, does not license the claim that intervening on one will change the other. That claim requires a causal argument, not just a statistical one.
How Causation Is Established
Causal claims come from intervention or from careful causal reasoning: randomized experiments that break confounding by design, or observational methods that explicitly model and adjust for confounders under stated assumptions. Prediction can rest on correlation; decisions to intervene cannot, because an intervention tests the causal claim directly.
In Modeling
A model that predicts well by exploiting correlations can fail catastrophically when used to guide an intervention that breaks those correlations. Distinguishing predictive models from causal ones, and knowing which a decision needs, is essential. A pattern that holds in observed data is not a lever until its causal status is established.