Spurious correlation

A spurious correlation is a numerically strong relationship between two variables that does not reflect any real causal or explanatory link between them. It arises from coincidence, a shared trend such as time or population growth that both variables happen to track, or a hidden confounding variable that drives both.

Spurious correlations are easy to manufacture: given enough variables, some pairs will correlate strongly purely by chance, especially over a short or trending time series. Ice cream sales and drowning deaths both rise in summer because of heat, not because either causes the other. The more comparisons made, the more spurious results appear, which is the same underlying issue behind the multiple comparisons problem.

Spurious correlation matters because it is a routine trap in dashboards and ad hoc analysis: a chart showing two lines moving together is persuasive, but it is not evidence of a causal link. The safeguard is to ask whether there is a plausible mechanism connecting the variables, to check for an obvious confounder such as a shared time trend, and to require a controlled comparison or a proper causal inference method before acting on the relationship. Simply having a high correlation coefficient is never sufficient on its own.