Correlation and Causation

Foundation · 12 min

Changing together is not causing

Correlation means two variables move together: when one rises and the other rises or falls with it, they are correlated. The classic example: in periods when ice cream sales rise, drowning incidents rise too. Ice cream does not cause drowning; a third factor lifts both — summer. In hot weather people eat more ice cream and swim more. The correlation is real; the causation is an interpretation, and a wrong one.

Correlation itself is a valuable finding; most research begins with a correlation being noticed. The problem lies in joint change being presented, on its own, as proof of cause and effect.

Three fillers of the gap

Between the observation “X and Y change together” and the verdict “X causes Y”, three alternatives always stand:

• A third variable: a common factor drives both — like the season in the ice cream and drowning example.
• Reverse causation: the arrow points the other way. “Confident people talk a lot” can also be read as talking a lot building confidence.
• Chance: compare enough variables and some entirely unrelated pairs will appear to move together by coincidence.

If a claimed link cannot close all three of these doors, a causal verdict is premature.

An example from the sky: birth-month relationships

From time to time, relationships are observed between birth month and certain traits — school performance, or representation in some sports. Is this evidence for the influence of star signs? Looking closer, earthly explanations come to the fore: school entry cut-off dates create nearly a year's difference in maturity between classmates; in sports with age-group categories, those born early in the year are bigger than their peers; seasons affect nutrition and daylight. None of these mechanisms needs the positions of the stars. When two explanations fit the same data, the one that shows a testable mechanism takes priority — a principle you may recall from the scientific claim lesson. This example is worth keeping as a template: whenever a relationship appears between a sky variable and a human trait, check the earthly consequences of the calendar first — school terms, seasons, age groups.

How is causation established?

For a causal verdict, researchers look for several supports: deliberately changing one variable and observing the other (experiment), making comparison groups alike (control), a plausible account of how the effect works (mechanism), and the result repeating across different data sets. In daily life we cannot run experiments; but we can at least build this habit: whenever we hear of a link, list three candidates for “what else would explain this?”. More often than not, one of the candidates is more convincing than the headline.

Sometimes you cannot find three candidates; that can be a sign the causal claim is strong. The point of the exercise is not to reject every relationship, but to scan the alternatives before passing judgement. The habit is a small mental exercise of a few seconds.