Correlation vs Causation — How to Tell Them Apart

Two words, one picture

A correlation is a pattern: when one number is high, the other tends to be high, or tends to be low. Causation is a mechanism: changing one thing changes the other, and there is a path you can point to. The ice cream example works because the pattern is real — the two lines genuinely move together, month after month — and the mechanism is still absent. Nothing about a cone puts anyone underwater. A third thing, summer heat, pushes both lines at once: more people buy ice cream, and more people swim. Every misleading statistic you will ever meet is some version of this picture with the third thing hidden.

The three traps that make them look alike

Almost every case where a correlation gets mistaken for a cause falls into one of three traps.

• A third variable moves both. Heat drives ice cream and swimming; wealth drives both private tutoring and exam results. The two visible things are connected only through the hidden one.
• The cause runs backwards. Fires with more fire engines at the scene do more damage — not because engines cause damage, but because bigger fires summon more engines. The direction of the arrow is the opposite of the first impression.
• It is a coincidence. Compare enough pairs of numbers and some will line up beautifully by pure chance, especially over short periods with few data points. A striking match between two curves means little until you ask how many curves were tried.

Three questions that tell them apart

When a claim says A causes B, run it through three questions in order.

1. Did A actually come before B? A cause has to precede its effect. If the data cannot establish which moved first, it cannot establish direction.
2. What else could move both? Name candidates deliberately — season, income, age, temperature, time itself. If a plausible third variable exists and was not accounted for, the correlation cannot carry the causal claim on its own.
3. What happens when someone intervenes? This is the decisive one. In an experiment, researchers change A on purpose for a randomly chosen group and leave a control group alone. Random assignment cuts the link to every third variable at once, which is why a controlled experiment can support a causal claim and a pattern found in existing data mostly cannot.

Reading the news with these questions

Headlines are written in the language of correlation and read in the language of causation. Linked to, associated with, tied to — each of these means the two things moved together in somebody's data, and nothing more. When you meet one, look for two facts in the article. First, was it an experiment or an observation? If researchers watched existing groups rather than assigning people at random, third variables are still in play, however large the study. Second, is this one study or many? A single striking result is where the coincidence trap lives; the findings that survive are the ones that keep showing up when other teams repeat the work. The replication questions in our psychology trivia quiz cover exactly why single studies mislead, and the methods round of the science quiz asks the control-group question in several forms.

What correlation is still good for

None of this makes correlation worthless — it makes it a tool for a different job. If you only need to predict, a stable correlation is enough: umbrella sales predict rain-related delays without causing rain, and no one is harmed by the confusion. The trouble starts only when you intervene. Buying umbrellas will not delay a flight, and banning ice cream will not prevent a single drowning, because intervening acts on the cause, not on the pattern. So the practical rule is short: correlations are for forecasting, causes are for deciding. Before acting on a pattern, you need the answer to question three — what happens when someone changes the thing on purpose — and not just a tidy pair of curves.

Where to practice

The distinction sticks faster when you meet it as questions rather than definitions. The science quiz has a full round on how we know things — control groups, correlation and causation, blinding, and why a theory is not a guess. The psychology trivia quiz devotes its last round to replication: publication bias, small samples, and famous results that vanished on a second look. And the environment quiz is practice for the magnitude half of the skill — its final round asks what actually changes how much, which is the question causal claims are ultimately for. None of them requires sign-up, and each answer explains itself, so a wrong guess teaches more than a right one.