OCKHAM Ockham
Article · Bias中文
Bias · Correlation

Correlation is not causation

Two things moving together shows they are related; claiming one causes the other takes a further step.

1. Three sources of correlation

When "A and B always occur together", there are at least three explanations:

  1. A causes B — the direction you want to prove.
  2. B causes A — the arrow reverses. "People who exercise catch fewer colds" may also be "healthier people find it easier to keep exercising".
  3. C causes both — a common factor. Ice-cream sales and drowning deaths both rise with summer.

And a fourth: plain coincidence, especially with small samples or short time windows.

2. Why we jump to causation

  • Narrative preference — causal stories are more memorable than statistics, so the brain fills in a plot.
  • Temporal adjacency — things that happen one after another feel causal, but sequence is not cause.
  • Confirmation bias — if you want a conclusion, you remember the supporting cases.

3. Five criteria for a causal claim

  1. Temporal order — the cause must precede the effect.
  2. Strength and stability — does the relationship hold across samples and settings?
  3. Excluding common causes — could a third factor explain both?
  4. Dose–response — more exposure, stronger effect, makes causation more credible.
  5. Intervention — change the cause and watch whether the effect follows. This is the strongest evidence.
Any single criterion is not enough; they are thresholds to be met together.

4. Using it during inference

Whenever you write "because X, therefore Y", do three things: ask whether the arrow could reverse; ask whether a third cause exists; ask whether changing X would change Y. Answer all three and the causal claim earns its weight.

These are exactly the checks Ockham applies in its causal-reasoning step. Related: base-rate fallacy, Bayesian thinking, abductive reasoning.

Read next