Bradford Hill Criteria

We repeat the mantra: "Correlation is not causation." But smoking correlates with lung cancer. Does it cause it?

In 1965, Sir Austin Bradford Hill gave us a checklist to decide.

The Checklist

1. Strength: A huge correlation (smokers are 20x more likely to get cancer) is more likely to be causal than a weak one.
2. Consistency: Do different studies in different places find the same thing?
3. Specificity: Does the cause lead to a specific effect, or just general "bad health"?
4. Temporality: Did the cause happen before the effect? (This is the only mandatory one).
5. Biological Gradient: Does more exposure lead to more effect? (Dose-response).
6. Plausibility: Is there a biological mechanism that explains it?
7. Coherence: Does it fit with what we already know?
8. Experiment: Does removing the cause stop the effect?
9. Analogy: Is it similar to other known causes?

The Diagnosis

You don't need all 9. But the more you have, the stronger your case.

If a study claims causation but fails the Bradford Hill test, it is just a correlation.