Why correlation isn't causation
Ice cream sales and drownings both rise in summer, but ice cream doesn't cause drowning—both are driven by hot weather, a confounding variable. There's also reverse causation (people who see doctors more are sicker, not sicker because of doctors) and coincidence, like the famous link between national chocolate consumption and Nobel Prizes.
How to establish causation
The strongest tool is the randomized controlled trial. When experiments aren't possible, researchers use regression with controls, natural experiments, or difference-in-differences. Epidemiologist Austin Bradford Hill proposed criteria such as strength, consistency, temporality, and dose-response. The coefficient r captures only linear association, and statistical significance (p-values) doesn't prove causation either.
FAQ
- Does r = 0 mean no relationship?
- Only no linear relationship; a U-shaped one could exist.
- Are observational studies useless?
- No—converging evidence, as with smoking and lung cancer, can support causal conclusions.
Sources reviewed: 2026-10-26
References
- Pearl, J., & Mackenzie, D. (2018). The Book of Why. Basic Books.
- Hill, A. B. (1965). The environment and disease: Association or causation? Proceedings of the Royal Society of Medicine, 58(5), 295–300.
- Messerli, F. H. (2012). Chocolate consumption, cognitive function, and Nobel laureates. New England Journal of Medicine, 367(16), 1562–1564.