Why Correlation Does Not Imply Causation
What You'll Learn
You'll understand why two variables moving together doesn't prove one causes the other, and learn to recognize this critical mistake in data analysis.
The Core Problem
When two variables correlate—meaning they move together in a predictable pattern—it's tempting to conclude that one causes the other. But correlation simply measures association, not causation.
Think of it like this: Ice cream sales and drowning deaths are highly correlated. Both peak in summer. Does ice cream cause drowning? Of course not. The hidden variable is warm weather, which causes both people to buy ice cream and to go swimming.
Classic Examples That Reveal the Trap
1. The Confounding Variable
- Shoe size correlates with reading ability in children
- Shoe size doesn't cause literacy—age causes both
2. Reverse Causation
- Does coffee cause stress, or do stressed people drink more coffee?
- Correlation alone can't tell you the direction
3. Pure Coincidence
- Nicolas Cage movies released per year correlates with swimming pool drownings
- Sometimes patterns emerge by pure chance, especially with many variables
Why This Matters in Data Science
Mistaking correlation for causation leads to:
- Wrong business decisions (launching ineffective interventions)
- Flawed predictions (models based on spurious relationships)
- Misleading reports (claiming impact where there is none)
To establish causation, you need controlled experiments, temporal ordering (cause before effect), theoretical mechanisms, and methods from causal inference—which build on correlation but go much further.
Key Takeaway: Correlation tells you two variables dance together; causation tells you which one is leading. Never assume correlation alone reveals cause and effect—always look for confounders, reverse causation, and alternative explanations.