Statistics for Everyday Life
Correlation vs Causation
"Correlation is not causation" is repeated so often it can feel like a slogan — but the distinction is one of the most useful ideas in all of clear thinking. Here’s what actually separates them.
| Aspect | Correlation | Causation |
|---|---|---|
| What it means | Two things tend to move together | One thing actually produces a change in the other |
| What it shows | An association or pattern | A cause-and-effect mechanism |
| Evidence needed | Observed data | A controlled experiment, or reasoning that rules out other explanations |
| Common trap | Assuming the link proves cause | Ignoring confounders or reverse causation |
| Example | Ice-cream sales and drownings rise together | Summer heat drives both — neither causes the other |
When to use correlation
Correlation is enough when you only need to predict or spot a pattern — a correlated signal can forecast without explaining why.
When to use causation
You need causation before acting to change an outcome: policies, treatments, and product changes only work if the link is genuinely causal.
Frequently asked questions
- Why doesn’t correlation prove causation?
- Because a third factor can drive both variables (a confounder), the causation can run the opposite way, or the link can be coincidence. A correlation is consistent with many explanations, only one of which is "A causes B."
- How do you actually establish causation?
- The strongest tool is a randomized controlled experiment, which breaks the link to confounders. Where that’s impossible, causation is argued by ruling out alternatives, checking that cause precedes effect, and identifying a plausible mechanism.
- What is a confounding variable?
- A confounder is a hidden third factor that influences both variables you’re comparing, creating a correlation without any direct link between them — like summer heat behind ice cream and drownings.