In summer, more ice cream is sold and there are more drownings. The two numbers rise together — but nobody would say ice cream causes drownings. This simple example hides one of the most expensive mistakes in data analysis: confusing correlation with causation.
What correlation is
Correlation means two variables move together: when one rises, the other tends to rise (or fall). It is a statistical pattern, useful for spotting relationships — but it does not say why they move together nor whether one causes the other.
What causation is
Causation is stronger: A causes B. Changing A changes B. This is what we really want to know in order to act — whether lowering the price increases sales, whether training reduces errors. But proving causation is far harder than observing correlation.
The hidden third factor
In the ice cream case, the common cause is heat: summer drives both up. This is called a confounding variable — something behind the scenes that moves both and creates the illusion of a direct link. Much of misleading correlation hides a third factor like this.
Why this costs money
- Wrong decisions: investing in a factor that was only correlated, not causing the result.
- False conclusions: "channel X brings the best customers" — when maybe the best customers seek out channel X.
- Overconfidence: a pretty chart with two lines rising together looks like proof, but it is not.
How not to fall into the trap
Be suspicious of convenient correlations, look for the third factor, and when you really need to know whether A causes B, run a controlled experiment (an A/B test). Testing is the safest way to move from "they go together" to "one causes the other".
In practice
Next time you see two numbers moving together, resist the easy conclusion. Ask: does one cause the other, or is there something behind both? That pause avoids expensive decisions based on statistical illusions. How many of the "causes" you assume are, in fact, just correlations?
