Ice cream sales and shark attacks are strongly correlated, but ice cream doesn't cause sharks. Explore confounding variables, generate spurious correlations, and learn when correlation hints at causation.
Both ice cream sales and shark attacks spike every year at the same time. Are they causally linked?
Pearson r
0.99
very strong positive
Correlation
Two variables move together, but neither necessarily causes the other.
Ice Cream Sales ↔ Shark Attacks
Bidirectional: no causal claim
Causation
One variable directly causes changes in another via a mechanism.
Temperature → Ice Cream Sales
Directional: mechanism exists
Key rule: Correlation does not imply causation. And causation does not even guarantee a visible correlation: opposing pathways or nonlinear effects can cancel the linear signal.
Common Cause (Confounder)
Z→X, Z→Y
A third variable Z causes both X and Y, creating a spurious correlation.
Reverse Causation
Y→X (not X→Y)
You assume X→Y but actually Y→X. Classic in economics and medicine.
Mediation
X→Z→Y
X causes Z which causes Y. The effect is indirect, but still causal.
Which causal structure best explains the ice cream / shark correlation?
Real data, zero causal meaning, and a generator to find your own "surprising" correlations.
Famous Real Spurious Correlations
Cheese Consumption vs Bedsheet Deaths
After Tyler Vigen, Spurious Correlations (data approximated)
Bedsheet entanglement deaths
Nicolas Cage Films vs Pool Drownings
After Tyler Vigen, Spurious Correlations (data approximated)
Pool drowning deaths
Spelling Bee Word Length vs Spider Deaths
After Tyler Vigen, Spurious Correlations (data approximated)
Spider-related deaths
Random Correlation Generator
Two completely unrelated random variables
Generated
0
With enough random pairs, you'll find high correlations by chance alone. This is called p-hacking or the multiple comparisons problem. The correlation is mathematically real; the causal meaning is not.
Click each card to expand and see the hidden confounding variable.
Each card uses a small illustrative dataset that sketches the real-world pattern; the r shown is computed from the plotted points, not taken from a published study.