What is the difference between correlation and causation

Correlation is a statistical metric that quantifies the strength and direction of a linear relationship between two variables, typically expressed by the Pearson coefficient $r$. A value of $r=0.8$ indicates a strong positive association, but it says nothing about which variable, if any, causes the other to change. Causation, on the other hand, asserts a directional influence: a change in variable X produces a change in variable Y, often demonstrated through experimental control or rigorous longitudinal analysis.

Why the confusion arises

People often mistake correlation for causation because the human brain looks for patterns and explanations. When two trends rise together, it feels natural to assume one drives the other, yet many hidden variables can create the illusion of a direct link. For example, ice cream sales and drowning incidents both increase in summer, but the true driver is temperature, not a causal relationship between the two.

Common sources of mistaken causation:

• Confounding variables that affect both measures
• Reverse causality where Y influences X
• Coincidental patterns in small samples
• Selection bias in data collection

How to test for causation using regression:

1. 1Formulate a clear hypothesis about directionality
2. 2Control for potential confounders by adding them as covariates
3. 3Check the statistical significance of the regression coefficient (p‑value)
4. 4Validate results with out‑of‑sample or experimental data

Sample data illustrating correlation vs. regression slope:

Worked example: Suppose we collect five observations of X and Y as shown in the table. The Pearson correlation computes to $r = 1.0$ because Y = 2X exactly, indicating perfect linear association. Running a simple linear regression yields $Y = 2X + 0$ with a slope of 2 and a p‑value far below 0.05, supporting a causal interpretation if the experimental design justifies it. If, however, we added a hidden variable Z that also influences Y, the regression slope might become non‑significant despite the high $r$, signaling that the apparent link is spurious.