Causation vs. Correlation

The fundamental distinction between two variables that MOVE TOGETHER (correlation) and one variable actually PRODUCING change in another (causation). Ice cream sales and drowning deaths are correlated (both increase in summer) but ice cream doesn't cause drowning—a third variable (warm weather) causes both. The distinction matters enormously: policies based on correlation without causation can be useless or counterproductive. The mechanisms that establish causation: controlled experiments (randomized controlled trials), natural experiments, instrumental variables, difference-in-differences, and regression discontinuity. Absent these: you have correlation, which is informative but NOT causal evidence.

When to use it

When correlation is being mistaken for causation in analysis or decision-making; when policy or strategy is being built on correlational evidence; when understanding why 'studies show X is associated with Y' doesn't mean X causes Y; when the causal mechanism needs to be established before acting on observed patterns.

How it can help

Before concluding that X causes Y: check for the three requirements of causation. (1) Correlation: X and Y move together. (2) Temporal precedence: X occurs before Y. (3) No confounding: no third variable Z causes both X and Y. The most common error: observing correlation and assuming causation without checking for confounders. The practice: for every observed correlation, ask 'what third variable could be causing both?' If you can identify plausible confounders: the correlation is suggestive but not causal. Only controlled experiments (or their statistical approximations) establish causation confidently.