Correlation and Causation
The statistical relationship between variables versus the proof that one causes the other, and why correlation alone cannot establish causation.
Correlation measures the strength and direction of the linear relationship between two variables. The most common measure is Pearson's r, which ranges from −1 (perfect negative relationship) through 0 (no linear relationship) to +1 (perfect positive relationship). Height and weight are positively correlated; hours of sleep and daytime sleepiness negatively. Correlation summarizes association, and nothing more: two variables can be perfectly correlated while neither causes the other.
The slogan correlation does not imply causation is a warning about inference, not a mathematical fact. The classic traps are confounding and reverse causation. A third variable may drive both: ice-cream sales and drowning deaths rise together in summer, not because ice cream causes drowning, but because warm weather causes both. Or the causal direction may be the reverse of what intuition suggests: exercise and health correlate, but the direction of the effect — and its magnitude — cannot be read off the correlation coefficient. Spurious correlations can also arise by chance, especially when many pairs are examined, and time series that both trend upward correlate strongly for no causal reason at all.
Establishing causation requires more than observation. The gold standard is the randomized controlled experiment: randomly assigning subjects to treatment and control groups breaks confounding, because randomization makes the groups comparable in expectation. When experiments are impossible (economics, epidemiology, astronomy), researchers use quasi-experimental designs, natural experiments, longitudinal studies with temporal ordering, and statistical controls — regression adjustment, matching, instrumental variables, difference-in-differences — each with assumptions that must be argued, not assumed. Temporal precedence (the cause must precede the effect) and a plausible mechanism are necessary but not sufficient evidence.
Quantitatively, correlation is connected to the concepts of standard deviation: r is the covariance of two variables normalized by the product of their standard deviations, so it is scale-free and dimensionless. Correlation also depends on the range of data observed — restricting to a narrow subpopulation weakens correlation (restriction of range). In the era of big data, the discipline of separating association from causation is more important than ever: predictive models can exploit correlation without understanding it, but interventions, policies, and explanations require causal knowledge.
Tags
causation correlation data analysis statistics