Correlation vs Causation: Why Association Is Not Proof

Direct Answer

Correlation is a statistical relationship: two variables are correlated when they change together — when one increases, the other tends to increase (positive correlation) or decrease (negative correlation). Causation is a relation of production: one event or state brings another about — the cause contributes to producing the effect. The two are connected but not equivalent. Causation typically produces correlation — a genuine cause and its effect will usually be correlated — but correlation does not imply causation, because an observed association can arise without any causal link between the two variables. "Correlation does not imply causation" is therefore the first principle of empirical reasoning.

Why can correlation exist without causation? There are three classic reasons. First, confounding: a third variable may cause both. Ice cream sales and drowning deaths rise together in summer, but ice cream does not cause drowning; the warm weather causes both. Second, reverse causation: the arrow may point the other way. People who exercise more may be healthier — but healthier people may also be more able to exercise. Third, chance: with enough data, coincidental associations appear; a noisy correlation between two unrelated variables is to be expected somewhere in a large dataset. These are not exotic exceptions but the normal hazards of reading causes off associations.

Historical Context

The distinction has deep roots in the philosophical study of causation. David Hume, in the Treatise of Human Nature and the Enquiry Concerning Human Understanding, subjected the idea of causation to skeptical analysis. We observe constant conjunction — event B regularly following event A — but we never observe the "necessary connection" that supposedly links them. Causation, Hume argued, is a habit of the mind: we infer the cause from observed regularity, and this inference rests on induction, whose justification he famously called into question. Hume's analysis is the philosophical origin of the modern idea that causal claims are inferences from observed correlations — and inferences that can go wrong.

The scientific response was methodological. Francis Bacon and later John Stuart Mill codified the methods by which causes are distinguished from mere associations. Mill's canons of induction — the method of agreement, the method of difference, and the method of concomitant variation — are the logical ancestors of the modern controlled experiment: vary one factor, hold everything else constant, and observe the effect. Charles Sanders Peirce and the statisticians who followed (from Francis Galton and Karl Pearson to Ronald Fisher) transformed the study of association into a mathematical discipline: correlation coefficients measure association, and experimental design, randomization, and statistical significance test whether observed associations support causal conclusions.

Philosophical Perspectives

Philosophers have offered competing accounts of what causation itself is. Hume's regularity theory reduces causation to constant conjunction: A causes B if events of type A are always followed by events of type B. The theory struggles with the difference between genuine causation and mere regularity (night follows day, but day does not cause night) and with the possibility of causal connections that occur only once. Probabilistic accounts analyze causation in terms of probability-raising: A causes B if the probability of B is higher when A occurs. Counterfactual accounts, developed in modern form by [David Lewis], analyze causation in terms of counterfactual dependence: A causes B if, had A not occurred, B would not have occurred. Manipulationist accounts, associated with James Woodward, tie causation to intervention: A causes B if manipulating A in the right way changes B. Each account captures part of the phenomenon, and the debate continues.

The epistemological question is how causal claims can be tested. The gold standard is the randomized controlled experiment: participants are randomly assigned to treatment and control groups, so that confounders are distributed evenly between them, and any difference in outcome can be attributed to the treatment. Where experiments are impossible or unethical — in epidemiology, economics, and much of the social sciences — researchers use observational methods: matching, regression with controls, instrumental variables, natural experiments, and longitudinal designs, all of which attempt to approximate the logic of the experiment with statistical tools. The philosophy of science contributes the crucial caution: no amount of statistical adjustment can eliminate unmeasured confounders, so observational causal claims are always provisional.

Modern Reflection

The confusion of correlation with causation is one of the most common errors in modern public discourse. It appears in health headlines ("coffee drinkers live longer"), in marketing claims, in social media's tendency to present associations as discoveries, and in the reasoning of everyday life. The growth of big data has made the problem more acute: with millions of variables to correlate, spurious associations are guaranteed to appear, and the "correlation mining" of large datasets produces a steady stream of false causal narratives. Machine learning models that excel at prediction often reveal nothing about causation, and the distinction between predictive power and causal understanding has become a central issue in data science and artificial intelligence.

The corrective habits are the habits of critical thinking. When confronted with a claimed causal link, ask: Could a third variable explain the association? Could the cause run the other way? Is the effect due to chance or selection? What evidence comes from controlled experiments rather than bare correlations? The scientific disciplines — randomization, blinding, replication — are institutionalized answers to these questions, and the scientific method is, at its core, the discipline of moving from observed association to tested causation. Hume's skepticism about necessary connection remains philosophically unsettled, but the practical lesson is secure: association is evidence, never proof — and the difference between the two is the difference between belief and knowledge.

Related Thinkers

• David Hume — The skeptical analyst of causation and induction
• Francis Bacon — The champion of controlled experimental inquiry
• John Stuart Mill — The codifier of the methods of experimental inquiry
• Charles Sanders Peirce — The philosopher of abduction and statistical inference
• Immanuel Kant — The defender of causality as a priori structure
• Daniel Kahneman — The psychologist of causal illusions and fast thinking

Related Quotes

• Hume on Custom — The habitual inference from constant conjunction to cause
• Bacon on Knowledge Is Power — Knowledge disciplined by experiment
• Einstein on Questioning — The relentless questioning of empirical claims

Further Learning

• Understand the underlying inference in what is inductive reasoning
• See how experiments test causation in what is the scientific method
• Recognize the psychological dimension in what is a cognitive bias
• Read the Stanford Encyclopedia of Philosophy on probabilistic causation
• Visit the Critical Thinking & Logic collection for the full learning path

Sources

• Stanford Encyclopedia of Philosophy, "The Problem of Induction" — https://plato.stanford.edu/entries/induction-problem/
• Stanford Encyclopedia of Philosophy, "Probabilistic Causation" — https://plato.stanford.edu/entries/causation-probabilistic/
• Stanford Encyclopedia of Philosophy, "Causation and Manipulability" — https://plato.stanford.edu/entries/causation-mani/