A statistical measure describing the degree and direction of association or dependence between two variables measured in a sample or population; correlation quantifies how values of one variable vary with another but, by itself, does not establish that variation in one variable causes variation in the other.

Principle

Correlation indicates statistical dependence and potential predictive utility but is insufficient for causal inference; it can arise from direct causation, reverse causation, confounding, shared trends, or measurement structure and therefore requires further analysis to interpret mechanisms.

Demonstration

Illustrative scenario (hypothetical): Situation — A dataset shows that variable X and variable Y tend to increase together. Recognition — Analysts compute a correlation coefficient or fit a model showing statistical association. Action — The association is used as a basis for further investigation (e.g., stratified analyses, temporal studies, or predictive modelling). Consequence — The correlation may guide hypothesis generation or prediction, but cannot by itself justify causal claims or interventions without additional evidence.

Misapplication

Mistaken interpretation: concluding that correlation proves causation. Why plausible: strong or consistent correlation suggests a relationship. Semantic error: conflating statistical association with causal effect and thereby endorsing interventions or mechanistic explanations that the data do not support.

Consequence

Correlation is useful for describing relationships, selecting predictors, and generating hypotheses; misinterpreting correlation as causal can lead to poor decisions, inappropriate interventions and misleading inferences.

Reversal

A correlation can become evidence for causation only after complementary designs and analyses (temporality, control for confounding, mechanism) support a causal interpretation; conversely, correlation may disappear after conditioning on a confounder or using different measurement scales.

Boundary

Clearly within: statistically significant association between two measured variables in a specified dataset, quantified by correlation coefficient or dependence measure. Boundary case: correlation in cross‑sectional data that may reflect reverse causation or confounding. Clearly outside: causal claim that one variable produces change in the other without supporting causal evidence.

Semantic Tension

Tension between descriptive statistical objectives (measuring and summarizing association) and inferential causal objectives (establishing causes), which demand different methods and standards of evidence.

Synthesis

Correlation is an informative descriptive and predictive tool but a weak basis for causal claims; treating it as such requires explicit additional evidence and sensitivity to alternative explanations.