A statistical phenomenon in which an extreme observed value on a variable subject to random variation is likely, on a subsequent measurement, to be closer to the population mean absent any new causal influence; it reflects conditional expectation under noise, not a causal corrective process.

Principle

Given measurements composed of a stable signal plus random error or a fluctuating component, an unusually high or low observed value has a conditional expectation closer to the true mean on re-measurement; the expected change depends on the relative sizes of signal and noise (reliability).

Demonstration

Illustrative scenario → A clinician measures a patient's blood pressure during an unusually high reading due to temporary stress; without intervention a repeat measurement later is expected, on average, to be nearer the patient's long-run average blood pressure, illustrating regression toward the mean. (Illustrative scenario.)

Misapplication

Interpreting an observed return toward average as proof that an intervention caused improvement (or worsening) when no control for natural regression or measurement error is used; the error lies in attributing statistical expectation to causal effect.

Consequence

Recognition of regression to the mean necessitates control groups, baseline adjustment and repeated measurements in study design to avoid spurious claims of treatment effect; failure to account for it can produce overestimates of efficacy or underestimates of harm.

Reversal

Does not apply when observed changes reflect true systematic change in the underlying signal (e.g., effective treatment, learning, disease progression) or when selection is based on a variable that itself causally changes afterward; in such cases observed movement away from the mean may be causal, not regression.

Boundary

Clearly within: repeated measurements of inherently noisy biological markers where selection occurs on an extreme observed value. Boundary case: moderately reliable measures with partial noise—regression present but attenuated. Clearly outside: deterministic trends, interventions with immediate causal effects, or selection on post-treatment variables.

Semantic Tension

Statistical expectation versus causal inference: the tendency to treat an expected statistical artefact as evidence of causal change competes with the need to detect genuine causal effects in practice.

Synthesis

Regression to the mean is an expected statistical consequence of measurement variability; properly distinguishing it from causal change requires study designs with controls, repeated measures and explicit modelling of reliability.