regression to the mean: a standalone construct

Francis Galton introduced the idea in his 1886 study Regression towards mediocrity in hereditary stature, where he found that unusually tall or short parents tended to have children whose heights fell closer to the population average, and he named that pattern regression. The modern statistical account, set out in reviews such as Barnett, van der Pols, and Dobson (2005), explains that whenever two measurements are imperfectly correlated, a value that is extreme on the first measurement will on average be less extreme on the second, because part of any extreme reading reflects chance or measurement error that is unlikely to repeat. It is a property of the measurement process itself and does not require any real change in the thing being measured, which is why it can be mistaken for a genuine effect when a group is chosen precisely because it scored very high or very low. The construct is a landmark because it accounts for a wide class of apparent effects across disciplines, from the illusory improvement of the sickest patients after any intervention to the so-called sophomore slump in sports, using only the statistics of correlation and variation.[1][2][3]

Unit

A statistical regularity in repeated or paired measurements, stated as the expected movement of extreme values toward the mean

Recovery means

Not applicable

Measured by

Quantified from the correlation between two measurements: in standardised units the expected value on the second measurement is the correlation times the value on the first, so the expected shift toward the mean is one minus the correlation times the initial distance from the mean