Regression to the Mean

Regression to the mean occurs when observations selected for extreme values on one measure have less extreme expected values on a related, imperfectly correlated measure under appropriate conditions. It concerns a conditional average, not an inevitable change in every individual case.

Regression to the mean describes how observations selected for an extreme value on one measure can have less extreme expected values on a related, imperfectly correlated measure. Under a simple standardized linear model, an initial score two standard deviations above average predicts a follow-up one standard deviation above average when the correlation is one-half.

The prediction concerns a conditional average, not a promise that every observation moves closer to the mean. Nor does it imply that exceptional ability disappears. Use repeated measurements and appropriate population comparisons to avoid extrapolating an extreme score as if it contained only stable information. Trends and changing conditions can alter the relationship, so state the model being used.

When to use it

When projecting from an unusually high or low observation, evaluating selected top performers, or interpreting change after selection on an extreme measurement.

How it can help

Use repeated evidence and the relevant measurement relationship to avoid extrapolating selected extremes or attributing ordinary change to an intervention.