Regression to the Mean
Regression to the mean is the tendency for unusually high or low measurements to be followed by values closer to their typical level when random variation contributes.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/regression-to-the-mean/
01What it means
Regression to the mean occurs when an extreme observation partly reflects transient variation, so a later measurement is likely to be less extreme even without an intervention. It is not a causal force or a guarantee that every value improves; it is a predictable source of misleading before-and-after stories.[1][2]
- 01Persistent levelUnderlying difference for the unit
- 02Transient variationNoise pushes one observation outward
- 03Select extreme caseEnrollment triggered by the unusual value
- 04Measure againNew transient variation is unlikely identical
- 05Looks closer to averageChange may occur without treatment effect
Regression to the mean is a source of misleading before-and-after stories, not a causal force or guarantee that every value improves.
02Why it matters
Product teams often target the lowest-performing cohort, feature, or period and then credit the intervention when the next observation improves. Recognizing regression to the mean protects prioritization and experimentation from false wins.
03When to use it
Consider it whenever selection is based on unusually high or low prior performance, especially in incident response, churn outreach, quality improvement, and dashboard diagnosis. Use a concurrent comparison group or repeated measurements to distinguish natural movement from impact.
04How it works
- Identify whether the population was selected for an extreme measurement.
- Separate persistent differences from transient measurement noise.
- Define a comparable untreated or historical reference where appropriate.
- Measure outcomes over a planned follow-up period.
- Interpret movement with uncertainty and alternative explanations.
05A practical example
A team offers coaching to accounts with an unusually bad support week. Their ticket volume falls the next week, but a comparison of similarly extreme accounts without coaching shows the same partial rebound, so the team does not attribute it to the program.
06Where it breaks
Regression to the mean does not explain every reversal and cannot replace causal analysis. A reference group may still differ in important ways, and operational changes can occur alongside natural variation.
Common mistakes
- Treating every rebound after an intervention as its effect.
- Selecting extreme cases and comparing only their next measurement.
- Calling regression to the mean an explanation without checking a comparison.
07Sources
- Regression toward the meanEncyclopaedia Britannica · Scholarly referenceRegression toward the mean
- Engineering Statistics HandbookNational Institute of Standards and Technology · Government statistical referenceEngineering Statistics Handbook