Extreme measurements tend to be followed by less extreme ones on retest, even without any real change.
When a variable is measured and an unusually high or low value is observed, the next measurement in the same person or group will tend to be closer to the average, simply due to natural variability and measurement error rather than any true underlying change. This is a major confound in clinical practice and research: patients are often selected for treatment precisely because they present at a symptom peak (e.g., worst pain), so improvement afterward may partly or wholly reflect this statistical artefact rather than treatment effect. Well-designed studies control for it using randomised comparison groups; its magnitude depends on how extreme the initial measurement was and how much random variability the measure has.
This guide was auto-drafted and is pending editorial review.