Abstract
Regression toward the mean effect (tendency of extreme data to resemble their mean when measurements are repeated) threatened internal validity when, in a group of extreme subjects, we attempted to test whether pre/post difference in values is due to our intervention. We are still una ware how and how much it affects: a) as a function of selection value, and b) statistical testing. This study demonstrates detection, measurement and correction of the effect. Its objective was achieved through a simulation of extreme groups, measured before and after a null treatment. The effect was measured by comparing simulated with assumed sample distribution. Correction was made by calculating the value of the alpha of the testing of the null hypothesis needed to cancel out the effect.
