Resolution of a Regression Paradox in Pretest‐Posttest Designs

Regression to the mean (RTM) of individuals is the tendency for extreme individuals to become less extreme on remeasurement; RTM of group means describes this same tendency among group means. Under the classical test model, in pretest‐posttest designs where subjects are selected on the basis of extreme values at the pretest, RTM of group means will always occur for the attribute used to select extreme subjects. For other attributes, however, RTM of group means requires a positive correlation between that attribute's measurement error and the measurement error of the attribute used in the selection. Thus, while all attributes will evidence RTM of individuals, extreme groups do not always regress to the mean. RTM depends most fundamentally on the magnitude of the pretest measurement error. Copyright © 1992, Wiley Blackwell. All rights reserved

Duke Scholars

Author Gregory P. Samsa Biostatistics & Bioinformatics, Division of Biostatistics