Explaining and Controlling Regression to the Mean in Longitudinal Research Designs This tutorial is concerned with examining how regression to the mean influences research findings in longitudinal studies of clinical populations. In such studies participants are often obtained because of performance that deviates systematically from the population mean and are then subsequently studied with respect to change in the trait used ... Tutorial
Tutorial  |   December 01, 2003
Explaining and Controlling Regression to the Mean in Longitudinal Research Designs
 
Author Affiliations & Notes
  • Xuyang Zhang
    University of Iowa, Iowa City
  • J. Bruce Tomblin
    University of Iowa, Iowa City
  • Contact author: Xuyang Zhang, PhD, Department of Speech Pathology and Audiology, The University of Iowa, Iowa City, IA 52242. E-mail: xuyang-zhang@uiowa.edu
Article Information
Research Issues, Methods & Evidence-Based Practice / Language / Tutorial
Tutorial   |   December 01, 2003
Explaining and Controlling Regression to the Mean in Longitudinal Research Designs
Journal of Speech, Language, and Hearing Research, December 2003, Vol. 46, 1340-1351. doi:10.1044/1092-4388(2003/104)
History: Received March 14, 2003 , Accepted July 15, 2003
 
Journal of Speech, Language, and Hearing Research, December 2003, Vol. 46, 1340-1351. doi:10.1044/1092-4388(2003/104)
History: Received March 14, 2003; Accepted July 15, 2003
Web of Science® Times Cited: 21

This tutorial is concerned with examining how regression to the mean influences research findings in longitudinal studies of clinical populations. In such studies participants are often obtained because of performance that deviates systematically from the population mean and are then subsequently studied with respect to change in the trait used for this selection. It is shown that in such research there is a potential for the estimates of change to be erroneous due to the effect of regression to the mean. The source of the regression effect is shown to arise from measurement error and a sampling bias of this measurement error in the process of selecting on extreme scores. It is also shown that regression effects are greater with measures that are less reliable and with samples that are selected with more extreme scores. Furthermore, it is shown that regression effects are particularly prominent when measures of change are based on changes in dichotomous states formed from quantitative, normally distributed traits. In addition to a formal analysis of the regression to the mean, the features of regression to the mean are demonstrated via a simulation.

Acknowledgment
This work was supported by a clinical research center grant P0-DC-02748 from the National Institute on Deafness and Other Communication Disorders.
Order a Subscription
Pay Per View
Entire Journal of Speech, Language, and Hearing Research content & archive
24-hour access
This Article
24-hour access