Activity Number:
|
662
- Methods for Meta-Analysis, and Longitudinal and Clustered Data
|
Type:
|
Contributed
|
Date/Time:
|
Thursday, August 1, 2019 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Statistics in Epidemiology
|
Abstract #307129
|
Presentation
|
Title:
|
Quantification and Estimation of the Regression to the Mean for Bivariate Distributions
|
Author(s):
|
Manzoor Khan* and Jake Olivier
|
Companies:
|
University of New South Wales and University of New South Wales
|
Keywords:
|
Regression to the mean;
Bivariate distributions;
Intervention effects;
Treatment effects
|
Abstract:
|
The phenomena of regression to the mean occurs when subjects having unusually large or small values are remeasured and found closer to the population mean. Not accounting for regression to the mean can adversely influence the conclusion of a pre/post study design. Expressions for quantifying and estimating regression to the mean are only available when the pre/post variables follow the bivariate normal, Poisson and binomial distributions. The total observed effect is the expected difference in pre/post random variables and can be decomposed into regression to the mean and treatment effects. Expressions for quantifying the total observed effect, regression to the mean and treatment effects are derived for any bivariate distribution. These formulae allow for unbiased estimation of the treatment effect when paired observations are influenced by regression to the mean. Maximum likelihood estimates are derived, and the unbiasedness, consistency and normality of these estimators are theoretically established where possible. The total observed change in paediatric blood lead levels from baseline to follow-up was estimated and decomposed into regression to the mean and treatment effects.
|
Authors who are presenting talks have a * after their name.