Effect Size in Mediation Analysis

• Activity Number: 296
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Elizabeth Newton*+
• Companies: Institute for Aging Research
• Keywords: effect size ; sample size ; mediation analysis

Abstract

In mediation analysis, many measures of effect size have been proposed, but each has drawbacks. Researchers often regard a 10% reduction in the regression coefficient for the initial predictor as evidence of mediation, thus focusing on proportion mediated. This has intuitive appeal, but also several shortcomings. Effect size measures based on differences in or products of regression coefficients or R-squared measures are less interpretable, however. This study examines the proportion mediated effect size measure. Simulations, with bootstrapping of confidence intervals are used to investigate the sample size required to achieve a given power to detect a 10% change in the coefficient. In addition, a formula for sample size based on a z-test is derived. Applications to real data are discussed as well.