Quantile Mediation Models: Methods for Assessing Mediation Across the Response Distribution

• Activity Number: 491
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ernest Shen*+ and Kiros Berhane and Chih-Ping Chou and Mary Ann Pentz
• Companies: University of Southern California and University of Southern California and University of Southern California and University of Southern California
• Keywords: Mediation ; Quantile Regression ; Longitudinal Data ; Latent Variables

Abstract

The traditional mean-based approach to mediation analysis may not suffice to capture potentially different mediating effects of risk factors across the distribution of outcomes. Three methods for quantile mediation are proposed, and compared with Imai's Average Causal Mediation Effect (ACME) and traditional approaches via a simulation study. The first is an extension of Baron and Kenny's Causal Steps Method, the second a variant of Amemiya's two-stage estimator (Fitted Value), and the third a fully Bayesian model. The methods are used to analyze data from the Healthy Places, a trial examining the effects of the built environment, specifically smart growth community planning, on resident obesity risk, for which factors like physical activity are possible mediators. Results indicate that the quantile-based Causal Steps, Bayesian, and ACME models perform equally well, while the Fitted Value does not, and that restricting the examination of mediation to the mean of the response distribution provides an incomplete picture of mediational relationships to outcomes. Extensions of the models to allow for multiple, possibly latent, mediators and longitudinal outcomes will be discussed.