Mediation Analysis Based on Propensity Scores

• Activity Number: 670
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Yeying Zhu*+ and Debashis Ghosh and Donna Coffman
• Companies: Penn State University and Penn State University and Penn State University
• Keywords: causal inference ; generalized boosting ; mediation analysis ; propensity scores ; sequential ignorability

Abstract

In clinical trials, many researchers are interested in examining whether a randomized treatment or intervention may affect the outcome through an intermediate factor. Traditional mediation analysis (Baron and Kenny, 1986) applies structure equation modeling (SEM) and decomposes the intent-to-treat (ITT) effect into direct and indirect effects. More recent approaches interpret the mediation effects as controlled effects (Robins and Greenland, 1992), natural effects (Pearl, 2001), and principal stratification effects (Frangakis and Rubin, 2002), and so forth. In practice, there often exist confounders, pre-treatment covariates that jointly influence the mediator and the outcome. We propose propensity-score-based methods to reduce the dimensionality of confounders under the sequential ignorability assumption. We assess the proposed methods through simulation studies. Furthermore, we show that combining machine learning algorithms (such as a generalized boosting method) and logistic regression to estimate propensity scores can be more accurate and flexible in mediation analysis, compared to logistic regression only.