Some Applications of Causal Mediation Analysis and Discrepancies with Traditional Mediation Analysis

• Activity Number: 24 - Recent Advances in Causal Analyses That Tell the Story of Complex Mediation Mechanisms
• Type: Topic-Contributed
• Date/Time: Sunday, August 8, 2021 : 1:30 PM to 3:20 PM
• Sponsor: Biometrics Section
• Author(s): David Peter MacKinnon* and Matthew Valente and Diana Alvarez Bartolo
• Companies: Arizona State University and Florida International University and Arizona State University
• Keywords: Causal Methods; Third-variable; Mediation; Counterfactual; Cognitive Psychology

Abstract

We describe results from applying traditional and counterfactual-based causal mediation methods to data from eight cognitive psychology studies investigating the same mediation hypothesis. The primary mediation hypothesis was that a randomized manipulation increases the use of mental imagery of words, which results in recall of more words. Another mediation hypothesis was that manipulation induced repetition of the word would not increase word recall. We estimated single and two-mediator models. Single mediator methods included regression-based, natural effects, Ratio of Mediator Probability Weighting, and Monte Carlo methods and all methods generally estimate the same mediated effect. For the two-mediator model, methods focus on the estimation of different mediated effects. Two mediator methods included parametric g-computation, natural effect models, interventional analogues, and Monte Carlo methods. Estimates were generally consistent across traditional and counterfactual methods, with most discrepancies between methods occurring in studies with smaller sample sizes.