Latent Mediators in Causal Mediation Analysis

• Activity Number: 230
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Jeffrey Albert*+ and Cuiyu Geng and Suchitra Nelson
• Companies: Case Western Reserve University and Case Western Reserve University and Case School of Dental Medicine
• Keywords: causal inference ; mediation formula ; measurement error ; potential outcome ; structural equations model

Abstract

In this research, we consider the role of latent mediators in a causally interpreted structural equations model, or more specifically, a causal mediation model. Such a model is of ostensible interest when multiple intermediate variables, possibly measuring a latent mediator, are available. Previous work has shown that the mediation formula may be used to estimate mediation effects for mixed types of outcomes following a maximum likelihood fit of the model. We first explicate the assumptions of the model and resulting interpretation of the latent mediator. A motivating problem is to determine the appropriateness and usefulness of the model for data from a dental caries cohort study. We discuss the possibility and limitations of empirically critiquing the model, and present results of simulation studies to determine the bias of alternative estimators under violations of model assumptions. Finally, we consider implications of the latent mediator model for predicting the effect of future interventions.