New Developments in Mediation Analysis
Methodology for assessing direct and indirect effects (i.e. mediation) has been used in the biomedical and social sciences for decades. More recently, theory and methods for mediation have been developed from the causal inference literature to extend traditional methods to more complex settings and to clarify the causal assumptions being made. The talk will (i) provide an overview of concepts, assumptions, and methods for causal mediation analysis, (ii) discuss sensitivity analysis methods to assess robustness of effect estimates to assumptions made, and (iii) present new results and methods on a 4-way decomposition that decomposes a total effect, in the presence of a mediator with which the exposure may interact, into four components: that due to just mediation, that due to just interaction, that due to both mediation and interaction, and that due to neither. The methodology will be illustrated by scientific and policy relevant examples from medicine, psychology, and genetics.
Further book-length discussion of this material can be found in: VanderWeele TJ. (2015). Explanation in Causal Inference: Methods for Mediation and Interaction. Oxford University Press. http://www.amazon.com/Explanation-Causal-Inference-Mediation-Interaction/dp/0199325871/ref=sr_1_1?ie=UTF8&qid=1423279902&sr=8-1&keywords=tyler+vanderweele