A0729
Title: Causal mediation analysis for optimizing interventions using factorial designs
Authors: Donna Coffman - University Of South Carolina (United States) [presenting]
Abstract: Optimization trials are used to design and build efficient, immediately scalable, and cost-effective interventions by selecting intervention components (i.e., any part of an intervention that may be separated out for experimental manipulation) for inclusion in an intervention. In most interventions, these components are designed to affect mediators (i.e., third variables) through which the components affect the outcome of interest. There is very little guidance on assessing mediation in the context of optimization trials. Methods are extended based on the potential outcomes framework for causal inference to draw more valid causal inferences about mediation in optimization trials that use a $2^K$ factorial design and effect coding. The focus is on defining, identifying, and interpreting causal mediation effects. For example, using effect coding, a mediated effect of a main effect may be interpreted as the effect of a factor on the outcome through the mediator, averaged over all combinations of levels of the remaining factors. These methods ultimately allow researchers to understand how their interventions work. With this knowledge, they are able to design more efficacious and cost-effective interventions by targeting the most relevant mediator(s). Understanding how interventions are effective is key to designing more powerful, efficient interventions.
