![[ Visit Client Website ]](images/banner.gif)
Two posters focus on assumptions of mediation. One assumption is no unmeasured confounders. Random assignment to intervention addresses potential confounding between the IV and mediator, but fails to address potential confounders between the mediator and DV. Confound It: Assessing the Impact of Confounders on Mediation demonstrates sensitivity analysis to assess how confounding affects the mediated effect. Effects of Violating the Homogeneity Assumption on Mediation Inferences evaluates violations of the assumption that relations between mediators and DVs are uniform across treatment conditions.
Not only is it assumed that the relation between mediators and outcomes is the same for all levels of the IV, most analyses do not take into account whether the mediated effect differs for individuals or groups. Exploring a Person-Oriented Mediation Measure describes and tests a person-oriented measure of mediation.
Extending standard methodology to mediation is not always straightforward as evidenced by the research in the next two posters. In a pretest-posttest design, a variety of analysis techniques exist including difference scores, residualized change scores, ANCOVA, and path analysis. There has been little work to formally evaluate these methods in mediation and Estimating the Mediated Effect in Pretest-Posttest Control Group Designs fulfills this gap. A Proposed Model for Mediation Analysis with Structurally Different Raters addresses issues with multimethod measurement in mediation. Although using multiple methods is highly valued in prevention science, a multimethod approach with mediation has limitations. This poster integrates modern statistical mediation methods with modern multitrait-multimethod methods.
Although it may seem counterintuitive, adding a mediator can increase power to detect effects between two variables. The poster, When the Three-Path Mediation Model Has More Power than the Single Mediator Model, compares power of the total effect to both single mediator and sequential mediator models. Results suggest that including theoretically supported mediators increases power, and more mediators can increase power further. This finding adds to the importance of the next poster, Monte Carlo Confidence Intervals for Specific Indirect Effects in Mediation Analysis, which evaluates methods to create confidence intervals for causal chains with more than two mediators.
Finally, there is increased interest in Bayesian methods for modern data analysis. Bayesian Mediation for Prevention Data outlines two Bayesian mediation methods and illustrates with data from a prevention study. SAS syntax is provided.
At the end of the session, the chair and discussant will summarize the posters and moderate a discussion between presenters and the symposium attendees. It is expected that the topics will be of interest to both methodological and applied substantive prevention researchers.