Mediation analysis has been used by social scientists for the last 50 years to explain intermediate mechanisms between an assumed cause and effect. During these years many advances in statistical mediation analyses were made, including the use of multiple mediators, models for limited dependent variables, latent variable modeling, improved standard errors, and the combination of mediation and moderation analysis. However, only very recently were the causal foundations and underlying assumptions of mediation analysis clarified. These more recent advances used potential outcomes notation and graphical causal models to illuminate the types of causal effects that can be estimated - and more importantly, which assumptions are needed to recover an unbiased causal effect. This course will briefly review the traditional approaches to mediation analysis, then review fundamental topics for causal inference, and then discuss the novel methods that fall under the rubric of “causal mediation analysis.” The causal mediation methods put the assumptions of the analysis front and center, and because the causal assumptions are often untestable, tools like sensitivity analysis become important.
The course is mostly lecture-based, but will also provide numerous opportunities to practice the studied concepts using applied data examples in R.

The proposed target group of this course are researchers in the social sciences at all career stages who want to learn more about causal mediation analysis.

By the end of the course participants will: