Investigating causal relations often leads to questions regarding the processes and mechanisms underlying a specific effect. Is an effect mediated by one or more other variables? In practice, this question is frequently assessed by analyzing changes in regression coefficients after adding the putative mediators to the model. The modern literature on causal inference demonstrates, however, that this approach yields valid conclusions regarding mediation only under specific substantive assumptions that are rarely made explicit in applied research.

This course uses graphical causal models and counterfactual definitions of direct and indirect effects to make transparent the conditions under which mediation analysis yields valid conclusions. In addition to classic approaches to mediation, the course also introduces modern regression-based methods of causal mediation analysis as well as formal sensitivity analysis. The course concludes with a discussion of further topics in causal inference and mediation analysis (e.g., instrumental variables, multiple mediators, alternative estimation approaches).

Bremer Häuser, Viertel, Bremen, Germany