ABSTRACT
This chapter focuses on identifiability of path-specific effects—that is, whether or not a certain set of causal assumptions suffices to identify such components from observed data. It provides a more comprehensive overview of the literature on causal mediation analysis. The chapter reviews definitions of path-specific effects, in particular of natural direct and indirect effects, which are the standard targets of inference in causal mediation analysis. In many applications across a wide range of scientific disciplines, scholars aim to understand the mechanisms behind established cause-effect relationships, as witnessed by the widespread usage of mediation analyses. Mediation analyses based on natural or path-specific effects are thus bound to rely on assumptions that cannot be empirically verified or guaranteed by any study design. The chapter shows that the distinct nature of nested counterfactuals calls for a type of assumption that cannot be verified empirically but that is naturally encoded in so-called non-parametric structural equation models.
