An Interventional Approach to Causal Mediation Analysis
Causal Mediation Analysis is concerned with distinguishing different causal pathways that may link a treatment and an outcome.
In the simplest case we may wish to know whether a treatment acts directly on the outcome or via a mediator.
In this talk I will first describe a new conceptual framework that is based on decomposing the treatment into sub-components.
I will then contrast this approach to a non-interventionist approach advocated by Pearl that is based on nested counterfactuals.
The new interventionist approach has several advantages over the counterfactual approach:
- It does not require the existence of well-defined interventions or counterfactuals on mediators;
- Identified effects are, in principle, empirically testable via interventions;
- The new theory preserves the dictum “no causation without manipulation”;
- It replaces the current complex definitions of path specific effects as nested counterfactuals with easily understood definitions in terms of concrete experimental interventions;
- The new approach facilitates communication with subject matter experts.
However, notwithstanding these differences, when both are identified from data, the identifying formulae under an interventionist view and the non-interventionist view are identical, even though the effects are different.
This talk is joint work James M. Robins (Harvard School of Public Health) and Ilya Shpitser (Johns Hopkins University).