Counterfactual and Graphical Frameworks for Causal Modeling

          @misc{ scivideos_PIRSA:24090084,
            doi = {10.48660/24090084},
            url = {https://pirsa.org/24090084},
            author = {Richardson, Thomas},
            keywords = {Quantum Foundations, Quantum Information},
            language = {en},
            title = {Counterfactual and Graphical Frameworks for Causal Modeling},
            publisher = {Perimeter Institute for Theoretical Physics},
            year = {2024},
            month = {sep},
            note = {PIRSA:24090084 see, \url{https://scivideos.org/index.php/pirsa/24090084}}
          }
          

Abstract

In the Statistics literature there are three main frameworks for causal modeling: counterfactuals (aka potential outcomes), non-parametric structural equation models (NPSEMs) and graphs (aka path diagrams or causal Bayes nets). These approaches are similar and, in certain specific respects, equivalent. However, there are important conceptual differences and each formulation has its own strengths and weaknesses. These divergences are of relevance both in theory and when the approaches are applied in practice. This talk will introduce the different frameworks, and describe, through examples, both the commonalities and dissimilarities. In particular, we will see that the “default” assumptions within these frameworks lead to different identification results when quantifying mediation and, more generally, path-specific effects.