Department of Statistics and Department of Economics, University of Washington
Potential outcomes (aka counterfactuals) are extensively used within Statistics, Political Science, Economics, and Epidemiology for reasoning about causation. Directed acyclic graphs (DAGs) are another formalism used to represent causal systems also extensively used in Computer Science, Bioinformatics, Sociology and Epidemiology. Given the utility of both approaches -- as demonstrated by many applications -- it is natural to to wish to unify them.
I will present a simple approach to this synthesis based on an intuitive graphical transformation: by 'splitting' treatment nodes in a causal DAG over the actual variables, we form a new graph, the Single-World Intervention Graph (SWIG). The SWIG encodes the counterfactual independences associated with a specific hypothetical intervention on the set of treatment variables.
I will contrast this new approach to previous attempts at unification such as the Non-Parametric Structural Equation Models with independent errors, proposed by Pearl (2000, 2009).
(Joint work with James Robins, Harvard School of Public Health.)