Social Science Statistics Blog

« App Stats: Grubb on "Cellular Service Demand: Biased Beliefs, Learning, and Bill Shock" | Main | Pop music is getting louder, possibly more boring (statistically speaking) »

9 September 2012

App Stats: Robins on "A Simple Unification of the Potential Outcome and Causal Graph Approaches to Causal Inference"

We hope you can join us this Wednesday, September 12, 2012 for the Applied Statistics Workshop. Jamie Robins, Professor of Epidemiology from the Harvard School of Public Health, will give a presentation entitled "A Simple Unification of the Potential Outcome and Causal Graph Approaches to Causal Inference". A light lunch will be served at 12 pm and the talk will begin at 12.15.

"A Simple Unification of the Potential Outcome and Causal Graph Approaches to Causal Inference"
Jamie Robins
Harvard School of Public Health
CGIS K354 (1737 Cambridge St.)
Wednesday, September 12th, 2012 12.00 pm

Abstract:

Potential outcomes are extensively used within statistics, epidemiology, and political science for reasoning about causation. Directed acyclic graphs are another formalism used to represent causal systems. They are extensively used in computer science, bioinformatics, sociology and epidemiology. It is natural to wish to unify them.

We present a simple approach to this unification. The approach is based on the idea of splitting nodes to construct graphs whose nodes are potential outcomes. The resulting graph can be used to read off counterfactual independencies. These independencies are satisfied by all previously proposed graphical and nongraphical causal models. We review many examples to illustrate the power of this approach.

This is joint work with Thomas Richardson at the University of Washington.

Posted by Konstantin Kashin at September 9, 2012 5:53 PM