Wednesday 12 February 2014

Johannes Textor - Postdoctoral Fellow (co-funded by DAAD and NWO), Theoretical Biology & Bioinformatics, Utrecht University

Some Words of Warning About Causal DAGs

Graphical causal models, also known as DAGs, are becoming increasingly popular tools for causal inference in the biomedical and social sciences. However, in contrast to the enthusiasm with which the powerful capabilities of these tools have been embraced, there has been relatively little work on the identification of limitations and potential pitfalls. My talk will illustrate two of these limitations. First, it has been recently claimed that DAG methodology allows for a post-hoc causal interpretation of classic structural equation models (SEM). I show an explicit counterexample to this claim. Second, I consider an a simple causal process from biology which, I argue, cannot be reasonably represented as a DAG, and attempting to do so would lead to wrong conclusions. In summary, while DAGs offer an attractive conceptualization of a given set of causal hypotheses and a powerful calculus to derive testable implications from these hypotheses, further work is needed to precisely identify which real-world causal processes can or cannot be represented as DAGs.

The meeting will be held in Room 9.27, Level 9, Worsley Building, University of Leeds. Starting 4pm until 5pm with Teas/Coffees available from 3:30pm.