Naftali Weinberger (University of Pittsburgh)
In my talk, I consider a strategy for causally modeling complex systems by dividing them up into semi-independent 'near-decomposable' subsystems. I use simple examples to show how descriptions of such subsystems in terms of their degree of interaction relate to descriptions invoking the complex temporal relations among the subsystems. The discussion sheds light on the usefulness of econometric time-series methods for causal inference and helps clarify the temporal assumptions that are implicit in causal models.
