Decision-Theoretic Foundations for Causal Reasoning

We present a definition of cause and effect in terms of decision-theoretic primitives and thereby provide a principled foundation for causal reasoning. Our definition departs from the traditional view of causation in that causal assertions may vary with the set of decisions available. We argue that this approach provides added clarity to the notion of acyclic graphs. We describe a special class of influence diagrams, those in canonical form, and show its relationship to Pearl’s representation of cause and effect. Finally, we show how canonical form facilitates counterfactual reasoning.