ABSTRACT
The statistical study of causal relationships and causal inference has taken off over the last few decades and the literature on this topic is now vast. This chapter explores how the framework of the Chain Event Graph (CEG) currently adds to the armoury of this methodological toolkit. It focuses on the problem of defining a sound causal algebra that can be used on the class of CEGs. The chapter suggests that the CEG can directly through its graph embody collections of conditional independence relationships. It provides the analogous conditional independences within a Bayesian network (BN)—or equivalently the implicit factorisations of a joint density—that were used by Pearl and others to formulate their causal algebras. These formally described what might happen if the modelled domain were to be. The chapter shows that the do-operation in BNs is only a special case of such a CEG manipulation.
