Part II looks in detail at two different kinds of theories of causality, Bayes-nets theories and various invariance or ‘modularity’ theories. I focus on these principally out of personal history. Both are closely associated with probabilistic theories of causality, which I have worked on in the past. In keeping with the pluralistic stance of this book, I take that it these are not best seen as alternative accounts of one kind of relation – the causal relation – but rather as descriptions of different kinds of systems of different kinds of causal relations requiring different methods to learn about them.

Both kinds of account are closely related to method. Bayes-nets methods provide a way to discover the different causal arrangements consistent with given input information, especially information about conditional independencies and any known causal relations. And at least one version of the invariance accounts – the one I discuss in ch. 10 – can be immediately translated into a method for testing for causality. What is nice in both cases is that it is possible to provide an explicit characterization of the kinds of systems to which the account applies and then to prove that the methods are appropriate to those kinds of systems. For Bayes nets this kind of proof has been there from the start. The account is presented axiomatically, then the methods are derived from it. The proofs in ch. 10 do the same kind of job for certain kinds of invariance.