Causation as Production and Dependence, or A Model-Invariant Theory of Causation


Many contemporary theories of singular causation (alternatively: token causation, or actual causation) are formulated within the framework of structural equations modelling (or causal modeling). These theories say whether a variable value $C=c$ caused another variable value $E=e$ only given a particular causal model. And the majority of these theories are model-variant in the following sense: they will say that $C=c$ caused $E=e$ in one model; but, when we remove an inessential variable, they change their tune and say that $C=c$ didn't cause $E=e$. In this talk, I develop a theory of causation which is capable of securing the intuitive verdicts in a wide range of cases from the literature and which is model-invariant in the following sense: if the theory says that $C=c$ caused (didn't cause) $E=e$ in a causal model $M$, then it will continue to say that $C=c$ caused (didn't cause) $E=e$ once we've removed an inessential variable from $M$. According to this model-invariant theory, causation is a hybrid of production (understood as the local propagation of deviant, non-inertial variable values) and counterfactual dependence. (Presented at the University of North Carolina, Chapel Hill, February 9th, 2018.)