Full Idea
All four probability versions of causation may need causation to be primitive: nomological - to distinguish laws from generalizations; statistical - to decide background; counterfactual - decide background; agent intervention - to understand intervention.
Gist of Idea
All four probability versions of causation may need causation to be primitive
Source
Jonathan Schaffer (The Metaphysics of Causation [2007], 2.1.2)
Book Reference
'Stanford Online Encyclopaedia of Philosophy', ed/tr. Stanford University [plato.stanford.edu], p.22
A Reaction
I don't need much convincing that the probabilistic view is wrong. To just accept causation as primitive seems an awful defeat for philosophy. We should be able to characterise it, even if we cannot know its essence.