Cases of genuine causal preemption may be rare but they are nonetheless interesting. They are rare because it is difficult to arrange the physics of events so that one event (C1) can cause another E, while preempting the efficacy of a third (C2), that would have had the identical effect (E) if C1 had not occurred. But rare as they mat be, genuine cases of causal preemption can be cooked up.1 Such cases are interesting because our intuitions about them are reasonably clear. As such they form a good test for proposed analyses of causation. In particular, they pose a challenge for any account of causation which requires that a cause be necessary for its effects. In this paper I shall argue that one such account, David Lewis' counterfactual conditional approach, fails to meet this challenge.
Lewis' proposed analysis of causation is that one event (C) is cause of another (E) if and only ifE depends counterfactually on C or there is a causal chain from the first to the second; where, a causal chain is a finite sequence of particular events (C,D,E...) such that D depends counterfactually on C, E on D and so on; and, an event E is counterfactually dependent on an event C in world w if some non-C world that is a non-E world is nearer to co than all non-C worlds that are E worlds. ~ Now suppose that C1 causes E and preempts C2 which would have caused E if C1 had not occurred. Then, it seems that on the proposed Lewis analysis, C~ is not the cause of E, since if C~ had not occurred, E would still have occurred. Lewis' defense against this line of attack is to deny the claim that if C1 had E, Q would have occurred but had failed to cause/causedE. His argument runs as follows: 3 Not every causally connected pair of events need exhibit the counterfactual dependence of the effect on its cause since counterfactual dependence need not be transitive. (If E depends counterfactually on D, and D on C, it does not necessarily follow that E depends on C counterfactually.) Assume that E depends counterfactually on D, which is in turn counterfactually dependent on C~, where C~ preempts C2. It might seem that without D, E would still Philosophical Studies 37 (1980) 115-124.