Unravelling Causality: A Critique of Lewis's Account of Asymmetry — Epoche B2
Lewis's test admits grounding too easily, and only one clause keeps it out A review with a narrow question David Lewis's 'Counterfactual Dependence and Time's Arrow' (1979 [1] ) is read today alongside a literature that did not exist when it was written. Metaphysicians now distinguish causation from grounding — the relation Jonathan Schaffer's 'On What Grounds What' (2009 [2] ) puts at the centre of metaphysics, on the ground that the discipline's question is what structures what rather than what exists: the rock causes the shattering, whereas Socrates grounds the existence of the set whose only member is Socrates. Both are relations of dependence, and both invite the same test — remove the one and see whether the other survives. So a question arises about Lewis's paper that Lewis was not asking [3] . If counterfactual dependence is what causation consists in [4] , and grounding relations also exhibit counterfactual dependence, what stops Lewis's analysis from counting a case of grounding as a case of causation [5] ? The usual answer is that Lewis's account is too coarse to tell dependence relations apart, and that his events, being spatiotemporal occurrences, blur the difference. This review argues that the usual answer is backwards on both points. Lewis's events are not coarse; the counterfactual test drives them towards fineness. And the fineness makes matters worse, not better, because it lets grounding relations pass the test more cleanly than many uncontroversial causal relations do. What actually keeps grounding out of Lewis's analysis is a single clause that is easy to overlook: the requirement that cause and effect be distinct existences. That clause, and not the event ontology and not the arrow of time, is doing all the work — and it is not obvious that it can bear the load. What the 1979 paper claims Lewis's analysis of causation is stated in 'Causation' (1973). Where c and e are distinct actual events, e depends causally on c if it is true that had c not occurred, e would not have occurred; and c causes e when a chain of such dependences runs from one to the other. The 1979 paper does not restate that analysis so much as defend the counterfactuals it needs. Its problem is that the analysis looks symmetric. If the window had not shattered, would the rock have been thrown? Read one way, no. Read another way, yes. Lewis has to say which reading is the standard one and why. His answer is the asymmetry of overdetermination. A determinant of an event is a set of conditions that, with the laws, suffices for it. Lewis's observation is that events typically have very many determinants afterwards and very few before. The shattering leaves fragments, a noise, a draught and a witness, each of which suffices for the shattering having occurred. Nothing comparable precedes the throw. So a world in which the shattering is absent but everything else is as similar as possible is easy to reach with one small local departure from law, while a world in which the throw is absent but its subsequent traces remain would need very many. Lewis's ranking of worlds therefore makes the backtracking reading non-standard. Direction follows. Two things should be said carefully here, because the point is often reported the wrong way round. It is the earlier event that is overdetermined, by its many later traces, not the later event by earlier conditions. And Lewis holds the whole asymmetry to be contingent: he treats it as a feature of how our world is arranged, not as a truth about causation as such. Lewis's events are fine, not coarse The complaint that Lewis works with a coarse-grained ontology of events misidentifies whose ontology is whose. The coarse-grained view is Donald Davidson's [6] , on which one and the same event can be described as flipping the switch, turning on the light and alerting the prowler — his own example, and his point is that these are one event under three descriptions rather than three events. Jaegwon Kim's view is the fine-grained one [7] , on which an event is the exemplification of a property by an object at a time, so that different properties give different events. Lewis's own view sits nearer Kim's, and it does so under pressure from the counterfactual test. If events were coarse, causal claims would come out false wholesale: had the rock been thrown a moment later, the window would still have shattered, so on a coarse individuation the throw is not necessary for the shattering. Lewis's response is that events have essences of variable strictness, some fragile and some robust, and that the shattering-at-exactly-3.02 is a different event from the shattering. He needs that resource. The finer the events, the more counterfactual dependences hold, and the analysis is designed to trade on dependences holding. Why fineness makes the grounding problem worse Now apply Lewis's test to a paradigm of grounding. Socrates exists; the singleton set containing Socrates exists. Had Socrates not existed, that set would not have existed. The counterfactual is true, and it is not merely true but necessarily true, which is more than any causal counterfactual manages. This is where fineness tells against Lewis. Causal counterfactuals are fragile things. They fail under pre-emption, where a backup cause would have brought the effect about anyway; they fail under symmetric overdetermination, where two sufficient causes operate at once; they need chains of intermediate events to be repaired. Grounding counterfactuals suffer none of this. Nothing pre-empts Socrates in the grounding of his singleton; there is no backup Socrates. So if counterfactual dependence were the mark of causation, grounding would not be a borderline case that the analysis struggles to exclude. It would be the analysis's best-behaved instance. Nor does the arrow rescue the situation in the way the standard story suggests. Lewis's asymmetry of overdetermination is a tie-breaker for choosing between readings of a counterfactual,