Source
Woodward, “Causation and Manipulability” (LIT-713), read in NOTE-585. The book it summarizes, LIT-723, is unread.
What was actually shown
Woodward states and defends the account in the encyclopedia entry (LIT-713).
- Why intervention and not action. A correlation can persist when A is produced by a free act without A causing B: the experimenter’s free act of giving a placebo is correlated with recovery. What is needed is a “surgical” change in A, such that any change in B comes only through A.
- The conditions. Relative to Y, an intervention I on X must (M1) be the only cause of X; (M2) not cause Y by a route avoiding X; (M3) not be caused by anything that affects Y by such a route; and (M4) leave the values of Y’s other causes, off the path from I through X, unchanged.
- The analysis. Whether X causes Y is what Y would do under such an intervention. X is a total cause if some intervention on X changes Y; a direct cause if Y changes when the other variables, including mediators, are held fixed by interventions too.
- The objections. It is not a reduction, since intervention is a causal notion, and it is not viciously circular, since it does not presuppose causal facts about X and Y themselves. It is not anthropocentric: a natural process with the right causal history is an intervention.
The account is philosophical, so nothing in the entry could have come out otherwise experimentally. What makes it more than a definition is that it disagrees with rivals on cases: on omissions and double prevention it gives judgments that causal-process theories reject, which Woodward takes to show it is not vacuous.
What this does not say
- Not that a cause must be manipulable in practice. Interventions need not be feasible, and observational evidence can confirm what an intervention would do.
- Not a reduction of causation. The entry says so.
- Not a theory of levels. This version of the entry does not discuss invariance, proportionality, supervenience or higher-level causation. Its only macro-level remark is that true macro-level interventionist counterfactuals plausibly have more fundamental explanations, without needing non-counterfactual truthmakers.
- Not settled on unmanipulable causes. Whether “being female” or “being a raven” can be causes depends on whether one takes setting or possibility-constrained interventions, and the entry presents both.
Connections
- List & Menzies (THEORY-133). Their difference-making account of higher-level causes applies this kind of counterfactual test across realizations. The interventionist account supplies the notion of a cause; realization-insensitivity is their criterion for when the higher-level property, rather than its realizer, is the cause. No relation is declared: their paper uses Lewis-style semantics, not M1–M4.
- French (THEORY-131). French establishes corporate intentional acts by licensed redescription, not by intervention. This account says what a causal version of his claim would need: an intervention on the corporate variable, such as its policy, that changes the outcome by no route through members’ independent reasons (M2, M4).
- The owner’s essay. For §5.3 it fixes what an agentive explanation of Northstar must show to count as causal. Whether such interventions are ever clean in an organization, where any change of policy is made by members, is the essay’s problem, not one the entry addresses.