Contribution
It replaced the agency theories’ appeal to free human action with a causally defined notion of intervention. That answers the anthropocentrism objection and turns the circularity objection into a non-vicious dependence on other causal facts.
Key insight
What makes a correlation causal is how it behaves when X is changed from outside in a way that touches Y only through X. A free act does not guarantee that (the placebo case); an intervention, by definition, does.
Assumptions
- Causal relata are variables, whose values the same system can take.
- An intervention is defined relative to a target variable Y (Woodward and Hitchcock), not, as in Pearl, by leaving the X→Y mechanism intact.
- Causal effect is relative to a background context.
Key results
- Intervention, M1–M4 (§5). (M1) I is the only cause of X, setting it entirely. (M2) I does not directly cause Y by a route that avoids X. (M3) I is not caused by any cause that affects Y by a route that avoids X. (M4) I leaves unchanged the values of Y’s causes other than those on the path from I through X to Y.
- Causal concepts (§7). X is a total cause of Y if some interventions on X change Y. X is a direct cause of Y if Y changes under an intervention on X while all other variables, including those on routes from X to Y, are held fixed by interventions.
- Non-vicious circularity (§6). The account says which other causal facts are needed, not the fact being characterized; “we do sometimes learn about causal relationships by performing experiments”. It is not trivial, because it disagrees with causal-process theories about omission and double prevention.
- Lewis explained (§9). Lewis’s similarity ordering works because it roughly picks out relationships “stable under interventions”.
- Setting versus possibility-constrained interventions (§§5, 10, 12). Woodward leans to the setting version: what matters is whether well-grounded theory says what would happen, not whether the intervention is physically possible. He states both views on unmanipulable causes such as race and gender and does not settle it.
- Reply to operationalism (§13). The account does not require that interventions can be carried out; observational data can confirm claims about what interventions would do. Woodward concedes it is “mono-criterial”.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Free action does not suffice to identify a cause | strong | placebo counterexample to Menzies & Price |
| C2 | X causes Y iff some intervention on X (M1–M4) would change Y | moderate | definitional proposal, defended against objections |
| C3 | The account is non-reductive but not viciously circular | moderate | argument, §6 |
| C4 | Interventions need not involve human agents | strong | follows from the causal definition; natural experiments |
| C5 | Whenever a macro-level interventionist counterfactual is true there is some more fundamental explanation of why | weak | stated as a plausible expectation, not argued (§13) |
Concepts
- intervention — an exogenous causal process on X meeting M1–M4 with respect to Y; “it has no essential connection with human action” (preamble).
- interventionist counterfactual — what Y would do if an intervention set X to a value.
- setting intervention — X set to a value by arrow-breaking, with no further restriction on possibility.
- soft intervention — an exogenous source of variation in X, uncorrelated with X’s other causes, that leaves those causes acting (Eberhardt & Scheines, §14).
Connections
It builds on Pearl and on Woodward & Hitchcock, and criticizes Menzies & Price’s agency theory. List and Menzies (LIT-722) build on Woodward’s interventionism, as the record’s reading of them notes. The book (LIT-723), unread, is where invariance is developed.
Bearing on the record
Source of THEORY-138. It is the general account behind the realization-insensitivity test of THEORY-133.
For the owner’s essay (§5.3), it says what an agentive explanation of Northstar would have to deliver to count as causal: an organizational variable that the same organization could take in different values, and an intervention on it that changes the outcome by no other route. The “Role” supplement’s example comes closest to institutions: on “being female causes one to be discriminated against”, intervening on the legal and cultural framework is “a quite different experiment” from intervening on gender. That is a model for separating an organization’s regime from its members’ attributes. The entry does not discuss organizations, collective agents or levels, so anything further is the record’s application.
Limitations
It is a survey. It does not adjudicate between rival characterizations of intervention, only sketches token causation, leaves open whether fundamental physics is causal, and admits that the notion of a possible intervention is unclear. Invariance, proportionality and higher-level causation are not treated in this version.
Open questions
What, for an organization, counts as an intervention on an organizational variable that leaves its members’ other causal influence on the outcome unchanged (M4); and what chapter 6 of the book adds on invariance.