Jim Woodward has extended a lot of his philosophical effort towards the task of understanding causation in the sciences (Making Things Happen: A Theory of Causal Explanation). Woodward is a primary exponent of the “manipulationist” theory of causation. He brings a counterfactual orientation to the problem of defining causal relations. If we assert that X caused Y, there is an implication that, if X had not occurred, Y would not have occurred. This implication isn’t universally valid, since some events or outcomes are causally overdetermined. (Both X and X’ may be a sufficient cause for Y — in which case removing X still allows for Y through the X’ pathway.) Notwithstanding this problem, the counterfactual nature of causal assertions is widely recognized. And this implies the association between causation and intervention or manipulation: if X causes Y, then we should be able to influence the occurrence of Y by manipulating X. This fact in turn underlies the logic of experimental design.
Woodward’s treatment of causation deserves fuller treatment than I’ll give it here. In this post I will focus on his application of these ideas to the notion of a causal mechanism. He lays this treatment out in a short but influential article, “What is a Mechanism? A Counterfactual Account” (link).
Here is the core idea. He focuses on the Machamer-Darden-Craver (MDC) definition of a causal mechanism (link):
Mechanisms are entities and activities organized such that they are productive of regular changes from start or set-up to finish or termination conditions. (3)
Woodward’s contribution is to give greater clarity to the idea of regularity or law by adding the idea of a relationship that is “invariant under intervention”. This idea models the notion of experimental testing of a causal hypothesis. We are interested in “X causes Y”. We look for interventions that change the state of Y. If we find that the only interventions that change Y, do so through their ability to change X, then the X-Y relation is said to be invariant under intervention, and X is said to cause Y. Here is how he expresses the idea in the article:
I understand this in terms of the notion of invariance under interventions. Suppose that X and Y are variables that can take at least two values. The notion of an intervention attempts to capture, in non-anthropomorphic language that makes no reference to notions like human agency, the conditions that would need to be met in an ideal experimental manipulation of X performed for the purpose of determining whether X causes Y. The intuitive idea is that an intervention on X with respect to Y is a change in the value of X that changes Y, if at all, only via a route that goes through X and not in some other way. This requires, among other things, that the intervention not be correlated with other causes of Y except for those causes of Y (if any) that are causally between X and Y and that the intervention not affect Y independently of X. Thus if A is a common cause of B and S as in the example above, manipulating B by manipulating A will not count as an intervention on B with respect to S since in this case the manipulation affects S via a route (the route that connects A to S ) that does not go through B. (369-70)
Here is how he applies this idea to causal mechanisms. A mechanism consists of separate components that have intervention-invariant relations to separate sets of outcomes. These components are modular: they exercise their influence independently. And, like keys on an accordion, they can be separately activated with discrete results.
So far I have been arguing that components of mechanisms should behave in accord with regularities that are invariant under interventions and support counterfactuals about what would happen in hypothetical experiments. (374)
Here is the proposal all of this leads up to:
(MECH) a necessary condition for a representation to be an acceptable model of a mechanism is that the representation (i) describe an organized or structured set of parts or components, where (ii) the behavior of each component is described by a generalization that is invariant under interventions, and where (iii) the generalizations governing each component are also independently changeable, and where (iv) the representation allows us to see how, in virtue of (i), (ii) and (iii), the overall output of the mechanism will vary under manipulation of the input to each compo- nent and changes in the components themselves. (375)
Woodward illustrates his theory of mechanisms with simple physical and biological examples. How does this theory work when we consider social mechanisms?
What seems most evident is that social mechanisms are not commonly as complex as Woodward’s examples would suggest. The sorts of mechanisms that crop up in sociology seem largely to be “simple” mechanisms: they don’t consist of multiple independent components leading to an outcome.
Here is the way that McAdam, Tarrow and Tilly (MTT) characterize mechanisms and processes in Dynamics of Contention:
- Mechanisms are a delimited class of events that alter relations among specified sets of elements in identical or closely similar ways over a variety of situations.
- Processes are regular sequences of such mechanisms that produce similar (generally more complex and contingent) transformations of those elements. (24)
These definitions imply that processes are compound, whereas typical mechanisms are simple.
Here are examples that MTT offer of mechanisms:
- resource depletion or enhancement affects people’s capacity to engage in contentious politics (25)
- commitment is a widely recurrent individual mechanism in which persons who individually would prefer not to take the risks of collective action find themselves unable to withdraw without hurting others whose solidarity they value (26)
- Brokerage … as the linking of two or more previously unconnected social sites by a unit that mediates their relations with one another and/or with yet other sites (26)
- Identity shift … alteration during contentious claim making of public answers to the question: “Who are you?” (27)
In each case we seem to have a simple relationship between one social or environmental fact and a typical outcome — not a complex concatenation of “cogs and wheels” of social interaction.
So when we consider typical examples of social mechanisms — free-riding (Olson), escalation (McAdam-Tarrow-Tilly), identity competition (Horowitz) — we commonly find that they are all basically one-step mechanisms. So the assumption that a mechanism consists of modular components doesn’t fit the social sciences well. There are complex social processes, to be sure, but it seems best to understand these as concatenations of distinct mechanisms rather than as a single complex mechanism. (Why? Because they are all too often unrepeatable.)
This doesn’t mean that we can’t understand social mechanisms along the lines Woodward suggests, if we are content to acknowledge that it is hard to find complex social mechanisms. But in order for even this to be the case, we would have to confirm that these simple mechanisms produce intervention-invariant regularities.
This requirement runs up against a different problem, however. The regularities that correspond to typical social mechanisms are soft regularities, not hard-and-fast laws. Social causation is probabilistic, not deterministic. The regularities corresponding to social causes derive from features of human agency and behavior, and they are deeply exception-laden. The free-rider mechanism tends to give rise to under-investment in the public good — except when people self-organize, semi-coercive organizations appear, or altruistic religious attitudes take hold. Social mechanisms are productive, in the sense that they “bring about” the associated outcomes. But they are not invariant across all or most cases.
This implies that there are no intervention-invariant relations to be had in the social world. And therefore we need some other analytical foundation if we are to persist in thinking there are social causal mechanisms.
Woodward addresses something very much like this possibility in conjunction with psychological mechanisms. And he draws a prescriptive conclusion: if the “mechanisms” cited in psychology do not have these characteristics of modularity and invariance, then they aren’t really mechanisms:
the standard boxological diagrams allegedly describing the operation of psychological mechanisms drawn by psychologists are rarely accompanied by convincing evidence that the parts corresponding to the boxes satisfy the modularity condition described above. If the argument of this paper is correct, this is a reason for being skeptical that these diagrams describe genuine mechanisms. (377)
It seems likely enough that he would reach a similar conclusion about the kinds of mechanisms offered by MTT.
(Here is an excellent review by Michael Strevens of Woodward, Making Things Happen.)