Скачать 126.01 Kb.
|cause for a less casual notion of causality. (Hedström 2005:23)|
Like Goldthorpe, Hedström also lays out a script for the construction of mechanisms, which is similar in its appeal to the principles of methodological individualism (although he steps back from Goldthorpe's strict reliance on subjective expected utility forms of rational choice theory, noting the promise of agent-based models, and similar approaches). Most importantly, Hedström does not advocate the indirect testing of mechanisms in the way that Goldthorpe does. Instead, he defends the explanatory utility of mechanisms in a much more general way, writing:
Mechanisms should be seen as theoretical propositions about causal tendencies, not as statements about actualities. An explanation may be perfectly correct if understood as a proposition about a causal tendency, and yet it may be inadequate for predicting actual outcomes if other processes are also at work ... Since it is the rule rather than the exception that concretely observed phenomena are influenced by several different processes, testing a theory by examining the accuracy of its predictions is likely to conflate the truth or relevance of the postulated mechanism with the importance of other processes, and this may lead us to mistakenly reject perfectly appropriate causal accounts. (Hedström 2005:108)
What, then, is the explanatory status of mechanisms that are theoretical propositions about causal tendencies? Here, Hedström seems to argue three basic points: (1) An explanation must be a causal account; (2) causal accounts must include mechanisms; (3) mechanisms may or may not explain actualities. The direct implication of this position is that a valid explanation does not necessarily have to explain actualities, only causal tendencies.9
In our view, Goldthorpe and Hedström have developed a wonderful appeal for the need to construct sufficiently deep theories of social processes. Their work, as well as the work of their compatriots, is filled with much insight on a variety of promising ways to formulate plausible theoretical mechanisms. And, they have helped to further convince many social scientists that specifying the social processes that account for how causes bring about their effects is a central goal of analysis. It is therefore unsurprising that this new wave of scholarship has inspired a movement of sorts in sociology and political science, which we and many others regard as both promising and very healthy for theory construction in the social sciences.10
But, to be seen as blueprints for explanatory causal analysis, both of these particular perspectives need to be augmented. And, we would argue that they can be usefully extended by embracing the counterfactual model of causality more directly. We explain this position in the remainder of this section, first by offering a slightly different reading of recent developments in the philosophy of science and then by raising the issue of how social scientists can retain the capacity to adjudicate between competing mechanistic accounts of causal effects.
Philosophy of Science and Explanation via Mechanisms
As we noted at the beginning of this chapter, the appeal to mechanisms as elaborations of causal claims has been entrenched in the sociological literature for many decades, and we suspect that this emphasis is true of all of the social sciences.11 Where perhaps the focus on generative mechanisms is new is in the proposed elevation of their status to the object of primary investigation. This elevation is not unrelated to shifting currents in the philosophy of science.
Some proponents of a mechanism-based social science have drawn inspiration from the demise of covering law models of explanation in the philosophy of science (e.g., Gorski 2004, Hedström 2005). The covering law model, as most famously explicated by Carl Hempel, maintains that all valid explanations can be formulated as logico-deductive entailment from invariant general laws.12 Since 1970 at least, the model has received near-continuous challenges to its basic premises (see Godfrey-Smith 2003, Salmon 1989, and Woodward 2003). The presupposition that general and exceptionless laws exist and that can be used to warrant all valid causal claims cannot be sustained in most scientific disciplines, and certainly not in the social sciences.
In response, a variety of alternative models of scientific explanation have arisen, with realist models attracting the largest number of adherents (see Psillos 1999). In general, realist models grant ontological status to unobserved quantities (sometimes conferring provisional truth status upon them). Most realist models reject causal nihilism and affirm that valid scientific explanation must necessarily be grounded in causal accounts of some form. Especially when invoked for the social sciences, realist models admit the possibility of inherent heterogeneity of causal relationships -- in time, space, and within populations. But, the depth of the appeal to unobservables varies across types of realist models, as does the level of provisional truth status conferred upon unobservables.
What we wish to emphasize in this section is how uneasily the focus on generative mechanisms sits within this new terrain. As we have just noted, the appeal for mechanistic explanation in the social sciences is attractive because it is claimed that it does not rest on the existence of general laws. But, some philosophers who endorse the mechanisms position disagree with this claim, arguing that laws are still essential, perhaps even constituting the defining characteristic of what a mechanism is. For example, having written on the role of mechanisms in scientific explanation for several decades, Mario Bunge (2004:207) concludes in a recent article: No law, no possible mechanism; and no mechanism, no explanation.13
Hedström (2005) considers this complication carefully, and he emphasizes that his mechanism-based approach to explanation is less reliant on invariant laws than the covering law models that he rejects (see pages 30-1). This position is certainly correct, but one must still consider how to grapple with what seem to be three fairly common declarative statements on the relationships between laws and mechanisms: (1) invariant covering laws do not exist, (2) mechanisms depend to some degree on the existence of laws that are weaker than general, invariant laws, (3) mechanisms are nested within each other. Accepting all three statements seems to delimit laws to the smallest possible configuration within each nested mechanism, such that the laws of a mechanism are dispensed with each time a new black-box is opened.
If a mechanism is designed to explain how an actuality comes about, then all seems to be fine with this perspective.14 But, if one instead maintains that mechanisms are the key to the explanation of causal tendencies only -- such that the validity of a mechanism cannot be undermined by its inability to explain anything in particular -- then this line of thought leads all too easily to the critical realist perspective on mechanisms. We suspect that most empirical social scientists would find critical realism uninspiring. Critical realism's pioneer, Roy Bhaskar, writes with regard to mechanisms:
The world consists of mechanisms not events. Such mechanisms combine to generate the flux of phenomena that constitute the actual states and happenings of the world. They may be said to be real, though it is rarely that they are actually manifest and rarer still that they are empirically identified by men. They are the intransitive objects of scientific theory. They are quite independent of men -- as thinkers, causal agents and perceivers. They are not unknowable though knowledge of them depends upon a rare blending of intellectual, practico-technical and perceptual skills. They are not artificial constructs. But neither are they Platonic forms. For they can become manifest to men in experience. Thus we are not imprisoned in caves, either of our own or of nature's making. We are not doomed to ignorance. But neither are we spontaneously free. This is the arduous task of science: the production of the knowledge of those enduring and continually active mechanisms of nature that produce the phenomena of the world. (Roy Bhaskar 1998:34-5)
If the social sciences sign on to the idea that mechanisms are general and transcendentally valid explanations that may not explain any particularities, we will be led inevitably to a fundamental premise of critical realism: the mechanisms that constitute causal explanations are irreducible to each other, even if they are nested in each other. This position is summarized by Andrew Collier (2005:335) as: Critical realism defends the idea that reality is many-layered, and each level has its own kind of laws, irreducible to those of any other layer.
For the social sciences, one might argue that such an irreducibility presumption could be unifying in offering protection against incursions from biology and physics. But, if accepted, it would undermine the work of social scientists who have at least some interest in developing causal claims that unite levels of analysis. If irreducibility were accepted, how then could methodological individualists such as Goldthorpe criticize those who seek to develop macro-level causal claims with only minimally sufficient reliance on proximate actors and institutions (e.g., Alexander 2003).15 Gorski (2004), for example, lays out a constructive realist model of explanation, built up from the causal process perspective of Wesley Salmon, that is completely at odds with the perspective of Goldthorpe (2000).16 But, Goldthorpe (2000) questions the explanatory utility of all second-hand historical analysis, in essence rejecting the capacity of historical analysis, as practiced in sociology, to sustain causal claims of any form. If Goldthorpe were to sign on to irreducibility, which we doubt he would, he could not thereby criticize macro-social claims of causal relationships.
Given that these extreme positions on mechanisms in the philosophy of science are likely to be unhelpful to practicing social scientists, and given that we suspect Goldthorpe, Hedström, and others in the generative mechanisms movement would agree, which type of philosophy of science gives the appropriate backing for causal analysis? If anything, it is the philosophical writing on the counterfactual model that provides a solid and pragmatic foundation, as best represented by Woodward (2003). The key to understanding why this is the case is to consider alternative ways to adjudicate between the rival mechanisms proposed by alternative investigators, which we turn to next.
Adjudication Between Rival Mechanisms
Imagine that social scientist A and social scientist B have proposed distinct mechanisms for how brings about . How do they determine whose mechanism is supported? According to Goldthorpe (2000, 2001), each scholar is expected to derive entailed hypotheses and test them with data. One might hope that scholars A and B will be able to agree on a clear critical test that could tip the scales in favor of one mechanism or the other, but the mechanisms of the two scholars may be so different that no such critical test can be derived and agreed upon (as would often be the case if scholar A is a sociologist and scholar B is an economist, for example). If no such agreement can be found, the two scholars may end up expending effort seeking to affirm their own entailed indirect hypotheses.
Consider the reaction that Goldthorpe's proposal elicited from the statisticians David Cox and Nanny Wermuth, whose prior work Goldthorpe had used to develop his proposal. They write:
Goldthorpe (2001) has argued for this .. view of causality as the appropriate one for sociology with explanation via rational choice theory as an important route for interpretation. To be satisfactory there needs to be evidence, typically arising from studies of different kinds, that such generating processes are not merely hypothesized. Causality is not be established by merely calling a statistical model causal. (Cox and Wermuth 2001:69)
Cox and Wermuth take the position that generative mechanisms must be directly evaluated, not evaluated only via indirect entailed hypotheses. Anything short of this analysis strategy could result in a flourishing of mechanisms in the social sciences, without an attendant sense of which ones are valid or not. The alternative to mechanism-anarchy could be even worse: mechanism warlordism. The mechanisms of the most industrious scholars -- those who can dream up the most hypotheses to affirm, who can recruit the largest number of students to do the same, and who can attract the funds to collect the data -- could receive the most affirmation. The only defense for out-of-favor mechanisms might then be to appeal to the hidden structures of alternative mechanisms, which one would be tempted to claim cannot be evaluated because of a lack of data.
In sum, the generative mechanisms movement in the social sciences is an admirable call for theory construction.17 When it does not verge into critical-realist transcendentalism, it is also a very useful call for the pursuit of sufficiently deep causal accounts. But, such depth, we would argue, is best secured when it is verified in empirical analysis grounded on the counterfactual model. In the next section, we work our way back to Pearl's front-door criteria, using the language of mechanism sketches and mechanism schemas.
The Pursuit of Explanation via Mechanisms that Bottom Out
Amidst the resurgence of writing on mechanisms, we find one statement more helpful than many others, the 2000 article Thinking About Mechanisms written by Machamer, Darden, and Craver and published in Philosophy of Science. In their article, Machamer, Darden, and Craver develop two particularly helpful lines of thought: (1) the distinctions between a fully articulated mechanism, a mechanism sketch, and a mechanism schema; (2) the process of bottoming out in mechanistic model building. To develop these concepts, Machamer, Darden, and Craver (2000:12) first note that: In a complete description of [a] mechanism, there are no gaps that leave specific steps unintelligible; the process as a whole is rendered intelligible in terms of entities and activities that are acceptable to a field at a time. But, they then explain that explanatory inquiry using mechanisms is not an all-or-nothing affair, in which every step is always specified. Variability in the representation of mechanisms is possible because:
Mechanisms occur in nested hierarchies ... The levels in these hierarchies should be thought of as part-whole hierarchies with the additional restriction that lower level entities, properties, and activities are components in mechanisms that produce higher level phenomena ... (Machamer, Darden, and Craver 2000:13)
In spite of such nesting, there is a natural bottoming out of mechanism-based explanations:
Nested hierarchical descriptions of mechanisms typically bottom out in lowest level mechanisms. These are the components that are accepted as relatively fundamental or taken to be unproblematic for the purposes of a given scientist, research group, or field. Bottoming out is relative: Different types of entities and activities are where a given field stops when constructing mechanisms. The explanation comes to an end, and description of lower-level mechanisms would be irrelevant to their interests. (Machamer, Darden, and Craver 2000:13)
Then, by thinking through the complexity of the nesting of mechanisms, and how scholars represent mechanisms to each other, they develop two related concepts. A