Two Concepts of Mechanism: Componential Causal System and Abstract Form of Interaction

Abstract

Although there has been much recent discussion on mechanisms in philosophy of science and social theory, no shared understanding of the crucial concept itself has emerged. In this paper, a distinction between two core concepts of mechanism is made on the basis that the concepts correspond to two different research strategies: the concept of mechanism as a componential causal system is associated with the heuristic of functional decomposition and spatial localization and the concept of mechanism as an abstract form of interaction is associated with the strategy of abstraction and simple models. The causal facts assumed and the theoretical consequences entailed by an explanation with a given mechanism differ according to which concept of mechanism is in use. Research strategies associated with mechanism concepts also involve characteristic biases that should be taken into account when using them, especially in new areas of application.

Acknowledgements

An earlier version of this paper was presented at the workshop, ‘Mechanisms in the Sciences: Concepts, Discovery, Explanation’, University of Helsinki, August 2007. I would like to thank all the discussants and the members of the Philosophy of Science Group for their comments. Special acknowledgements go to Peter Hedström, Daniel Steel, and Petri Ylikoski for their detailed and insightful comments. I also owe thanks to the three anonymous referees of this journal for their valuable suggestions and to the Finnish Cultural Foundation for generously supporting this research.

Notes

[1] This is one sense of near decomposability in Simon's classic article. I am not going deeper into exegetics here in trying to question whether there is a singular concept of near decomposability behind the different definitions and uses in the article (such as the near decomposability of a flow matrix or the independence of long‐term aggregate dynamics with respect to short‐term component dynamics).

[2] Craver uses the term ‘levels of mechanism’ to distinguish these from other kinds of ontological levels, such as mereological levels or levels of aggregation. Renate Mayntz uses the more ambiguous term of ontological level in order to draw a contrast to levels of abstraction, which order representations, not things in the world.

[3] With abstraction, I refer to both vertical and horizontal isolation, as discussed by Mäki (1992). Horizontal isolation means completely omitting or excluding some causal factors or features from a model and vertical isolation means simplifying but still retaining some causal feature by stripping away ‘particularities’ (e.g. moving from a parametric to an unspecified functional form). Although Mäki characterizes the latter as vertical, both forms of theoretical isolation remain within the original level of mechanism.

[4] AFI concepts can, and often do, contain references to causal factors in multiple levels of mechanisms. Many mechanism types in the social sciences refer simultaneously to groups, agents, preferences etc. The constellations of causal properties picked out by AFI concepts are thus examples of entities that Wimsatt describes as ‘being in between levels’ (Wimsatt 2007, 217). The use of AFI concepts is also not necessarily linked to methodological individualism: AFI concepts could be applied to the interaction of social units of meso or macro level. However, it should be stressed that these kinds of spatial metaphors can be extremely misleading: mechanisms themselves do not reside in any particular ‘level’ of reality (see also Craver 2007, ch. 5). Mechanisms are constellations of causally interacting objects and are thus always simply located in (the one and only) space–time.

[5] Although many seem to treat it as such, aggregation in itself can hardly be seen as a type of mechanism.

[6] As an anonymous referee correctly pointed out, sometimes heterogeneity can itself be a causally relevant property (selection is a good example). However, in most cases what is relevant is the heterogeneity itself (a relational or system‐level property), not which particular units have which particular properties.