Hunting Causes and Using Them: Is There No Bridge from Here to There?

Abstract

Causation is in trouble—at least as it is pictured in current theories in philosophy and in economics as well, where causation is also once again in fashion. In both disciplines the accounts of causality on offer are either modelled too closely on one or another favoured method for hunting causes or on assumptions about the uses to which causal knowledge can be put—generally for predicting the results of our efforts to change the world. The first kind of account supplies no reason to think that causal knowledge, as it is pictured, is of any use; the second supplies no reason to think our best methods will be reliable for establishing causal knowledge. So, if these accounts are all there is to be had, how do we get from method to use? Of what use is knowledge of causal laws that we work so hard to obtain?

Acknowledgements

This paper was presented at the Society for Philosophy of Science in Practice at the University of Twente in the summer of 2007. It develops ideas presented in a more roundabout way in Cartwright (2006) and in Cartwright (2007). Support for the research and completion of the paper was provided by the AHRC project Contingency and Dissent in Science, by the University of California at San Diego Senate, by the University of California Humanities Research Institute, by the Institute for Advanced Study at Durham University, and by the Spencer Foundation; we are grateful to all for their help.

Notes

Among philosophers, we discuss Woodward rather than other accounts (e.g. Menzies and Price 1993, Gasking 1955, or Von Wright 1971) because his two separate conditions directly illustrate the distinction between characterizations of causality rooted in hunting methodologies and those that work to guarantee use.

Cf. Cartwright's early paper, Cartwright 1979 (also published in Cartwright 1983) or the spate of work at about the same time on causal decision theory (for instance in Harper, Stalnaker, and Pearce 1981).

The range of permitted variation of variable must be specified as well. Mention of the relativization to these ranges is suppressed as well.

Cartwright takes these axioms to be fairly innocuous and to be true of causal laws even if her singular-causings account of causal laws is mistaken. Cartwright notes that there has been some objection that the axioms are not so innocuous because a transitivity axiom is included. But the transitivity axiom assumes only that if x appears as a cause of y in a linear deterministic causal system, we still have a causal law for y if we substitute for x the right-hand side of any causal law that has x as effect. This is necessary unless we are willing to assume that causation in nature is not continuous in time, so that there is a notion of direct causal law (the ‘last’ law in operation before the effect is produced) that is not representation relative, and that it is this notion of direct causal law that we are trying to characterize.

It also supposes a relatively clear separation between the structure (what Cartwright calls the ‘nomological machine’) that gives rise to a set of causal laws and the laws themselves. Cartwright has defended this division in Nature's Capacities and Their Measurement (1989) and in The Dappled World (1999). Cartwright does not think it is universally applicable. But she does think it can be made practically everywhere where there are causal laws at work that can be represented in the usual triangular array of equations, whether those equations represent deterministic or probabilistic causality and whether they are linear or not.

The formulation given here still isn't quite right, because K must not hold fixed any causal intermediaries by which C causes E on a given occasion. Cartwright's own best attempt relies on reference to singular causings even in the formulation of the probabilistic theory. See Cartwright 1989, §§2 and 3, and Cartwright 2007 for a fuller discussion.

As usual the population relativity is buried in the probability measure. The laws hold for any population for which the assumed probability measure holds.

Efstathiou's account of ‘founded’ concepts could help understand the problem that arises here. Efstathiou proposes that there is a process of conceptual modification through which some common concept becomes ‘founded’ in a scientific context. Founding modifies common ideas to become relevant and precise enough to be usable in already available theoretical frames (Efstathiou 2009). Concepts of ‘causation’ may come to differ across hunting and using shores and differ from the more general notions we are interested in in philosophy because causal notions are founded in distinct—hunting and using—scientific contexts.

Note as in Section 2.2 that we can drop the relativization to the arrangement of confounders by relativizing instead to a population. Clearly C will cause some Es in a population if it is guaranteed to cause some Es in some subpopulation of that population (i.e. a subpopulation that is homogeneous with respect to confounding factors).

Except for chancy capacities, which produce their contributions only spasmodically.

The description given here of Hoover's account is not one he is happy with. Cartwright claims it is what his definitions say and takes the kind of causal relation described by the definitions as a very important one different from more ‘mechanical’ kinds of causal relations. He maintains that he intends his account to cover the more conventional notion of ‘mechanical causation’ and that various caveats he offers allow his definitions to do so. For further discussion, see Cartwright 2007, ch. 14.

Hoover causation thus is closely associated with the kind of ‘implementation neutral’ counterfactual that Hausman proposes for investigating casual claims, but with the range of implementations restricted to implementations we are able to bring about. See Cartwright 2007, ch. 16.

Or better, with note 11 in mind, ‘Hoover as described here’.

This is not exactly how Woodward defines modularity but it is how he uses the notion sometimes and especially to do just the job discussed here. See Woodward's definition of modularity in Woodward 2003, 329.

Actually, he gives the same reason—causes must be usable to manipulate their effects—for both level invariance and for modularity. We cite it only for modularity because level invariance does not provide manipulability unless modularity is added and Cartwright at any rate has an alternative defence of level invariance.

Also, in line with the probabilistic theory of causality there is in general the assumption that causes and effects are probabilistically dependent.

A good many capacities are derivative however. These too will depend on the underlying structure, or ‘nomological machine’, that gives rise to them.

Indeed, in most cases it is just because we know both a rule of composition and the contribution of a full set of causes towards the effect that we can make sense of the idea of a contribution from any one of them.

Additional information

Notes on contributors

Nancy Cartwright

Nancy Cartwright is at the Centre for the Philosophy of Natural and Social Science, London School of Economics, and the Department of Philosophy, University of California, San Diego.

Sophia Efstathiou

Sophia Efstathiou is at Philosophy, University of Southampton