Interventionism and Higher-level Causation

Abstract

Several authors have recently claimed that the notorious causal exclusion problem, according to which higher-level causes are threatened with causal pre-emption by lower-level causes, can be avoided if causal relevance is understood in terms of Woodward's interventionist account of causation. They argue that if causal relevance is defined in interventionist terms, there are cases where only higher-level properties, but not the lower-level properties underlying them, qualify as causes of a certain effect. In this article, I show that the line of reasoning supposed to establish this claim does not succeed and that interventionism is not better capable of dealing with higher-level causal claims than other accounts of causation. According to Woodward, higher-level causal claims are nonetheless more adequate than lower-level ones if they describe a realization-independent dependency relationship and, hence, meet the requirement that causes should be proportional to their effects. I argue, however, that combining interventionism with proportionality considerations raises difficulties and that, therefore, Woodward's account does not vindicate higher-level causation.

Acknowledgements

I would like to thank two anonymous referees of this journal for helpful comments on an earlier version of this article.

Notes

[1] Throughout the article, I follow Woodward (2003, 39) in using the notions of causing and being causally relevant to interchangeably.

[2] One might object that this misrepresents the issue. The question is whether N₁ competes with M₁ for causal relevance, that is, whether Anna's actually having N₁ is a cause of her actually having P₁, and not whether the set of properties represented by some variable N or N* is causally relevant to the set of properties represented by some variable P. This can be interpreted as the difference between type-causal and token-causal claims (see, however, Raatikainen 2010, 356n12): that N is causally relevant to P is a type-causal claim, whereas what is of interest with respect to the causal exclusion problem is actual causation, that is, token-causal claims. However, type-causal claims imply token-causal claims (at least as long as the causal structures under consideration do not involve overdetermination). N = N₁ is an actual cause of P = P₁ since there is a possible intervention changing the value of N from N₁ to a different value which changes the value of P from P₁ to a different value (for a more detailed account of the relationship between type-causal claims and token-causal claims, see Woodward 2003, 74–86). Accordingly, the observation that a variable N, having N₁ among its values, is causally relevant to P is sufficient for concluding that N₁ is a cause of P₁—and, hence, threatens M₁ with causal pre-emption.

[3] According to Baumgartner, if variables standing in a supervenience relation to each other are included in the same causal graph, this gives rise to an interventionist causal exclusion problem. According to the interventionist account, M can be causally relevant to P only if the value of M can be manipulated without changing the value of any other variable which is causally relevant to P, but not on the causal path between M and P. Since the value of M cannot be manipulated independently of the value of N, M does not satisfy this condition and, hence, does not qualify as a cause of P (Baumgartner 2009, 169–172; 2010, 366–370). Woodward argues that variables standing in the supervenience relation to each other may still be included in the same causal graph if the notion of an intervention is modified in an appropriate way. He admits, however, that this proposed modification does not provide a complete theory of how to deal with such cases (Woodward 2011, 22–25). This implies that the problems which interventionism encounters with respect to higher-level causation would even become worse if variables standing in the supervenience relation to each other were included in the same causal graph. Therefore, I do not pursue this line any further.

[4] According to Shapiro (2012), there is reason to doubt that such realization-independent dependency relationships exist. However, since I argue below that this notion raises problems within the interventionist framework anyway, I leave this consideration aside.