1 The Interventionist Account of Causation and Non-causal Association Laws Max Kistler Université Paris 1 mkistler@univ-paris1.fr Erkenntnis, March 2013, DOI 10.1007/s10670-013-9437-4. http://link.springer.com/article/10.1007/s10670-013-9437-4 Abstract The key idea of the interventionist account of causation is that a variable A causes a variable B if and only if B would change if A were manipulated in the appropriate way. This paper raises two problems for Woodward’s (2003) version of interventionism. The first is that the conditions it imposes are not sufficient for causation, because these conditions are also satisfied by non-causal relations of nomological dependence expressed in association laws. Such laws ground a relation of mutual manipulability that is incompatible with the asymmetry of causation. Several ways of defending the interventionist account are examined and found unsatisfying. The second problem is that it often seems to be impossible, in a model that contains variables linked by an association law, to satisfy the conditions imposed on interventions on such variables. Various ways to solve this second problem, most importantly the analysis of manipulability in terms of difference making, are examined. Given that none solves the problem, I conclude that the interventionist conditions are neither sufficient nor necessary for causation. It is suggested that they provide an analysis of nomological dependence, which may be supplemented with the notion of a causal process to yield an analysis of causation. 1. The interventionist analysis of causation According to the interventionist account, causation is a relation between variables. Its fundamental hypothesis is that a variable A causes a variable B if and only if it there are circumstances in which it is possible to manipulate B by intervening on A. According to Woodward (2003; 2008), this idea underlies scientific research for causes across all sciences. He gives the following example from social science. One can observe, in the contemporary US, a statistical correlation between children’s attendance of private schools (P) and their scholastic achievements (A). A randomized experiment would be a straightforward way by which a social scientist could try to find out whether this correlation stems from the fact that attendance of private schools causes better scholastic achievement or whether both variables are effects of some common cause, such as the parents’ higher socio-economic status (S). Such an experiment requires attributing children from a group of fixed S randomly to two sub-groups: one sub-group is sent to a public school, the other to a private school. This is equivalent to attributing one value of P to the individuals in the experimental group and another value to those in the control group. Making the attribution to the two subgroups random is intended to make it independent of any other factors that could influence A independently from P. After a suitable lapse of time, A is measured in the two subgroups. Any correlation that is found between A and P can be taken to reflect the existence of a causal influence of P on A. The possibility that A and P be the effects of some common cause such as S has been excluded by randomizing the attribution of a value to variable P for each individual. This is supposed to ensure that P is statistically uncorrelated with S, and indeed with any other variables that might be common causes of P and A.