Manipulationism, Ceteris Paribus Laws, and the Bugbear of Background Knowledge

ABSTRACT

According to James Woodward’s manipulationism, to explain an event is to show how it could be changed by manipulating its cause. The relevant intervention must be a ‘serious possibility’, distinct from mere logical or physical possibility and approximating something I call ‘scientific possibility’. I argue that background knowledge is indispensable for judgements of scientific possibility and that ‘invariant’ generalisations, the primary vehicles of explanation in manipulationism, are not well adapted to encoding this often implicit knowledge, especially in the social sciences. A survey of key social scientific research methods (case and comparative studies, randomised control trials, ethnography, and structural equation modelling) shows that the output of these methods is generated by causal and non-causal background knowledge meshing in a way that is better encapsulated in an updated theory of ceteris paribus generalisations.

Acknowledgements

I would like to thank three anonymous referees for helpful and constructive comments on a previous version of this article.

Notes

1 Hempel (1965, 249) also considered the difference between explanation and prediction to be purely pragmatic.

2 See Woodward (2014) for scepticism about whether Salmon’s last proposal can work.

3 Earlier manipulationist accounts are Collingwood (1940) and von Wright (1971); contemporary contributions include Hausman (1998), Hitchcock (2001), as well as Woodward and Hitchcock (2003) and Hitchcock and Woodward (2003), and Ylikoski and Kuorikoski (2010).

4 Woodward in recent work uses ‘interventionist’ to refer to his theory of causation (Woodward 2009, 2014, 2015).

5 Woodward credits among others Frisch ([1938] 1995), Haavelmo (1944), Spirtes, Glymour, and Scheines (1993), and Pearl (2000) with different variants of the concept.

6 For criticism of this conception of ‘explanatory depth’, see e.g. Strevens (2009); Imbert (2013).

7 Woodward (2003), 79. Woodward defines a type variable X as a direct cause of type variable Y relative to a variable set V iff a possible intervention on X will change Y or its probability distribution while all other variables in V are held fixed at some value (Woodward 2003, 59, emphasis mine). Using this definition, he says that X’s taking value x is an actual cause of Y’s taking value y, iff X = x and Y = y (i.e. the change actually takes place) and ‘there is at least one route R from X to Y for which an intervention on X will change the value of Y or its probability distribution, given that other direct causes Z_i of Y that are not on this route have been fixed at their actual values’ (where ‘x’ and ‘y’ range over particulars, and ‘R’ refers to a directed path, defined as an ordered set of variables Z₁ … Z_n such that X is a direct cause of Z₁, Z_n is a direct cause of Y, and for all 1 ≤ i > n in between, Z_i is a direct cause of Z_i + 1; Woodward 2003, 77). He calls X a contributing cause of Y with respect to V iff ‘there is a directed path from X to Y’ (Woodward 2003, 57). The rather complex definition of ‘actual cause’ successfully explains how transitivity of actual causation fails in the one-boulder scenario. The fall of the boulder is an actual cause of the ducking, and the ducking an actual cause of survival, but the fall cannot be an actual cause of survival: given that the hiker ducked, an intervention on the boulder falling would make no difference to her survival; and there being (in that scenario) no other direct causes of survival on a route that bypasses her ducking, ‘there is no well-defined operation of fixing such variables’ in order to evaluate the influence of the boulder’s fall via the route that does not bypass it (Woodward 2003, 80; see also Hitchcock 2001, 276ff). The two-boulder scenario provides such a route, and needs to be ruled out. Some audiences, incidentally, declare no discomfort with calling the fall of the first boulder the actual cause of the hiker’s survival in that scenario; they appear however to be in the minority.

8 In a manner that Woodward (2003, 384n50) anticipated.

9 Adopting a strategy of ‘divide and conquer’, completer accounts seek to explain how the laws of nature could have exceptions without necessarily resolving all other outstanding issues with ‘laws’ (e.g. Fodor 1991, 22). They have been proposed in different forms by Hempel (1965; arguably also Hempel 1988), Hausman (1992), Fodor (1991), Pietroski and Rey (1995), Carrier (1998), Cartwright (2002), Kowalenko (2007, 2011, 2014), Steinberg, Layne, and Steinberg (2012), Strevens (2012, 2014), Pemberton and Cartwright (2014), among others.

10 See e.g. Mill (1843–1882, book 6, ch. 3); also Popper (1935, 1957), Nagel (1952), and Feigl and Broadbeck (1953, ch. 7).

11 Kuhn (1962) taught us, of course, that if the existence and constant operation of a hitherto unnoticed force were robustly confirmed in numerous subsequent observations across numerous different experiments, a theoretical and philosophical tipping point could be reached (all else being equal) that would force us to conclude that our initial model—and paradigm—must have been misspecified.

12 Others have provided strong arguments for the view that manipulationism is at odds with the methods of the social sciences. Boumans (2003), for example, points out that manipulationism is ill-equipped to account for inference from passive observation, a much more common practice in the social sciences than active manipulation; Cartwright (2007) says it gives a sufficient condition for causal inference only for ideal experimental settings far removed from the conditions under which it takes place in real life; Reiss (2007) notes that it assumes the existence in social contexts of robust causal relations that do not break down under intervention; and Russo (2011) claims it does not provide a suitable causal test in contexts in which interventions are not physically or ethically possible (see Russo 2014 for a summary). However, none of these authors contrasts manipulationism with a theory of explanation based on CP laws specifically in terms of its treatment of social scientific background knowledge.

13 The criteria used to establish the typology were themselves exception-prone. For example, the authors considered absence of ‘careful maintenance of the family budget’ to be a characteristic of one category of families, while acknowledging that all unemployed families were found to display varying degrees of irrational spending (Jahoda, Lazarsfeld, and Zeisel 1971, 55).

14 The question how the content of the CP clause in ‘CP₃ As cause Cs’ could be derived from the ceteris paribus conditions in ‘CP₁ All As are Bs’ and ‘CP₂ Bs cause Cs’ is, however, complicated. It seems that CP₃ cannot be simply the conjunction ‘CP₁ & CP₂’, for it is as unlikely that relations between CP clauses are summative as that relations between models are (cf. Kowalenko 2014, 149).

15 See applications of this idea e.g. in decision and learning theory: Boutilier, Bacchus, and Brafman (2001); Boutilier et al. (2004); McGeachie and Doyle (2004).