theory dependence of accounts of causation

Jeffrey Barrett has said that causation is a theory-specific concept. alternatively, that there is no one notion of causation which makes sense outside any particular physical theory. rather, he suggests that different physical theories have different notions of what causation is. this seems problematic, because it would seem that if causation is really something out there in the world, the question of whether a causes b should make sense independent of specific human theories. at the same time, notions of causality can and do change in response to developments in physics.

in this paper i seek to explore Barrett's claim further. i consider in some detail Daniel Hausman's account of causation and the manner in which he is forced to adapt it in response to different physical theories. finally, i give a fuller account of the ways that i believe causation is and is not theory dependent.


Barrett's claim is that causation is a theory-specific notion. but this is rather different from the usual statement that some notion is theory-specific. for example, one might note that mass is a theory-specific notion, having somewhat different character in a newtonian theory from an einsteinian one. there is no theory-neutral account of what mass is; rather, mass is one of a great complex of ideas that form particular theories. the question of what mass is in a newtonian theory depends on the interrelationship of that concept to the other ones in newtonian theory and there is no absolute sense to the question.

the situation is not that Einstein and Newton offered different answers to the question what is mass?—rather, they both include different concepts each called mass which have only loose analogic relations to each other. a later physical theory might well have no need for anything closely resembling newtonian or einsteinian mass.

but if causation is like this, we run into rather a difficult bump. neither newtonian theory nor einsteinian theory, nor any other, include a term causation. causation, unlike force, mass, phlogiston, and the like, is usually taken to be a metaphysical term rather than a physical one. moreover, in a totalizing deterministic theory, it seems initially that there is no room whatsoever for anything approaching traditional metaphysical notions of causation.

one could augment physical theories with specific metaphysical models, and then causation might well form part of those models. given notions of causation could then be relative to specific models, and thus in an indirect way, to the physical theories they in turn model.

i propose now to turn to the particular account of creation that Daniel Hausman gives in his book causal assymmetries. in two ways, the account that Hausman gives needs to be adjusted in response to specific theoretical developments in physics. we can see then a sort of theory-dependence in practice, and better evaluate the nature of the theory-dependence of causation.


Hausman analyzes causation into two components: causal connection, and causal priority. a causal connection is taken to be the undefined intuitive notion of a nomological linkage.

we then have the first important posit, the connection principle [cc]:

cc: for all events a and b, a and b are causally connected if and only if they are distinct and either a causes b, b causes a, or a
[cc] justifies calling this notion causal connection, by giving the three sorts of connection that count, asserting that causation and nomological regularity are coupled. [cc] thus plays a role similar to the humean notion of constant conjunction.

but [cc] is not itself enough to understand causes. it tells us that when we see a causal connection (a nomological linkage) we should expect some causation, but it doesn't tell us which of the three possibilities obtains. leaving aside the question of being effects of a common cause, the central problem is that causation is an asymmetric relation, but nomological linkage is not.

Hausman discusses many sorts of asymmetry that causation has, and settles on one particular asymmetry as being the cardinal attribute of causation, from which the others can be derived. he begins by considering transitivity: that (properly described) if a causes b and b causes c, we must have that a causes c. this establishes a necessary condition on causes: a causes b only if a and b are causally connected and everything ... connected to a is connected to b. (p. 62) if we know that a and b are causally connected, and we further know that b and c are causally connected, but a and c are not connected, we conclude that it is not possible for b to cause a.

the presence of the third event c, connected to b but not a, thus breaks the symmetry of the causal connection between a and b. but this is not by itself enough to give a sufficient condition for causality. but if we knew the independence condition [i], we would:

i: If a causes b or a and b are causally connected only as effects of a common cause, then b has a cause that is distinct from a and not causally connected to a. (p. 64)
if we have [i] and [cc], we have a sufficient condition for causation: if a and b are causally connected and nothing causally connected to a is independent of b, then a causes b (p. 70). combining the two sufficient and necessary conditions gives us [cp], the independence theory of causal priority:
cp: a causes b if and only if a is causally connected to b and everything causally connected to a and distinct from b is causally connected to b. (p. 70)
but why should we believe [i]? there is no good reason to. in Hausman's words, as a metaphysical claim about patterns of lawlike connections found in nature, [i] seems incredible, and its truth miraculous (p. 64). instead, [i] has a very different status. according to Hausman, if [i] obtains, then we can meaningfully identify causes and use causal explanations. when [i] does not obtain, there is no causal asymmetry, and causal explanations cannot be given (p. 64). Hausman devotes considerable space to [i], showing that [i] is implicated in other philosophical accounts of causation.

if one is dealing with a totalizing physical theory, however, it is generally likely that everything will be causally connected to everything else, at the least, as effects of a common cause. this makes [i] unusable, and we become unable to talk satisfactorily about causation. Hausman deals with this by the notion of a causal field. a causal field is something like a domain under discussion, relative to which external common causes are ignored.

Hausman uses Elliott Sober's example of bread prices in england and water levels in venice. both are rising. but there is no causal connection between them, we want to say. however, from a broad enough perspective, they are in fact both effects of a common cause: the continuing influx of energy from the sun.

Hausman treats this with the notion of a causal field: relative to the terrestrial short-term causal field, the system within which we are thinking, there is no causal connection between these two increasing series because there is in fact no terrestrial short-term common cause of the two series.

another challenge to the account thus far comes from probabilistic theories. indeed, this is a famously contentious area in discussion of causation. Hausman's answer is that we don't need the notion of probabilistic causes; we can fully make do with the deterministic causation of probabilities.

without going into detail, one significant advantage of this account is that it is univocal with respect to causation: all causation is deterministic, all causation is of a piece and understandable on the same conceptual terms. Hausman details many other important advantages he finds in his theory as well.

for my purposes, i wish merely to point out that one rhetorical motivation for Hausman's consideration of probability at all is that according to contemporary physics, many [sic] occurrences are not determined (p. 185).

indeed, the most serious obstacle for any account of causation will be how it deals with epr-like experiments in quantum mechanics. Hausman's last topic in his book is to address epr experiments. these pose a significant problem for his account, because in epr experiments we seem to have lawlike regularities in spin correlations which cannot be explained as effects of a common cause, nor as mutually caused. this would seem to contradict [cc].

he addresses this problem by modifying the theory in three ways. first, we replace the notion of causal connection by nomic connection, which is intended to have the same intuitive content; and then [cc] is replaced by [cc']:

cc': a is nomically connected to b if and only if a and b are distinct and either a causes b, b causes a, a and b are effects of a common cause, or a and b are mutually dependent. (p. 252)

if the notion of mutual dependence were too broad, then this modification would remove all content from [cc']. but if we restrict mutual dependency to relations of strong symmetry, then [cc'] is still able (says Hausman) to fill the role in the theory that [cc] did. so we add the axiom [ssmd]:

ssmd: mutual dependence is symmetrical, and if a and b are mutually dependent, then everything nomically connected to a and distinct from b is nomically connected to b. (p. 253)

as Hausman puts it, strong symmetry implies that if a and b are mutually dependent, then they will have the same causes and effects (p. 253). so the new case in [cc'] does not open the gates to any kind of relation whatsoever, but has some specific narrow content.

this is satisfactory to handle epr-like experiments. the peculiar correlations which did not fit in [cc] but surely are nomically connected, do fit the requirements of [ssmd]. armed with these two modifications, the theory as a whole gets its restatement in [cp']:

cp': a causes b if and only if a is nomically connected to b, everything nomically connected to a and distinct from b is nomically connected to b, and something nomically connected to b is independent of a. (p. 253f)
here the notion of nomic connection has replaced the causal connection of [cc']. as Hausman puts it the separate measurements in [epr-like experiments] are nomically connected, but they are not related as cause and effect or as effects of a common cause. causal explanation is out of place because the separate measurements do not have independent causes (p. 253).

this concludes my brief tour through Hausman's account. i now turn to the task of relating his theory to Barrett's claim that accounts of causation are theory-relative.


in several ways we have seen that the development of Hausman's theory changes in response to the details of the physical theories concerned. upon examination, i intend to show that these modifications form a sort of evolutionary lineage, in which the account of causation must be adapted as new physical theories come on the scene, but without the need to jettison entirely what came before.

the notion of a totalizing physical theory is relatively recent. it is unclear whether Newton imagined such a theory, though the notion was certainly in place by the time of Laplace. totalizing deterministic theories wreak havoc with accounts of causation, and Hausman's is no exception. his theory hinges upon [i], but [i] simply fails outright for any plausible totalizing deterministic theory.

the rescue is to relativize to causal fields. we then have that a can be a cause of b in one causal field, but not in a larger field. in any causal field which can be described fully deterministically, we will generally find it impossible to talk of causes under Hausman's account. this means that, if such a theory is true for the universe as a whole, we cannot talk about any absolute sense of causation, and, therefore, we cannot sensibly ask if a causes b full stop.

next comes the consideration of probability. probability is a common topic for accounts of causation, and Hausman's principal project here is to show that there is no need for probabilistic causation: that traditional notion of necessary connection does just as well in probabilistic contexts and in deterministic ones. he is clearly motivated to do this by the existence of a physical theory which (under some interpretations) is thoroughgoingly indeterministic.

most examples used to discuss the relationship of probability to causation are mundane and do not seem to implicate quantum theory. but those examples are usually trotted out by advocates of probabilistic causation. if all bottom-level physical theories were strictly deterministic, then Hausman would have no need to ever broach the subject. all probability would be merely epistemic doubt. Hausman wants to show there is no need for a concept of probabilistic causation. epistemic considerations play no role in his theory; if all probability is only epistemic doubt, then his case would be made quickly.

but all doubt is not merely epistemic, thanks to the unexpected advent of quantum mechanics. and so Hausman is forced to consider in great detail the question of probabilistic causation. the easy dismissal is not open to him. as it happens, he shows that his theory has an account of probabilistic situations. here, no change turns out to have been needed in the theory. yet, the account in its presentation must do considerably more work.

quantum mechanics is not merely probabilistic, but also has peculiar nonlocalities in its probabilities, which give rise to further puzzles for causation. epr-like situations directly contradict crucial presuppositions of Hausman's account. but with a gentle twist, the structure of his account can adapt to the subtleties of quantum mechanics.

i have tried to make the case that this is not the only way in which his theory needed to alter and shift in response to different physical theories. it is certainly the most obvious one (because it forces the rewriting of key propositions), but both with the consideration of totalizing deterministic theories, and then with deeply probabilistic theories, he was also force to adapt his account. and, importantly, note that the final form of the account might have to adapt still further, if a later theory allows for non-symmetric mutual dependence.

the need to adapt over time in response to changes in physical theories is not unique to Hausman's account. it is not because of some special defect or deficiency that his account must alter and shift as physical theories arise. for example, the minor miracles of Lewis's theory also require changes and alterations depending on the nature of the physical theory held.

this points out one important sense in which Barrett's claim might be misleading. it might be taken to suggest that there is such a thing as a quantum mechanical account of causation, and a general relativistic account of causation, and a galilean account of causation. Lewis's and Hausman's accounts, and most others, all have ways of addressing the peculiarities of quantum mechanics. none is so obviously superior that it could claim to be the quantum mechanical account of causation.

this is markedly different from the normal notion of theory-relative terms. there is such a thing as the general relativistic account of mass. mass is a theory-relative concept, but for a given theory, there is one thing which is that theory's account of mass.

there are still reasons to prefer one account over another. Hausman can plausibly believe that his account is the true account of causation, and then the hausmanian account of causation in quantum mechanics is the true one, if quantum mechanics is the true theory of physics. Lewis can also have the same belief, and the decision between Hausman and Lewis will take place on familiar philosophical grounds, not on physical ones.

however, as new physical theories arise, certain accounts of causation might become so strained by ad hoc modification that they become progressively implausible. (some might say that Hausman's account of epr-like experiments has already passed that point.) in addition, there might be more than one way to adapt a given account in response to a new theory. i envision such adaptations branching out, with some getting killed off because of growing implausibility; this further justifies the use of an evolutionary metaphor.

we may have some hope that the true physical theory, the goal of our physical investigations, only fits well with a single account of causation: and that this account of causation could then be reasonably termed the true account of causation. such an account would still be theory-relative, and indeed, it would take different forms between the ultimate true theory of physics and its predecessor inexact theories. but in a deeper sense, it would promise to give a correct absolute understanding of causation, relative only to the truth and not to particular faulty physical descriptions.


all page references are to
Daniel M. Hausman, causal asymmetries, cambridge university press, 1998.

things i wrote for my m.a.