The Philosophers' Cocoon

A safe and supportive forum for philosophers.

Owned & Moderated by Marcus Arvan (University of Tampa)

Contact: marvan@ut.edu

Blog mission & moderation policy

IMPORTANT NOTE on how to ensure anonymity in commenting
Job-Market Reporting Thread (2025-26)
Job-Market Discussion Thread (2025-26)

recent posts

about

A Very Different Worry About Counterfactual Theories of Causation

I've been trying to work out some longstanding concerns I've had with certain types of metaphysics, and counterfactual theories of causation in particular — and the resulting conversation led me to a new worry about counterfactual theories. The worry may be completely off-base, but I'd like to run it by everyone nonetheless, so that at least if I am off I can quickly learn why.

Here's the worry: counterfactual theories of causation aim to understand causation, obviously, in terms of certain counterfactual relationship. The standard semantic Lewis-Stalnaker analysis of counterfactuals, in turn, understands their truth-values in terms of what's true in relevant nearby possible worlds. So, 'if p were the case, q' is true just in case in the closest p-worlds, q is true; and 'if p were not the case, q would not be the case' is true just in case the closest ~p worlds are ~q worlds.

So, in order to analyze causation using counterfactuals, one must be able to account for which worlds are closer than which in "causal" cases.  But how is this to be done?

Suppose we adopt a hard-core Humean empiricism and so don't posit any kind of necessitation relation between constantly conjoined events — in other words, we adopt a Humean regularity view of events.  In that case, we have the problem of induction.  Even if all F's in the past have been followed by G's, the fact that they have all been in the past means that you don't have any empirical evidence that future F's will be followed by Gs. For all your (past) empirical evidence establishes, F's are regularly followed by G's only up until now, and something very different will happen from here on out (e.g. every F will be followed by G's absence). So, for example, even if every time a person has ever dropped a ball in the past, the ball accelerates downward under the force of gravity, on a Humean empiricism one has no evidence that the next time one drops a ball, the ball will go downward; it is just as likely, given your evidence, that the ball will shoot upward, or to the side; etc.

There's another way to put all this: on a Humean view of reality, a world in which a ball accelerates upward is, strictly speaking, no further away (or less like the actual world) than a world in which the ball accelerates downward. All worlds in which a ball does different things must be equally close to the actual world, for there is no necessary connection between events.

But now if that's true, then a counterfactual theory of causation can't get off the ground. One can't analyze Billy's rock-throw breaking a window in terms of anything like the counterfactual, "If Billy hadn't thrown the rock, the window wouldn't have broken" — for again, on a Humean view of reality, if Billy hadn't thrown the rock, worlds in which the window breaks are as close to worlds in which it doesn't.

How, then, can a counterfactual theorist get the relevant counterfactuals to come out right (such that it is in fact true that if Billy hadn't thrown his rock, the window hadn't broken)?  Answer: only by assuming some kind of necessary connection between events (such that there are objective grounds for thinking that a world in which Billy doens't throw his rock will be a world in which the window doesn't break). But in that case the counterfactual theory of causation presupposes a Kantian necessitarian metaphysics of causation — in which case the counterfactuals aren't causation itself but rather a reflection of an underlying necessitation of events.

Here, then, is the dilemma:

  1. Either the counterfactual theorist assumes no necessary connection of events, in which case a counterfactual theory of causation can't get off the ground (for necessary connections are needed to specify closeness of nearby worlds), or else 
  2. The counterfactual theorist assumes necessary connections between events, in which case counterfactuals don't explain causation but are instead explained by primitive causal necessitation.

Is there a way around the dilemma? Can the counterfactual theorist account for closeness of nearby worlds without assuming a necessitarian metaphysics? How? 

Posted in

12 responses to “A Very Different Worry About Counterfactual Theories of Causation”

  1. Daniel

    I don’t see the support for premise 1. In particular, I don’t see why the only way to specify closeness of nearby worlds involves appeal to necessary connections.
    In fact, it seems to me positively unattractive on grounds independent of Humeanism to try to ground closeness of worlds (in the sense relevant to counterfactuals) in facts about necessary connections. One reason is that counterfactuals are notoriously context-sensitive. Think Caesar, nukes, and catapults. Causal talk seems to inherit a lot of that context sensitivity–see Eric Swanson’s “Lessons from the Context Sensitivity of Causal Talk.” But presumably facts about which necessary connections between distinct existences hold are not context sensitive. So I’d doubt that you could get an extensionally adequate account of causation or counterfactuals if you gave a big role to context-insensitive necessary connections between events.
    One could dismiss all this, and say that the metaphysically interesting notions in the neighborhood of causation/counterfactuals won’t be context sensitive, even though the way we talk about them is. And I suspect that this may be a fair thing to say to a certain sort of analytic metaphysician who sees what they’re doing as being entirely “about the world” and not involving a hint of anthropology. But I tend to take a more pragmatist line (I like Huw Price a lot); maybe the best way to understand counterfactuals and causation is to understand how our talk of causes and counterfactuals actually works, and to maybe see why it has to be context sensitive in order to serve the role in our lives that it does.

    Reply
  2. Marcus Arvan

    Hi Dan: Thanks for your comment, which I think is very helpful in a lot of respects. My claim was that in the absence of deep metaphysical commitments — Humeanism or necessitarianism — there are no objective grounds for considering one world closer to the actual world than another. This is entirely consistent with your reply, which is that pragmatic considerations can be used to (contextually) decide which worlds are closer or further away from the actual world. But in that case, here’s my point: this is to import subjective, pragmatic considerations into inquiry. In which case now the counterfactual theorist has to say causation itself is constituted by our pragmatic, contextual-based concerns. It would really surprise me, though, to hear any counterfactual theorist on causation actually accept this — for I take it that (or at least hope that!) we can all agree that causation is something out there in the world, not just (or primarily) a pragmatically useful way of talking about the world. If you’re prepared to accept that, my argument has no force — but it seems to me a big bullet to bite.

    Reply
  3. Daniel

    I think Lewis would be happy to accept something like what you call a “big bullet to bite”. Namely, in any given context, “subjective”, pragmatic considerations in that context, determine which similarity relations we’re talking about in that context, and thus, which counterfactuals and causal claims made in that context are true. But that doesn’t obviously mean that causation facts about causation are constituted by our pragmatic, contextual based concerns. Maybe you’ll think the following distinction isn’t important, but it may be that our concerns determine which similarity relations our talk is sensitive to in any given context, but that once our concerns fix which similarity relations we’re talking about, they don’t do any further work in determining which counterfactuals are true.
    As for causation being “something out there in the world” rather than “a pragmatically useful way of talking about the world”, I think the contrast is probably much harder to draw than your remarks suggest, but to the extent that you can draw it, I don’t think it ought to be uncontroversial which side of the line causation falls on. You might find this paper by Huw Price interesting:
    http://prce.hu/w/preprints/bertinoro.pdf
    It addresses exactly the sort of worries you’re talking about, namely, that a broadly pragmatic approach to causation is “too subjective” or “not sufficiently realist” (his words). As he puts it: “I hope to show both that the view is not subjectivist in the sense most inimical to realism, and that to the extent that realism is an attractive position about causation, it is compatible with the kind of pragmatism I have in mind.” Section 7 in particular tries to draw a disanalogy between the sense in which it’s attractive to be realist about theoretical entities (e.g., electrons) on the one hand, and a different sense in which its attractive to be realist about causation and modal notions, on the other.

    Reply
  4. Lewis Powell

    As I understand it, the standard Lewisian approach to doing this is to take similarity of worlds as determined, in part, by similarity in the Humean regularities. Thus, the fact that worlds 1 and 2 are both worlds where (in fact) all Fs are followed by Gs is a feature that counts in favor of their similarity. World 3, in which almost all Fs are followed by Gs, but there are some exceptions, is, in that regard, less similar to world 1. So far, none of this relies on anything stronger than Humean regularities and Humean laws.
    Maybe you are suggesting that “similarity of which Humean regularities there are” isn’t really a kind of similarity? In which case, I think the Humean is on good footing to say, “well, it sure seems like something worlds can have in common or not”. There is a sort of conventionalism built in here. It is true that if we cared more about similarity of local facts than similarity of the patterns among those facts, the Humean has no way to say that the nomologically similar worlds 1 and 2 are “closer” to each other than the nomologically dissimilar worlds 1 and 3, but that is just to say that which similarity ordering is relevant for a given counterfactual claim depends on which similarities we are more interested in tracking, when we make those claims.

    Reply
  5. Marcus Arvan

    Hi Dan: Thanks for your comment, and for the link to the Price article. I look forward to reading it! In any case, I’m apt to agree that the distinction between “pragmatically useful” and “out there” isn’t as clean cut as my remarks suggest (we tend to think science gets at what’s out there in large part because it works in pragmatic ways — its predictions enable us to build computers, etc.). Still, until I read the Price article, it still seems “too pragmatic” to me. But I get the response. Thanks!

    Reply
  6. Marcus Arvan

    Thanks Lewis, that’s really helpful. I guess I’m still a bit skeptical that a true Humean can make that move for problem of induction reasons. After all, the problem is that even if all rock-throwing are in fact accompanied by window breakings, on a Humean empiricism it still seems to me an entirely open question whether, in any given individual case, the regularities can establish one counterfactual (e.g. If the rock weren’t thrown, the window wouldn’t have broke “) over another (“the window would have broken even without the rock”). For again, the regularities are just that: mere regularities. What in the world (from a Humean perspective) licenses a jump beyond what is observed to the positing of counterfactual relationships. In short, it still seems to me that counterfactuals must be grounded in something above and beyond mere regularities: say, necessitation or pragmatic concerns.

    Reply
  7. Lewis Powell

    Marcus, I think you are conflating an epistemological question about the grounding of induction with a metaphysical question about the space of possibility and the truth conditions for conditionals.
    The Humean says, if in fact all Fs in the actual world are followed by Gs then other worlds where all Fs are followed by Gs share the feature of having a Humean regularity of F’s being followed by Gs.
    Now, there is an entirely separate worry for Humeans, which is levied by Armstrong and Tooley (there is a third person I am forgetting who goes in this list), about whether Humean laws can underwrite induction. They complain, contra Lewis and other Humeans, that if laws are Humean, nothing would make it the case that we could project observed regularities into the future (while a more robust sense of the laws would license such projection). They suggest that Lewis has to assume that there are simple patterns in nature, to justify induction, and complain that this is an additional unsupported assumption of his theory.
    Lewis responds to this by pointing out that the anti-Humeans also need to posit simplicity, but in this case, simplicity of laws, rather than simplicity of patterns among local facts (think about a law like, “All Fs are followed by Gs for the first half of time, then followed by H’s for the second half” or the like).
    I don’t want to prosecute this dispute here, though I do think the anti-Humeans have some compelling things to say against Lewis’s parity argument there. Rather, my point is that the justification of induction is an epistemological question, rather than a metaphysical one. At worst, you are forcing the Lewisian to say that we can’t know whether the most similar world to ours is one where all Fs are followed by Gs (because we can’t know that our own world is one where all Fs are followed by Gs). But, this does not impugn the metaphysical determinacy of the claim that once you fix all the local facts in a given world, the Humean generalizations that count, for Lewis, as laws are also determined, and then the various possible worlds exhibit objective similarity relations in proportion to the degree to which they have similar patterns among their local facts. There is no induction at all playing a role at this stage of the Lewisian project.

    Reply
  8. Marcus Arvan

    Thanks, Lewis — sounds right. I think I was tacitly assuming that the Humean empiricist is a scientific anti-realist, in which case (I thought) the epistemological problem of induction has metaphysical implications. But I think I now see I was wrong about that. Thanks for setting me straight!

    Reply
  9. aVap

    Hi Marcus-
    [I tried posting this earlier, but either it got lost in the cyberspace (presumably due to length), or else it was too immoderate, and so was moderated away. If the latter, I apologize for re-posting. If the former, I’ll try breaking it up into parts.]
    There are a few different purposes a theory of causation might serve, and it seems to me that one of these cuts across these issues you’re having about causation in particular (Humean vs. non-Humean) and metaphysics in general (world-based vs. concept-based).
    Allow me to explain. Though one might think of David Lewis as a metaphysician with something quite opposite of a desert landscape ontology, his ontology is actually quite sparse in terms of types (rather than tokens). Basically, Lewis just has concrete individuals, worlds, and sets. (Arguably, worlds themselves aren’t even a separate type of commitment; for Lewis, worlds are basically just maximal mereological sums of concrete individuals, modulo spatial relations, etc.).
    Within this economy of types, of course, Lewis has a plenitudinous conception of tokens: every way a way world could be is a way some world is (modal realism); every possible combination of objects is itself an object (mereological universalism); and every way any group of objects can be collected is a set (sets are also maximally abundant).
    Given this economy of types and plenitudinous conception of tokens, Lewis professes to explain or do the work of pretty much every other ontology. Who needs universals when you have sets of possiblia? Who needs propositions when there are sets of worlds? And who needs non-Humean causal relations when you have worlds with highly similar patterns and regularities?
    cont’d…

    Reply
  10. aVap

    [cont’d from above]
    Put another way, part of the Lewisian project is show that given his resources (and only his resources), all the important philosophical work can be done. And what this amounts to, in practice, is showing that (virtually) all of our talk, and all of our intuitions, can essentially be modeled or accounted for given only his ontology.
    So, for example, you wonder exactly where the boundaries of a cloud, or a cat, are? Well, we just help ourselves to the mereological sum of every collection of atoms in the vicinity. Consequently, it is simply a matter of semantic decision (and so conventional) as to how to precisify ‘the cloud’ or ‘the cat’. But to ask where the boundaries of the cat are really is to ask a senseless (or unimportant) question (for each mereological sum really has the precise boundaries it has, and which one of these sums is the cat is conventional).
    Or perhaps you wonder whether the lump but not the statue can survive being smushed. Well, we have an infinite number of objects in other possible worlds, similar to the this-worldly lump, some of which survive the smushing, and some who don’t. And what we do here in this world is call the former sort ‘lumps’ and the latter sort ‘statues’. Thus, whether the statue/lump survives is also to some extent a matter of semantic decision- which existent objects are considered (lumpish) counterparts, and which are not. But again, to ask whether the statue really survives the squashing drops out of the picture.
    And the same goes, of course, for causation: is some event e really a cause of event f? Well, we have a plenitude of worlds- and in some, e’ (e’s counterpart) is followed by f’, and in some it isn’t. To the extent that worlds in which e is followed by f are (by some metric) “closer” to this world, we will then make the semantic decision to call e the cause of f (in this world). But to ask if e really causes f is to ask the wrong question- for what there is, really, is just a bunch of events (e.g. properties instantiated at regions), in a bunch of different worlds, forming various patterns, similar to others in some ways, and dissimilar in others. (And so whether some are even “really” closer to our may not even be crucial…)
    cont’d…

    Reply
  11. aVap

    cont’d from above…
    If something like this is right, I hope it puts into perspective some of the worries you’ve expressed about the counterfactual theories of causation in particular, and the concept/world-based metaphysics in general. For it’s not that the Lewisian starts with concepts, say, as opposed to the world. Rather, what is fundamental is a distinctive picture of the world: viz, that there (really) exists just individuals, worlds, and sets. But to show that that’s really all we need – and not e.g. universals, abstract possiblia, unique cats or clouds, lumps and statues with intrinsic persistence conditions, or non-Humean connections in addition- the nuts and bolts project is to show that all of our intuitions and theoretical desiderata can be modeled or recovered given only Lewisian resources- i.e., with only the world-stuff that Lewis thinks is ontologically cheaper, or less weird (such as sums of concrete individuals instead of unactualized possibles, or distinct existences and free recombination instead of non-Humean necessary connections).
    What I think can happen as a result, though, is that the project comes to look like conceptual analysis (over the course of doing Lewisian “normal science”, in the Kuhnian sense of the term). For now the project is to say e.g. ‘given that you intuitively think of causation like this, let’s see if we can satisfy those intuitions with a counterfactual model/analysis of causal claims (only in terms of worlds, individuals, and closeness, say) And to the extent that this sort of conceptual modeling can be done, then the original metaphysical thesis- that one need not posit non-Humean connections, say – is vindicated (and the need for positing non-Humean connection is obviated).
    Of course, many will argue, as you are, that something more is in fact needed: perhaps some non-Humean causal relation is needed to ground induction, or e.g. intensionally individuated universals or propositions do some work that sets of individuals or worlds cannot do. And if so, you might well think that there’s something else to causation besides counterfactual dependence. But here the objection is metaphysical, not meta-metaphysical, as you have seemed to suggest: that is, the objection concerns whether there are non-Humean connections, not whether the appropriate methodology is conceptual analysis. For on my way of reading this theory, the conceptual analysis methodology falls out of the metaphysical view, rather than supporting it.
    Now, I’ll grant you that many actual defenders of counter-factual theories of causation may not frame it quite this way, and perhaps others really are just doing conceptual analysis. But what I’ve tried to do is show it’s possible to defend the a methodological concern with concepts in a way that doesn’t reduce the metaphysics to the conceptual.
    I look forward to your response…

    Reply
  12. Marcus Arvan

    Hi aVap: Thanks so much for your detailed comment. I haven’t thought very carefully about Lewis’ overall metaphysical and semantic picture in over a decade (!), so I found your summary very helpful. It’s also clear that you’ve thought hard and deeply about it. Here are a few brief thoughts of mine in reply.
    While I can’t help but have admiration for Lewis’ overall picture — viz. there is just a plenum of actual and possible individuals (constituting worlds), sets, and finally, semantic decisions on our part — I have two general worries about it.
    My first worry is that it seems to obliterate any notion of “carving the world at its joints.” If I understand him correctly, there are (A) just swarms of particles that blend into one another (e.g. there is no clear boundary to a cloud), and (B) we make (vague) semantic decisions to call various swarms “clouds”, “tables”, “chairs”, etc. Is this right? If so, I have the worry that the picture is too “semanticized.” I think the world does have joints — some objects and properties are really “out there”, and some are simply things we semantically decide to call by various names and concepts.
    This brings me to my second (and related) worry, one that I evidently share with Douglas Kutach: which is that I think Lewis may be operating with an outmoded view of physics. Although I can’t really go into all of the details here, the picture of “particles in the void” — with no objective boundaries between objects — seems to me incompatible with modern physics. In modern particle physics, “particles” are really just minimum excited states of fields (e.g. the Higgs field, the electromagnetic field, etc.). So, they’re not really “points” in space (viz. the classic image of a point particle), and macro objects like clouds, tables, and chairs are not vague collections of atoms (as Lewis suggests). But I may be wrong about this. I’m sure Douglas could do a much better job of explaining this than I.
    In brief: I thank you for clearing up Lewis’ meta-metaphysical views, and showing how they might be used to resolve the worries I’ve tried to raise. At the same time, I’m still concerned about whether they’re the correct meta-metaphysical views to have. But I’m not sure I’m enough of a (meta-)metaphysician to make much more headway here. I would definitely encourage everyone to check out Douglas’ forthcoming book on causation through Oxford University Press. From the little I’ve read, it seems to me a very important step forward for metaphysics generally, and the metaphysics of causation in particular.

    Reply

Leave a Reply to Marcus ArvanCancel reply

Discover more from The Philosophers' Cocoon

Subscribe now to keep reading and get access to the full archive.

Continue reading