9 thoughts on “Correlation and causation

  1. That probably depends on your willingness to accept non-strict causal explanations as giving real insight into underlying causes. (That sentence is badly worded.)

    If you think that the only true causal statements proceed from exceptionless laws, then there had better be a correlation between the cause and the effect. If you’re willing to tolerate statements with implicit ceteris paribus clauses, then there could in principle be very little actually empirically observable correlation.

    For instance, you may be willing to accept as true the statement “being submerged in water causes sugar cubes to dissolve”. However, it is not necessarily true that there is a high correlation between sugar cubes’ being submerged in water and the subsequent dissolution of the submerged cube. For instance, the water might already be saturated, or the sugar might be covered in wax; in either case, the sugar won’t dissolve. Because of this, dedicated pedants can affect the correlation between being submerged in water and dissolving. That, however, doesn’t, or at least shouldn’t, affect the truth of the statement about being submerged, “for it has the capacity in so far as it is a capacity for acting, and that is not in any and every condition, but just in some circumstances, in which external things preventing will be ruled out as well; for these are set aside by some of the things present in the specification of the capacity”.

    Then there will be a correlation between submersion and dissolution in (not fully specifiable in advance) the right conditions, because in those conditions the effect is guaranteed, given the occurrence of the event described as the cause. But plainly an event under that description can occur without the event described as the effect.

    Of course, you might think that the statement about dissolving is really just a heuristic, and what we do in experimentation, for instance, is find out what the right circumstances are, so that eventually we can either: (a) formulate a new statement at the same level of abstraction with explicit defeating conditions enumerated (“being submerged in water causes sugar cubes to dissolve, unless …”) or (b) move to a lower level of abstraction, talking say about interactions at the molecular or atomic or subatomic level, where, hopefully, there are no defeating conditions to be worried about.

    And, of course, you might think (not implausibly) that even if we don’t think of the above statement as just a heuristic, the existence of any particular causal link between events implies that there’s some exceptionless law relating them, even though we might not presently know what that is. In that case, causation would again imply correlation. (In fact, the view quoted above seems to have that implication—if the explanation of the sugar’s dissolving is that the sugar has the capacity to dissolve, and that capacity was exercised, then the sugar and the water came together in the right way, and when they come together in that way or relevantly similar ways, the effect necessarily occurs.*)

    In conclusion, this comment is quite rambly.

    *”It is not the case, then, that sugar has the property of being-soluble-under-all-circumstances-provided-nothing-external-interferes. Rather it has under all circumstances the property of being-soluble-under-certain-special-circumstances, namely, the ones under which it will invariably dissolve”, sez Moline in “Provided Nothing External Interferes”, p 253.

  2. Causal relations can be conceived of as “one-off”; the conditions for causal generalizations (linguistic items, if you follow, say Davidson) are time order, correlation (sometimes termed association) and non-spuriousness. So, yes, I think there is conceptual space at least for the idea that relation between event a and event b can be causal without it being such that there is a causal generalization between events of type A and events of type B.

  3. So, yes, I think there is conceptual space at least for the idea that relation between event a and event b can be causal without it being such that there is a causal generalization between events of type A and events of type B.

    This is undoubtedly true. Davidson has a pretty clear example of this in I think “Thinking Causes”; you could have a table indexed by row and column whose entries describe events. You could then truthfully say something like this: “(9,16) caused (7,1)” without that being any support at all for the generalization “events of type (x2, y2) cause events of type (x+y, y-x)”.

  4. I read Davidson much more drastically: if we ordinarily speak of an event caused by another event this has nothing to do with correlation. In fact, correlation is a statistic notion and in everyday life we speak about events caused so they are in two very different places. The link correlation-to-causation is entirely due to a technical use of the word causation to stand for physical laws that apply ceteris paribus. Sure, when we attribute a certain cause to a certain event we use the background knowledge of (supposed) universal laws but these are never sufficient to explain or justify such an attribution. Shorter: Hume.

  5. That sort of plays havoc with the nomological character of causation as assumed in “Mental Events” and defended in “Laws and Cause”, doesn’t it?

  6. Any cause that cannot be mathematically expressed cannot be said to be “correlated.” If we can think a cause that is not so quantified, then we can think your counter-example.

Comments are closed.