How about the three rockets being a case of both joint causation and overdetermination? Each rocket contributes 2/3 to the joint causation of the destruction of the planet and 1/3 to the overdetermination. I guess I see this as helpful in that it not only does it preserve the salient facts (that there were more than enough rockets to do the job and that no single rocket would have sufficed), but also how much was necessary and how much, er, overkill. Fractional causation also seems easier to reason about than n-tuples of counterfactuals.

That sounds like it works, but most of the uses of ‘overdetermination’ in the philosophical literature seem to assume that it’s an on/off concept. I don’t know how one would rewrite exclusion arguments under the assumption that something can be partially an over-determiner and partially a joint cause.

In the 2-rocket example, you have joint causation of the confusion of the system and so joint causation of the failure of the defense mechanism by F1 and F2. The two rockets get through once the failure has occurred, so the failure allows the rockets to continue on their paths and then overdetermine the eventual destruction of the planet. So you have joint causation at an early stage of a causal process that allows two other causal processes (one initiated by F1 and one by F2) to continue uninterrupted and overdetermine a later event. Seems OK to me—why does this seem odd to you?

I too am puzzled about what overdetermination is supposed to be, but for a simple (perhaps simple-minded) reason that doesn’t require clever examples to illustrate. Consider (O1) and (O2):

(O1) If c1 had occurred and c2 had not, e would (still) have occurred.
(O2) If c2 had occurred and c1 had not, e would (still) have occurred.

Take the textbook firing squad example: c1 and c2 are two shots and e is the victim’s death. In her brilliant paper on the exclusion problem, Karen Bennett says that to decide whether or not the death was overdetermined, we would ask whether these counterfactuals are true: Would the victim have died if the first gunman had fired without the second? The second without the first? Suppose the answers are ‘yes’. It hardly follows that the same death occurred. That the victim died is a fact that could have obtained by virtue of the occurrence of either e, e1, or e2, where e1 (e2) is the perhaps very similar death that would have been caused by c1 (c2) alone. Why suppose that e = e1 = e2? This seems unrealistic, and an argument is needed for it. I’m not assuming essentialism about events’ causes, but I guess I am assuming some sort of essentialism about the properties of events. That is, on each scenario the death would have had different properties, and I see no particular reason to suppose that it would have been numerically the same death that occurred on each scenario. (Bennett is trying to remain neutral on the nature of events (n.27), but it seems to me she’s being too neutral.)

Anyhow, I can’t come up with a clear and realistic case of overdetermination that isn’t really a case of joint causation, i.e., where the two causes don’t partly contribute to the determination of the actual effect, even if merely one of the causes would have caused an effect of much the same sort. Or else the case is one of pre-emption, not overdetermination. This is probably a familiar point to you experts on causation. So please tell me what I’m missing and where to find it.

I grant that there is a weaker sort of overdetermination, where ‘e’ in (O1) and (O2) is replaced by ‘an event very similar to e’, and maybe this all that people who worry about the exclusion and related problems really care about.

Considerations of this sort led Paul Humphreys to doubt that there is any genuine overdetermination. He says, “A death by simultaneous action of cyanide and strychnine is a different kind of death from one by either poison alone” (*Scientific Explanation*, ed. Kitcher and Salmon [1989], p. 301). He adds, “A factor which left no trace on the effect would have contributed nothing to that effect,” and so would not really be a cause of it. One can quibble: isn’t it possible for something to leave a trace on the effect without leaving a unique or distinctive trace on it, by leaving exactly the same trace as some other cause? But I too find it hard to come up with a clear and realistic case of this. I guess it’s easy to do so if one allows for determinable effects, so that, say, counts as an effect, distinct from . This is a departure from neutrality about events (unless one takes something other than events, like facts, as the causal relata).

Sorry about the mangled second-last sentence. It should read: so that, say, the-button’s-being-pushed counts as a distinct effect from its-being-pushed-with-just-that-force,etc.

Please forgive this post - my intuitions about overdetermination have been brought into question lately. Nevertheless, for some reason I cannot resist responding …
I don’t see how the three rockets example is a case of overdetermination at all. For overdetermination to occur, it was my understanding that you need (at least) two sufficient causes for the same effect, so that if one of the causes didn’t occur, the effect still would.
But with the three rockets example, I’m only counting one sufficient cause. The third rocket is overkill, but it would not be sufficient to cause the effect all on its own. Maybe what your example shows is that the intuitions behind overdetermination aren’t adequately expressed by (O1) and (O2), because your example can get around them. But then, maybe your example can’t. It depends on how you look at it:
In identifying the first sufficient cause, you say this: (O1a) If c1 and c2 had occurred and c3 had not, e would (still) have occurred. So we see that c1 and c2 together form a sufficient cause. But where’s the second? Presumably, in what we omitted. So we reverse the formula: (O3b) If c3 had occurred and c1 and c2 had not, e would (still) have occurred. This is false. So c3 is not a sufficient cause, and we’re left with only one in this case. For any sufficient cause you take (c1 and c2, or c1 and c3, or c2 and c3) you will find that you’ve got only one in that situation.
Maybe you’re thinking that the three sufficient causes I just mentioned (c1 and c2, c1 and c3, and c2 and c3) make this a case of overdetermination. I’m not quite sure what to say to that - it seems incorrect to me, because those are simply different ways of describing the same situation. Saying they are 3 distinct sufficient causes is like looking at a foot long ruler and, after adding the first 10 inches and the last 10 inches, saying that there are 20 inches there.
But then, I could be completely off. Am I?

This is pretty belated but, in my defense, I just came across this page. In response to Shieva I would ask, what about the case where the third rocket is a sufficient cause—say it’s very well-armored and the defense system barely scratches it. Then it’s true that, if c1 and c2 occurred, and not c3, e would occur, and it’s true that, if c3 occurred, and not c1 and c2, e would occur. So it looks like there are two sufficient causes, (c1 and c2) and c3, so it’s overdetermined (and I think that makes a lot of sense).

But that only works if you choose that grouping. If the heavy-duty rocket were fired first, along with with one of the ordinary rockets, there would be the same kind of joint causation leading to overdetermination as is mentioned in the main article. But then when you add the third ordinary rocket in, you get the result that if c1 and c2 occurred, but not c3, e would have occurred, but if c3 occurred, and not c1 and c2, e would not have occurred—even though the same three rockets are fired. The reason this happens is that in each case you’re essentially considering two causes, one of them being compound, and which two causes are compounded leaving which cause behind makes a difference to the result, even though it doesn’t seem as if it should intuitively.

So I would suggest as a generalization this: if there are n causes, then if you can partition that group of causes into two groups of p and n-p causes such that the groups satisfy (O1) and (O2), then those groups of (possibly joint) causes overdetermine the effect.

A nice example to test this on is elections under majority voting. Suppose that candidate A wins with a majority greater than one. If overdetermination has a meaning, I would want this to be an instance.