To: Mark S.
From: Geoffrey Klempner
Subject: Counterfactual analysis of causal connection
Date: 7th January 2010 13:50

Dear Mark,

Thank you for your email of 18 December, with your essay for the University of London Logic module, in response to the question, 'What might or might not have happened does not explain what had to happen. So we cannot rely on the former to give a satisfactory analysis of causal connection.' Discuss.

You have written an excellent answer to what is, in my view, a rather indifferent question. The statement, interpreted as an argument against the counterfactual analysis of causation, doesn't even look plausible. What reason is there to think that you can't explain the necessity of P (it's not being the case that it might have been the case that not-P) in terms of 'what might or might not have happened'? A basic ground rule for composing this kind of question is that the 'So...' has at least a prima facie claim to plausibility.

As always, my strategy would be to focus on the question. Why would anyone think the claim was plausible? or even made sense, for that matter?

However, I would probably end up saying more or less what you say. You organize your answer into two separate sections. The first deals with obvious counterexamples to the simple counterfactual analysis of causation and responses to those counterexamples. The second section (which I wish had been longer) raises the more foundational objection that our intuitions regarding the truth or falsity of counterfactual conditionals depend on our prior grasp of causation.

Regarding the details of the counterfactual analysis, there does seem to be a basic methodological problem -- which, being charitable, one might regard the question as hinting at -- of distinguishing causes from background conditions. How do you pick out 'the' cause? If all sorts of conditions, both large and small were needed in order for e to occur, why pick on c? Why not c' or c''?

Cases of overdetermination, pre-emption etc. are additional problems, but the fundamental difficulty is that our intuitions tell us that a 'cause' is more than a 'condition'. The driver was distracted by his mobile phone a second before the pedestrian walked onto the road. It's obvious who is to blame, and what caused the pedestrian's death. The death could have been avoided if the mechanic at the last service had paid more careful attention to the brakes, but that doesn't give us the slightest motivation to point the finger at the mechanic.

It's tempting to say, as you do, that the notion of causation is 'primitive' but I think that's far too easy a solution. Moreover, for all the cases where it is clear how one identifies a cause, there are just as many others where it is far from clear. Causation is something we need to understand, just as much as it is something that enables us to understand other things (like counterfactual conditionals).

Which brings us to the second section, and your claim that the counterfactual analysis of causation is circular. I would have said more here, in particular about Lewises total reliance on an unanalysed notion of 'similarity'. Goodman in 'Languages of Art' mounts a devastating critique of the idea that we know what it is for some thing to be 'similar' to or 'resemble' another. Similarity is useless as a logically primitive notion, especially so in the possible world analysis of counterfactuals.

There is a solution, but it requires going back on the model of analysis which you give at the beginning of your essay. If A is 'analysed' in terms of B, then what we expect is a biconditional, where A is on the left hand side, and the right hand side refers only to B. This is the model of 'reductive' analysis.

A possible alternative in the case of causation would be to see causal relations and counterfactual conditionals as co-primitive. To analyse a causal statement, 'c caused e' in terms of counterfactuals is, in effect, to analyse it in terms of other causal statements. To analyse a counterfactual statement in terms of alternative causal chains, is to analyse it in terms of other counterfactual statements.

You mention the contextual dependence of counterfactual statements. It is the context which tells us which are the relevant factors to vary when we consider the possible world where the antecedent is true. Explanation is relative to interest (Putnam). So is causation, for exactly the same reasons.

Meanwhile, if the earlier Davidson is right (as I think he is) in claiming that events are individuated by their causal relations, then what counts as an 'event' is similarly contextual.

Our intuitions about counterfactuals are messy. I don't want to say that counterfactual statements, or statements about causal relations can't be 'true', or have 'truth conditions'. Which only goes to show that 'truth' is not a simple notion -- not all truths are the same, or have the same status. At one end of the spectrum is pure make-believe, while at the other end are theories considered worth publishing in a scientific journal.

All the best,

Geoffrey