Counterfactuals and temporal direction

A forward counterfactual conditional is one whose consequent is about a later time than any the antecedent is about: ‘If Stevenson had been President in 1953, the Viet Nam war would not have escalated in the 1960s.’ The consequent of a backward conditional is about a time earlier than any that its antecedent is about: ‘If Stevenson had been President in 1953, he would have won the election in 1952.’ In this paper I shall offer a partial analysis which shows there to be no difficulty about allowing forward and backward conditionals in the same breath, as it were. This is in sharp contrast with a theory of David Lewis’s1 which provides for forward conditionals but not backward ones: when we conditionalize from times to earlier times, Lewis thinks, we adopt standards which are not those we use for forward conditionals and which he does not undertake to describe. My theory also contrasts with Frank Jackson’s, in which each kind of conditional is given an official theoretic basis, but the bases are different and we are not allowed to combine backward and forward conditionals freely in a single operation. In short, Lewis’s theory is asymmetrical, Jackson’s is symmetrical but split down the middle, and mine is unified and symmetrical. In expounding various theories, I shall assume that something like this is true: (P < Q) is true iff Q is true at all the P-worlds which are closest to the actual world. What marks off one theory from another is its view about what makes a world a ‘closest’ P-world. Despite the superlative form of the word, I do not take it for granted that closeness is a matter of degree and that ‘closest’ means ‘more close than any other.’ Lewis thinks that closeness is a matter of degree, but Jackson doesn’t, and I am not sure. More about that in section 15. On some theories in which closeness is a matter of degree, there are no closest P-worlds, even if ’closest’ means merely that none are closer. See Lewis’s Counterfactuals 1.4 for a clear pointer to how all my uses of ‘closest P-worlds’ could be adjusted so as to take in those theories as well.