First, some background. Hare argues that identity theorists should hold that who we take to exist in a counterfactual world (or state of affairs) depends on which world is actual. As a toy example, suppose that we have three worlds that are qualitatively identical except that a boy called 'Jack' in one is replaced by a better-off girl called 'Jill' in another, whereas the third world instead contains a gender-indeterminate amalgam of the two called 'Janus' (whose welfare level is similarly in between the other two}. Further suppose that a counterpart theorist would say that Janus is a counterpart of Jack and of Jill, but Jack and Jill are not counterparts of each other. An identity theorist can correspondingly claim the following:
(1) If W_Jack is actual, then W_Janus contains Jack but W_Jill does not. (So W_Janus pareto dominates W_Jack, but W_Jill is separate.)
(2) If W_Janus is actual, then W_Jack and W_Jill both contain Janus. (So W_Jill pareto dominates W_Janus, which in turn pareto dominates W_Jack.)
(3) If W_Jill is actual, then W_Janus contains Jill but W_Jack does not. (So W_Jill pareto dominates W_Janus, but W_Jack is separate.)
Hare illustrates the first and last cases as follows (where, for maximal qualitative propositions Pi, Pk, "Pi@Pk" denotes "the state of affairs that, supposing Pk is true, would have come about if Pi had been true"):
At this point, the partialist may plausibly insist that what preferences you ought to have depends on which states of affairs actually pareto dominate others, which in turn depends on which world is actual. So, e.g., if W_Jack is actual, we're required to prefer W_Janus, but we need not go all the way to preferring W_Jill.
Hare's argument against this partialist view appeals to the (putative) principle of rationality that your preferences supervene on the qualitative (PSQ):
Whether or not you prefer one state of affairs to another supervenes on the qualitative features of the state of affairs. There are no states of affairs S1, S2, S1*, S2*, maximally specific qualitative propositions P1, P2, such that P1 accurately describes S1 and S1*, P2 accurately describes S2 and S2*, and yet you prefer S1 to S2, and do not prefer S1* to S2*. (p.186)
Since P_Jill@P_Jack is qualitatively identical to P_Jill@P_Jill, and P_Janus@P_Jack is qualitatively identical to P_Janus@P_Jill, and we must prefer P_Jill@P_Jill to P_Janus@P_Jill, we are thus likewise committed to preferring P_Jill@P_Jack to P_Janus@P_Jack (and hence to P_Jack@P_Jack).
It's a clever argument. But is PSQ really a requirement of rationality? To defend it, Hare imagines a violator of PSQ who takes non-qualitative haecceities to matter morally:
"You have told me, in glorious detail, all the qualitative facts about [the options], but I need to know some further, irreducibly haecceitistic facts about them in order to know which I prefer. For me, these irreducibly haecceitistic facts matter.” I feel about such a person the way Parfit feels about someone who is future-Tuesday-indifferent. He is loopy. He takes things to matter that just don’t.
I have two objections to this. Firstly, I wouldn't be so quick to dismiss the potential moral significance of haecceities. (It reminds me of those bad arguments against moral realism that ask why we should care about non-natural normative properties.) Of course, if all we know about them is that they are non-qualitative properties of some sort, then it'd be mysterious why they matter. But the haecceitist claims more than this: according to them, haecceities are the essential locus of our individual identities. And it isn't obviously "loopy" to think that the identities of affected individuals matters -- that's just the partialist's view, after all! If they didn't think identity mattered, they would be impartialists already. (Of course, we may doubt whether identity consists in non-qualitative properties, but I take it that's to question whether haecceitism is true, not whether it would matter if true.)
Secondly, and more importantly, you can reject PSQ without relying on non-qualitative haecceities. To see this, consider the following counterexample (developed from related suggestions by Adam Elga and especially Helen):
We're in a mirror universe, exactly symmetrical until time t at which one child will get tortured (either harshly or moderately), whereas their mirror duplicate will not. And suppose I'm the parent of one of these children. Then the following states of affairs constitute a counterexample to PSQ:
S1: My child tortured a bit
S1* = S1
S2: My child tortured a lot
S2*: The mirror child tortured a lot
I must prefer S1 to S2, but (unless impartialism is true) I'm not obviously required to prefer S1* to S2*. I could prefer that a duplicate on the other side of the universe get tortured a lot rather than my own child right here getting tortured a bit.
We can thus produce counterexamples that rely on "self-locating" or de se desires, without any appeal to haecceities. And the point generalizes: which qualitative outcomes I prefer may depend on my 'location', including my location in modal space, i.e. which world is actual. One needn't believe in haecceities to prefer W_Jack to W_Jill; one simply needs to have gotten attached to the particular cluster of qualities that make up the actual person Jack. Appeal to PSQ to rule out these sorts of preferences would be entirely question-begging.