Traditionally, philosophers have wanted to analyze dispositions in terms of counterfactuals. We might say, for example: a substance is soluble iff it would dissolve were it to be placed in water. But it was soon suggested that such analyses commit the "conditional fallacy". A cheeky interpretation of Kvanvig's complaint might have us characterize this as the fallacy of utilizing subjunctive conditionals in an analysis. But I take it the idea behind Shope's original diagnosis was that satisfaction of the conditional's antecedent conditions might have a confounding influence on the case. This is most obvious in simple ideal agent theories, when the agent's ideality influences their decision. (My idealized self would believe "I am ideal", but that doesn't mean that I ought to believe it!) Interested readers can find a thorough discussion of the Conditional Fallacy in this paper [PDF] by Bonevac et al.
Many standard counterexamples to the simple counterfactual analysis of dispositions involve situations where the antecedent condition would add or remove the disposition. Suppose a fragile glass has a divine protector who would step in and temporarily strengthen the glass were it to be struck. This example shows us that one cannot analyze fragility as the counterfactual would break if struck. David Lewis, in 'Finkish Dispositions' elaborates the counterfactual analysis to accommodate such examples, by appealing to the "causal basis" of a disposition, and building the persistence of this property into the conditional's antecedent. (This pre-empts 'finkish' counterexamples like those above, where the causal basis of the disposition gets temporarily removed.)
Still, this seems insufficient to me. There remain an interesting class of counterexamples which don't involve manipulating the dispositional property itself. The object retains its disposition throughout, but the counterfactuals fail for reasons of modal deviancy. Let me elaborate.
I'm interested in counterfactual-based analyses that require an assumption of modal 'normalcy'. This came up in my response to Timothy Williamson on thought experiments. Counterfactuals depend on our position in modal space, and it's possible for really quirky and weird counterfactuals to be true, if (perhaps unbeknownst to us) our modal position is deviant enough. Intuitively, 'modal deviancy' is a matter of counterfactual coincidence. It is when surprising, random, or unlikely events take place in our neighbouring possible worlds. For example, if through some quirk of modal space, we happened to be modally situated such that the nearest possible world where I click my fingers is one where I am subsequently struck by lightning, then the counterfactual "I would be struck by lightning were I to click my fingers" would be true.
Often we want to disregard such oddities, and I'll argue below that this is so for dispositions in particular. But the concern I have here is more general (and, I think, tolerably distinct from the "conditional fallacy"), so I want to give it a name. Let us say that one commits the counterfactual fallacy by offering a counterfactual-based analysis when this would only yield plausible results under conditions of modal 'normalcy'. (That is, the fallacy of assuming that modal normalcy obtains, or of neglecting the possibility of modal deviancy when offering an analysis involving counterfactuals.)
Now, I want to suggest that counterfactual-based analyses of dispositions (even Lewis' more complicated version) commit the counterfactual fallacy. For something to have a disposition is instead for it to support the corresponding counterfactuals under normal modal conditions. Suppose that salt normally dissolves in water, but through some modal quirk (let's blame quantum mechanics!) it just so happens that if I were to now place this salt in a glass of water, it would - by sheer incredible quantum coincidence - pass right through the glass and end up sitting, undissolved, on the table. That wouldn't stop the salt from being soluble. Not even temporarily, as in the case of Finkish dispositions. We had perfectly soluble salt sitting in water and refusing to dissolve for reasons of sheer coincidence or modal deviancy.
My point generalizes. We see that the truth of counterfactual conditionals may be influenced by arbitrary considerations and random chance. Insofar as the things we want to analyze are not beset by such quirks, we should be wary of employing counterfactual conditionals in their analysis.