According to George Stigler and Gary Becker (in a 1977 article) – not to mention Donald Green and Ian Shapiro (who use this to bash rational choice) – it is scientifically wrong to explain empirical anomalies that do not fit the expectations of a particular model by positing changing preferences or by using so-called ad hoc add-ons to typical rational actor models. Dennis Mueller has a nice response – though not one that will surprise most professional social scientists – in his very interesting (but not so new ) survey of public choice theory, Public Choice III (certainly not the most exciting title, but it will probably be an intellectually exciting book for those who did not cut their teeth on this stuff in grad school. Indeed, I think many well-educated non-social scientists might enjoy it as well):
There is nothing in the rational actor methodology that demands that we assume that there is only one argument in an actor’s objective function, and that the analyst is constrained in her choice of what this one argument should be by the choices made by previous analysts. This point is particularly important to keep in mind when considering the application of rational actor modeling to politics (pg. 661).