We’ve had various versions of the case for and against the use of (micro)economic rational actor models in the social sciences lately, so I thought I would weigh in with my version of the case against. It has three main elements
First, most rational actor models assume that “rationality” can be represented as “maximization of self-interest”. This assumption is either false or vacuous. Those committed to egoistic rationality tend, when challenged, to oscillate between falsehood and vacuity, in much the manner of the function sin (1/x) as x approaches zero.
A very good example arises with voting. In this 1987 paper pointing out the empirical weakness of rational actor models, I observed out that, while egoistic rationalists will not vote (or rather will vote only in very small numbers), quite a limited amount of altruism suffices to making voting rational. Andrew Gelman made the same point more recently.
Second, the fact that egoistic rationality assumptions work well in a lot of microeconomic applications proves little or nothing about their viability in other contexts; these include macroeconomics as well as sociology, political science and other targets of microeconomic imperialism. Abstractions that work well in one context don’t work well in another. In particular, deviations from given assumptions that roughly cancel out locally may nonetheless be significant in the aggregate. I conjecture that something like this underlies the notion of emergent phenomena, beloved of critics of reductionism in general.
Finally, game theory is much more problematic than is commonly realised. To derive a Nash equilibrium, it is necessary to define the strategy space. In real games this is not a problem. In social, economic and political operations, however, it requires that the participants have shared understandings of the problem, accessible to the modeller. In practice this is hardly ever true, and game theoretic analyses typically proceed with an essentially arbitrary assignment of strategies to players.