Introduction

Given theoretically predicted instability, why the empirically observed stability? There are several types of answers: that stability is an illusion because we are unable to detect the manipulation that occurs (Riker 1982); that stability is due to institutional devices (e.g. Shepsle and Weingast 1984); that such institutional devices are themselves pervasively unstable (Riker's rejoinder 1980a); that stability is due to similarity in preference rankings among the population, and to preferences for fair distribution; or is due to some other defect in the models. The counterempirical outcome of Arrow's theorem puts us on notice that one or another of its conditions must be misconceived.

I begin with similarity among individuals' preference rankings, a challenge to the realism of Arrow's condition of unrestricted domain. The theorem's impossibility result is a logical possibility but not an empirical probability, I shall argue. One kind of similarity in preference rankings is disastrous though: if majority-rule voters divide up a fixed good, and if each is motivated solely by self-interest and not at all by fairness, then we are guaranteed instability. Contrary to theoretical prediction, however, democratic legislators are typically universalistic rather than factional on distributional questions. This may be due to uncertainty about the future, or a direct concern for fairness, or independently motivated reciprocity, or public deliberation, or due to some combination of these devices. Empirical work shows that citizens vote judgments of general welfare rather than personal welfare.