Since the publication of The American Voter (Campbell et al. 1960), most analyses of electoral behavior have focused upon three factors: party identification, issues and candidate attributes. According to the Michigan model, voters have long-standing psychological ties to political parties, but particular elections can be decided by short-term forces such as issues and candidates. More recently, the hypothesis of retrospective voting (key 1966; Kramer 1971; Fiorina 1981) has found some empirical support in U.S. presidential elections, while the hypothesis of moderating elections (Alesina and Rosenthal 1989) rationalizes the empirical regularity of the midterm Congressional cycle. Both hypotheses pose a challenge to the primarily social psychological emphasis of The American Voter by proposing more rational economic and strategic considerations as the basis for electoal behavior.