The issue context of modern American politics: semiparametric identification of latent factors from Discrete data

Abstract

A new methodology that estimates attitudes semiparametrically and estimates actions nonparametrically, as a function of the resulting attitudinal measures, is used to examine the behavioral eects of 'cultural' and 'economic' preferences in the Presidential elections of 1984 and 1992. The results suggest a shift toward 'cultural politics', achieved …rst among the highly educated but spreading throughout society by the later election. One consequence is that both parties are now consistent in their policy alignments—the Democrats being liberal on both scales, the Republicans conservative. Despite this aggregate consistency, dierent social groups are attached to the parties in dierent ways, thereby heightening the potential for intraparty con‡ict while sharpening the problem of fashioning a platform that is broadly attractive. These problems, …nally, express themselves very dierently within the Democratic and the Republican parties. 1. The Issue Context of Modern American Politics Social science is on its most …rm ground when it deals with concrete behavior - the patterning to that behavior, the measurement of this patterning, the comparison of those measures. Nevertheless, many of the central notions of social life are not directly observable, in principle. Students are said to have aptitudes. Consumers are said to have tastes. Citizens are said to have policy preferences. And education, consumption, or, in this case, voting can hardly be understood in the absence of these attributes. This produces a great and familiar, two-part psychometric problem: how to infer an actual distribution of unobserved attributes from responses to a series of questions or items, and how to infer the relationship between that distribution and individual behavior or aggregate outcomes. In this paper, we attempt a fresh approach to deriving an attitudinal distribution, where the attitudes involve preferences for public policy. We proceed, in essence, by specifying individual responses as a probabilistic function of a (small) number of attitudes. And we then use the result to construct an issue context for modern American politics, in which attitudinal distributions shape voting behavior. This allows us to a¢rm some conventional