ABSTRACT
Justin H. Gross Department of Political Science, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA
Daniel Manrique-Vallier Department of Statistics, Indiana University, Bloomington, IN 47408, USA
Political scientists have long observed that members of the public do not tend to exhibit highly constrained patterns of political beliefs and values in the way that partisan elites often do. In answering survey questions designed to measure latent ideology, they may act as if drawing responses randomly from different perspectives. In the American context, survey respondents frequently defy easy categorization as prototypical liberal or conservative, and yet their response patterns reflect structure that may be characterized by reference to such ideal types. We propose a mixed membership approach to survey-based measurement of ideology. Modeling survey respondents as partial members of a small number of ideological classes allows us to interpret the “mixed signals” they seem to send as a natural consequence of their competing inclinations. We illustrate our approach by reanalyzing data from a classic study of core beliefs and values (Feldman, 1988) and find that the most dramatic difference between prototypical members of the two main ideologies identified is not their vision of what society should be but rather their belief in what American society actually is.
