A Entropy-Based Measure of Mass Political Polarization and Its Implications

• In Event: Political Methodology Virtual Poster Session

Abstract

Political polarization has become an extremely important topic in political science, yet existing studies are significantly limited by the measurement issues of polarization. Studies have reached many consensuses on party and elites polarization, but to what degree the mass public is polarized is still an ongoing scholarly debate. With few exceptions, most of the existing studies of mass polarization are conducted in a single-country context (mostly the U.S.) and systematic knowledge about the variations and sources of polarization across countries is still very limited. Many of these gaps and limitations in the existing studies can come down to the measurement issue of mass polarization. Existing studies rely on conventional statistics such as variance, bimodality coefficients to measure mass polarization and tend to reconceptualize polarization based on whatever measure is available rather than the other way around. While these measures may capture the characteristics of the polarization under certain circumstances, they are also largely limited by the assumptions about the data type, distribution, etc., or simply the fact that they are not originally designed to measure polarization.

In this paper, we propose a nonparametric, entropy-based measure of mass polarization. The measure takes into account the specific features of ordinal data from the public opinion survey. We follow the conventional definition of polarization in the literature but design the measure precisely to capture the dynamics of mass polarization. Contrary to the previous research that relies on the unnecessary distinction between dispersion and bimodality and unrealistic assumptions of continuous data and specific modality, our measure emphasizes the ordering and distribution of the ordinal data, which are more central to depict polarization. The proposed measure is free of central tendency, number of choices, as well as assumptions about specific distribution and intervals between categories, etc. We use large-scale simulated data to demonstrate the validity of this new measure and benchmark its performance with multiple conventional measures of mass polarization. We then employ this new measure to three empirical examples: (1) ideological and issue polarization in the U.S.; (2) mass polarization of European integration in European countries; and (3) economic development and mass polarization in Latin America.

Authors

• Le Bao, American University

• Jeff Gill, Washington University