Author Description

The past decade witnessed growing scholarly interest in mass polarization and concerns that divided electorates around the world are pulling political systems apart. However, comparative studies of mass polarization are limited by the lack of appropriate research methods and scholarly debate remains with regard to what causes mass polarization and whether countries are indeed ideologically more polarized. The dissertation consists of three chapters on the methods of studying polarization and their implications. Chapter 1 focuses on the measurement issue of mass polarization. We propose a new nonparametric, entropy-based measure of mass polarization that is theoretically and conceptually more appropriate for the intuition and structure of polarization. It exploits the specific structure of ordinal distributions and is able to capture both the ordering and distribution in public opinion data. The analytical comparisons, simulation, and empirical examples all demonstrate the reliability of the proposed measure and its capability of revealing the nuanced and complicated dynamics of polarization. Chapter 2 introduces a compositional data analysis approach to study macro-level political behavior. We show that log-ratio transformation, especially isometric log-ratio transformation, is more methodologically appropriate for compositional data, and further provides the leverage for exploring the richer underlying dynamics between different compositional parts. We use Monte Carlo simulations and a real-world example of democratic support to compare the proposed method with conventional political behavior approaches. Chapter 3 applies the preceding methods to the pressing question of how socio-economic inequality impacts mass polarization. We argue that, with increasing inequality, the growing lower class is the primary cause of mass polarization. We also consider party polarization and group-based redistribution as two possible moderation mechanisms that influence the relation