Isaac Mehlhaff (University of North Carolina, Chapel Hill)
Abstract: Political polarization has become a key concern in many important topics within comparative politics, yet past research has reached little consensus as to its substantive causes and effects. Much of this disagreement, I argue, stems from the use of diverse measurement strategies that do not reliably capture the two key dynamics of polarization: distance and concentration. I use unsupervised machine learning methods to derive the cluster-polarization coefficient (CPC), a novel measure of multimodal data structuration that scales to high-dimensional analysis and accepts a wide variety of data structures. I use Monte Carlo simulations to show that the CPC predicts polarization with substantially greater accuracy than current measures and I offer a substantive application by replicating and extending the well-known work of DiMaggio, Evans, and Bryson (1996) using data from the American National Election Studies.