New publication: State-of-the-art network measure on polarization by SODAS PhD
SODAS PhD student Marilena Hohmann is first author on the newly published paper “Quantifying Ideological Polarization on a Network Using Generalized Euclidean Distance”. The paper is the product of an interdisciplinary collaboration with mathematician Karel Devriendt and computer scientist Michele Coscia. In this paper, the group develops a state-of-the-art network measure of polarization. Tested against already-known polarization measures, the measure developed in this paper more adequately captures three theoretical components of polarization: the opinion component (how extreme are people’s opinions), the structural component (to what extent are people embedded in echo chambers), and the opinion-structural interplay (how do these echo chambers organize in the network).
How can we accurately quantify the level of ideological polarization of a social system? This question taps into the debate of how to measure polarization in offline and online contexts - if such polarization even exists, of course. The paper seeks to synthesize the two main approaches within polarization theory: a traditional approach of opinion divergence embedded within political science, and a newer approach from computational studies, focused on network structures.
One of the main accomplishments of the paper is to construct a new measure based on a single numerical value that describes the level of ideological polarization of a social network. This polarization measure has promising potential for comparisons of different social networks. The measure captures all relevant theoretical dimensions derived from the literature, where existing measures would miss at least one of them. It is based on the Generalized Euclidean distance between vectors, which measures whether two different characteristics of nodes (e.g. liberalism and conservatism) tend to have the same pattern of spread in a network. In other words, it can measure whether people with liberal and conservative opinions tend to be close to each other, or far away from each other, in the same network.
To test and validate the polarization measure, the authors apply it on a wide range of six different Twitter discussion-networks including Obamacare, gun control, abortion, and the US vice-presidential debate, 2nd presidential debate, and election night in 2020. The new polarization measure shows that the election debate networks were overall more polarized than the other Twitter debates. Based on US House Representative data, the authors also show that polarization has increased since the 1950s, which is in line with prior results
To read the full paper, follow this link.