Multidimensional polarization dynamics in US election data in the long term (2012–2020) and in the 2020 election cycle

We use a network-based method to explore bifurcation in the multidimensional opinion-based political identity structure from 2012 to 2020 in American National Election Studies data. We define polarization as ideological clustering which occurs when attitudes are linked or aligned across group-relevant dimensions. We identify relevant dimensions with a theory-driven approach and confirm them with the data-driven Boruta method, validating the importance of these items for self-reported political identity in these samples. To account for data sets having different sizes, we bootstrapped to obtain comparable samples. For each, a bipartite projection generates a network where edges represent similarity in responses between dyads. The data provide us with preidentified groups (Republicans and Democrats). We use them as our network communities and to calculate an edge-based polarization. Results show bifurcation progressively increasing, with a striking increase from 2016 to 2020. We visualize these identity-related shifts in opinion structure over time and discuss how polarization results from both between- and within-group dynamics. We apply a similar method to a smaller data set (N = 294) to explore short-term fluctuations before and after the 2020 election. Results suggest that between-group polarization is more evident after than before the election, because in-group opinion dynamics result in a more synchronized opinion-space for Republicans.