In recent years, political discourse and election results appear to be more polarized in western countries but is this associated with increasing attitude polarization of their general public? To answer this question, many different polarization measures have been proposed in the literature but no systematic empirical comparison exists. In an exploratory analysis of 4155 attitude distributions on 11-point scales from the European Social Survey, we find that most polarization measures for single attitude distributions correlate strongly with the average attitude discrepancy between randomly selected pairs. We propose this as a catch-all measure for polarization because it can be decomposed into components related to different groups. By analyzing attitude distributions of the left–right political self-placements and several other topics, we find that distributions are typically not unimodal or bimodal, but show more so a structure with up to five modes. We exploit this structure by fitting a model with five latent groups of moderates, extremists, and centrists. Finally, we use the decomposition of polarization with respect to these groups to analyze polarization and its different aspects across topics, countries, and time establishing an overview and new perspectives on single attitude polarization in Europe.

In recent years, political discourse and election results appear to be morepolarized than in the years before. Empirical evidence for opinion polarization hasbeen found regarding specific topics but is there a general trend in society? Wecompare various polarization measures and find that in empirical data most ofthem correlate strongly with the average attitude discrepancy between randomlyselected pairs which we propose as a catch-all measure for polarization. In anexploratory data analysis of the European Social Survey, we analyze distributionsof individual responses on the left-right political self-placements and several otherattitudes. We find that distributions are typically not unimodal or bimodal, butshow more structure with up to five modes.We exploit this structure by fitting a new model to distributions of answers onan eleven-point attitude scale, and demonstrate that distributions can beapportioned into moderates, extremists, and centrists. We use the model todecompose general polarization into empirically meaningful components which weuse to analyze the complete data set across topics, across countries, and withrespect to time-trends establishing an overview and new perspectives onpolarization in Europe.