Assessing the Extent of Populist Attitudes among Voters Using Machine Learning

Even though research into populist attitudes among voters is still in its infancy, there is evidence suggesting that views exist among voters which reflect the core concerns of ideational populism, and validly tap into a single concept even across countries. Yet, it is precisely because of the novelty of this research strand that we still lack comparative evidence regarding the extent of populist attitudes and scope of their effects on political behavior especially over time. Have voters always been populist and did political actors merely politicize and activate such (latent) attitudes? Or does the level of populism hinge mostly on the presence of populist parties within a political system? In this paper, we first seek to predict populist attitudes of respondents of the German General Social Survey (ALLBUS/GGSS), before assessing the validity of our prediction and the effect of our predicted attitudes on vote choice in a second analytical step. To this end, we draw on the 2018 round of GGSS for which populist attitudes are available. We single out predictors that are theoretically related to populism yet distinguishable from host ideologies, to separate the two. Drawing on Machine Learning, we then assess the explanatory power of these (socio-economic and political) indicators to estimate populism within the 2018 round, before using these models to calculate a populism score for individuals going back in time. Finally, we assess the validity of our prediction and the effect of predicted populism scores on political behavior and voting decisions.