Exploring the predictors of the populist vote using random forests

Abstract

In this paper, we use random forests to determine what individual attitudes and opinions are the best predictors of the vote for a populist party. We used data from the European Value Survey, Wave 7, and after coding European parties as populist or not, we carried out a preliminary analysis on two peculiar nations, France and Poland, highlighting the basic steps of a random forests application. The analysis reveals that populist voters have different attitudes. In Poland, the populist vote is mostly predicted by the adherence to religious and traditional values. In France, vote is mostly predicted by strong discontent on the actual practice of democracy. However, we show that predictions can be biased when imbalanced data are used, that is, data in which the minority class contains few observations. We discuss how to obtain a balanced dataset and we show that in this way the predictive power of the random forests is improved. Next, we extend the analysis of the populist vote to all the European nations available in the survey, to determine what are the most important predictors both at supranational and national level. The use of the random forest allows determining what are the most common global predictors, and the role of some local predictors as well.