Measuring Populism in Context: Introducing a Supervised Approach

• In Event: Poster Session II: Comparative Politics

Abstract

Populism is a worldwide phenomenon. However, much of the academic literature has focused on just a few regions, notably Western Europe and Latin America. One major obstacle in studying populism from a global perspective is coming up with a measure of populism that is comparable across countries and over time. In this paper, I introduce a computer-assisted supervised method that `learns' the features or characteristics of populist discourse from a small number of human-coded training texts, and is able to classify new documents based on the learnt features in the training set. Because populist rhetoric and ideas are embedded in a particular context and cannot be understood in terms of individual words, the `bag of words’ assumption underlying the typical automated text analysis methods used in political science is problematic. To address this issue, I incorporate word-embedding models that vectorize documents based on their `context’ rather than on individual words. This method is highly generalizable and is not constrained by language. I have already successfully applied the method to measure populism in China using tens of thousands of articles from the state-controlled media. In this paper, I will extend my measure of populism to Europe and America, and compare it to alternative measures of populism in the existing literature.

Author

• Yaoyao Dai, Pennsylvania State University