Dynamic exploratory graph analysis of emotions in politics

Abstract

This study explores the dynamics of emotions in political leaders’ communication using network psychometric methods applied to facial expression recognition (FER) data extracted from YouTube videos. The analysis covers 220 videos of global political leaders and employs zero-shot machine learning via the transforEmotion R package. It focuses on six emotions (happiness, excitement, hope, anger, fear, and sadness) and a neutral expression. Dynamic Exploratory Graph Analysis reveals a two-dimensional network structure for FER scores and their rate of change, showing distinct patterns between positive and negative emotions. The first derivative model indicates a negative correlation between anger and most other emotions, suggesting a more autonomous expression of anger. Significant differences in network structure emerge between leaders with varying degrees of populist rhetoric. More populist leaders exhibit less connected and more autonomous expression of anger, while happiness becomes more contingent on other emotions. In the discussion, we consider the universality of the network structure, the autonomy of anger expression, and the implications of emotional connectivity within the estimated models. The results offer valuable insights for future computational studies of affective political communication, particularly in the context of rising global populism.