Have the US Partisans Become More Predictable? A Machine Learning Approach

• In Event: Political Sociology Section Roundtables
  In Refereed Roundtable (60 minute): Measuring Politics (Table 5)

Abstract

Have the US partisans become more predictable? This study investigates whether Democrats and Republicans have become more distinct social identities and if so, what drives their discernibility. A machine learning prediction model can inductively unveil how characteristics of Democrats and Republicans changed over time. Using the General Social Survey (1988-2024), we employ average 185 variables for the preliminary model, including attitudes, socioeconomic status, demographic group, lifestyle and taste. AUROC score of XGBoost model has increased from average 0.692 in 1988 to 0.915 in 2024, indicating that political identity has become dramatically predictable. Feature importance and additional analyses show that variables capturing specific attitudes make the most decisive contribution to this predictive power. The seemingly ever‑widening divide between the right and the left is in fact anchored in a relatively narrow set of contested issues.

Authors

• Inseong Jung, Korea University

• Eun Kyong Shin, Korea University