Demographic Influences on U.S. Election Outcomes Across Levels

A Machine Learning Approach

Abstract

Demographic factors such as race and education are strongly correlated with party affiliation in the United States. Using machine learning techniques, this paper uncovers how the relationship between those demographic factors and voting behavior differs across different election jurisdictions by analyzing those relationships across senatorial, gubernatorial, and presidential elections. Previous studies surveyed do not incorporate modern machine-learning techniques; incorporating this into the model allows for a numerical scoring of each demographic in its importance of determining election results. The result was a separate importance score per demographic factor for each election jurisdiction analyzed. Among other results, the analysis in this paper showed that racial demographics tend to have the strongest predictive power among all demographic features studied, but have greater importance in presidential elections than in gubernatorial or senatorial elections. This study found that despite large shifts in party platforms and candidate profiles, demographic patterns still hold strong predictive power in how voters will choose candidates in elections across different levels of government.