Uncovering partisanship at the national and state level by identifying unique voter blocs in congress

Singular value decomposition and non-negative matrix factorization are machine learning methods that can uncover hidden patterns in large datasets. We apply these methods to detect partisanship in voting behavior in national and state (Illinois) legislatures. We find that there are two clear voting blocs at both the national and state level that align with party affiliation. Party lines clearly define voting behavior at the national level, but do not predict voting behavior as well at the state level, suggesting a more fluid party affiliation at the state than national level.

Note: this work was done in partnership with Evelyn Siu (esiu@uchicago.edu) as a final project for the Mathematical Foundations of Machine Learning course taught by Prof. Rebecca Willett at UChicago in Fall 2022.