Ananya Balaji
Research Mentor: Joshua Ashkinaze
Mentor Department: University of Michigan, Information
Author(s): Ananya Balaji, Joshua Ashkinaze
Session: Session 1 (9:00 AM – 9:50 AM)
Presentation Type: Poster 44
Abstract
Ballot measures are a critical part of direct democracy. However, voters are highly uninformed about these measures, performing worse than random chance when answering binary questions of basic ballot information. Moreover, many ballot measures pass on thin margins. Consequently, our research group is developing scalable AI tools that enable voter education on ballot measures to improve overall outcomes and strengthen democracy. But in order to do this, the tools need to be robustly trained, demanding a large database of ballots. My work involved putting together all the 2026 ballot measures and understanding what the 2026 ballots look like as a whole via topical modelling. Using a custom web scraper built on Ballotpedia data, I collected metadata and vote percentage information across active and upcoming ballot measures, validating the pipeline on both historical and future elections. I then conducted exploratory analysis of the 2026 dataset — examining geographic distribution across states and applying topic modeling using scikit-learn to identify the major policy domains represented. Ongoing work involves collecting raw measure text for a population-weighted sample of 30 ballot measures to enable deeper language-level analysis. Future work entails completing the topical modeling and any other cleaning for the database such that it’s ready to train the AI tools that were built.
