Ranking Swing Voters in Congressional Elections

TL;DR

This paper introduces a model to identify and rank swing voters in congressional elections by predicting individual voting likelihood and preferences, then refining these predictions with aggregate data to enable targeted campaigning.

Contribution

The paper presents a novel approach combining individual and aggregate data to accurately rank swing voters for targeted political campaigns.

Findings

Effective identification of likely voters and swing voters

Enhanced prediction accuracy through aggregate data incorporation

Potential for improved campaign targeting and vote maximization

Abstract

We present a model for quantitatively identifying swing voters in congressional elections. This is achieved by predicting an individual voter's likelihood to vote and an individual voter's likelihood to vote for a given party, if he votes. We make a rough prediction of these values. We then update these predictions by incorporating information on a municipality wide basis via aggregate data to enhance our estimate under the assumption that nearby voters have similar behavior, which could be due to social interaction or common external factors. Finally, we use a ranking scheme on these predictions to identify two key types of voter: 1) Voters who are likely to vote that we can convince to vote for a given party; and, 2) Voters who are likely to vote for a given party, if they vote, that we can convince to actually turn out to vote. Once these voters have been identified, a political campaign can use this information to micro-target voters and win more votes.