Modeling Text in Time and Space

Abstract

What is the relationship between electoral campaign duration and electoral campaign content? While the 2015 United Kingdom general election was scheduled four years ahead of time, the 2017 and 2019 snap elections were called just weeks in advance, giving parties and candidates limited time to prepare their campaigns. We leverage this variation to examine how campaign duration affects what aspiring legislators talk about on social media. Using unsupervised machine learning, we analyze all tweets sent by candidates in the 31 days preceding the 2015, 2017, and 2019 United Kingdom general elections. Our findings suggest that longer campaign periods allow candidates to cover substantive issues related to party platforms, while shorter campaign periods are associated with mass mobilization, get out the vote statements, and responses to current events. Regardless of campaign duration, incumbents with safe seats are less likely to use Twitter, and challengers are more likely to talk about policy than incumbents.

Candidates running in the 2017 United Kingdom general election, by party, with the size of each electoral district weighted by number of tweets Candidates running in the 2017 United Kingdom general election, by party, with the size of each electoral district weighted by number of tweets