Voting the District or Talking the District

• In Event: Legislative Productivity in the U.S.

Abstract

In this paper we examine how members of the U.S. House of Representatives during the 115th (2017-18) and 116th (2019-20) Congresses represent the preferences of their district – by how they vote or how they talk. This question of constituency representation is central to the study of the U.S. Congress and has generated a literature that is both extensive and influential in the study of American politics. Here we leverage new developments in text-as-data methodology to examine this classic question of whether legislators more closely reflect the preferences of their district through what they say or the votes they cast. In doing so, we contribute to the growing body of political science research that uses computational linguistics to gain new insight into political behavior and dynamics (e.g., Grimmer 2013; Lauderdale and Herzog 2016; Monroe et al. 2009; Wilkerson and Casas 2017).

We do this by comparing the position of each House district (calculated using election-based measures) to three measures of legislative behavior. First, we employ the widely-used vote-based DW-NOMINATE measures of legislators’ ideal points to determine how legislators are positioned based on the votes they cast. Second, drawing on recent advancements in text-as-data analyses in the political context (e.g., Vafa, Naidu, and Blei 2020; Hoyle, Goel, and Resnik 2020), we develop a text-based ideal point model (TBIP), which is an unsupervised probabilistic topic model that uses the Congressional Record from the 115th Congress (2017-18) and the 116th Congress (2019-2020) to estimate the political positions of all House members based on the language they use in the official setting of the House floor debates. We also employ this text-based ideal point model to estimate a separate set of political positions for all House members based on the text of their official Twitter accounts. The paper then first compares the three distinct measures of legislators’ positions, before examining district representation, and whether legislators vote like the district, talk like the district, or tweet like the district.

This paper shows that the choice of ideal point measure makes considerable difference for most legislators. These findings are similar to what Vafa et al. 2020 find for their sample of U.S. Senators (2015-16), and provide additional evidence that text-based ideal points make a unique and valuable contribution to the study of legislative politics. Moreover, our preliminary results reveal that different types of legislators represent their districts differently. We find that some House members are indeed quite consistent across the three ideal point measures, and these legislators also tend to take a position close to the preferences of their district, which indicates a high degree of representational fidelity. However, our data also show a set of legislators who are much bolder and extreme when looking at the text-based ideal points than they are in their voting behavior. Here, legislators are talking the talk that matches their district, but when it comes time to vote, they are much more moderate than one might have expected. There is also a third group of legislators who appear quite moderate by text-based ideal point measures, but have a more extreme (i.e., very liberal or very conservative) vote-based ideal point. This paper then uses regression analysis of all members of the House during the 115th and 116th Congresses to determine which types of members are more likely to represent their district through their votes, their talk, or their Tweets. We incorporate relevant characteristics of the legislator (e.g., race, gender, seniority, etc.) as well as characteristics of the district (e.g., district education levels, media markets, etc.).

Authors

• Kris Miler, University of Maryland, College Park

• SoRelle Wyckoff Gaynor, University of Maryland

• Philip Resnik, University of Maryland

• Pranav Goel, University of Maryland, College Park, USA

• Alexander Hoyle