2025/09/17 by Hendrik Erz · 1 voice
Social Sciences · #Law, Rights, and Freedoms #Law in Society and Culture #Judicial and Constitutional Studies
paper · pdf · doi:10.3384/9789181182828
Political institutions and the policies they create are central to modern societies, and have a strong influence on many aspects of everyday life. Lawmaking more specifically describes the process of creating and adapting such policies by drafting legislation. This process takes place in a country’s legislature, in which elected officials –representatives – come together to debate policies. This thesis explores the process of lawmaking in U.S. Congress. Using state-of-the-art text analysis methods, it ex-amines the transcripts of the floor speeches of representatives in order to understand this process from a substantive perspective. By extracting relevant information from the available textual data, it explores the discursive dynamics in U.S. Congress, how the topics which representatives talk about can contribute to polarization, and how the representatives’ speech patterns predict their voting behavior. This thesis is motivated by a desire to understand not just what policy changes are made by the U.S. legislature, but also why and through which processes. By combining political speech with vote results data, it aims to understand which mechanisms drive the policy outcomes that we observe. It aims to not just deepen our understanding of how the representatives in the two chambers of Congress have voted, but why. What influence do their social origins play, their education and occupational history? What role do the institutions in Congress – the committees, the whips, the parties – play in these decisions? How can we explain the large shift in economic policy that took place in the second half of the 20th century? To answer these questions and contribute to a deeper understanding of the lawmaking processes in U.S. Congress, this thesis operates under the framework of analytical sociology. It designs testable hypotheses about how lawmaking happens at various points in time. In total, the thesis explores more than 130 years of Congressional lawmaking and examines over 17 million speeches by more than 7,000 representatives, and over 100,000 roll-call votes. To enable such a large-scale analysis of text and vote data, the thesis draws on the toolkit of computational social science to ex-tract signal from the large amounts of data this thesis tackles. More specifically, it draws on recent advances in text analysis methods, including Large Language Models (LLMs) to extract relevant information from text that can give us insight into the preferences and motivations of the representatives. Across three essays, this dissertation analyses the reasoning of representatives and how policy changes over time. The first essay examines the discursive dynamics in U.S. Congress around a time when the U.S. economic system underwent exogenous shocks. It determines how the shared experiences of representatives and the institutional setting in which they operate influenced their policy preferences. The second essay explores polarization in Congress based on what topics representatives talk about, highlighting that party polarization is only one aspect of the diverging trends we observe in Congress. The third essay proceeds to explore the relevancy of the contents of floor speeches for the representatives’ voting behavior. The essay predicts the likelihood of representatives breaking ranks with their party in roll-call votes based on their speeches in advance of the vote. It uses this information to derive a measure of party pressure, highlighting how the institutional context of Congress restricts lawmakers in their choices, but unevenly so over time. This thesis contributes to, and draws upon, a large amount of literature in sociology, political science, and computational linguistics. It uses modelling frameworks developed by political science to analyze the lawmaking process, develops hypotheses based on an analytical sociological perspective on social change, and adapts three different computational text analysis methods to align with the theoretical and methodological assumptions and preconditions of its work.