An Odyssey Through Political Space and Time: Tracing 130 Years of Swiss Legislative Work Through Computational Approaches

Abstract

This thesis marks a significant advancement in the quantitative examination of the Swiss Parliament. By examining over 130 years of legislative work of the Swiss Federal Assembly, this work stands as the most comprehensive study conducted to date. Central to this thesis is the development of the DemocraSci relational graph database, a comprehensive database of over a century's Swiss parliamentary activities, ranging from 1891 to the present. A core focus and contribution of this thesis is demonstrating how text embeddings can be leveraged to extract the meaning of text for substantive political science research questions. The application of text embeddings from large language models represents a novel and crucial advancement in the field of legislative research. We introduce a novel, comprehensive approach to quantitatively capture the multifaceted aspects of member of parliament's (MPs) legislative activities, including the content of their speeches and legislative texts, MP's embedding positions. Our approach provides nuanced understanding of ideological stances. Specifically, we leverage high-dimensional embedding spaces created from text data to establish a political space that is comparable over time, capturing the evolution of language and ideological expression. Furthermore, we leverage our embedding approach to analyze over 130 years of polarization in the Swiss political landscape. For this, we employ the content of legislative bills to trace party divergence, i.e. polarization, within the high-dimensional embedding space. Our approach reveals a strong increase in polarization over recent decades. This confirms that even when studying polarization over more than 130 years, we are currently experiencing unprecedented high levels of polarization in Switzerland. A detailed examination across 21 political topics shows different patterns of polarization over time, with notable increases in areas like International Affairs from the 1960s and Transportation in the 1980s. This thesis presents the first study to utilize bill texts as a primary source for measuring polarization, highlighting the utility of legislative content in identifying ideological divergence from the earliest stages of the legislative process. By doing so, it broadens our understanding of polarization, demonstrating its evolution across different political themes over time. After having identified polarization at the beginning of the legislative process by studying the content of legislative bills, we consider the deliberative process of the parliament. For this, a generative agent-base model is employed. This thesis probes the intricate interdependencies between MPs' topic preferences and polarization. It illustrates the influence of the chamber's size and the diversity of political topics on polarization and specialization. Additionally, this thesis develops a text-based framework to analyze MPs' engagement across political topics, addressing the challenge of accurately capturing the MPs' diverse interests and the complexity of interconnected topics. Utilizing a high-dimensional embedding space, it maps Swiss political topics over time, allowing for a detailed examination of MPs' topical breadth. This approach highlights MPs' increasing focus and specialization in their legislative work, also aligning with the observed professionalization trend. By addressing the non-orthogonality of topics through bill text embeddings, this method offers a nuanced understanding of topic similarities, enabling a more precise measurement of MPs' issue engagement. This approach represents a significant contribution in evaluating the depth and diversity of legislative efforts.