Machine Learning Approaches to Understanding Legislative Economic Discourse

This dissertation provides an investigation into legislative economic discourse across multiple countries and institutional settings, as well as communication channels. It relies on several machine-learning techniques to analyze legislative texts and social media posts. The analyses included in the various chapters offer insights into how economic rhetoric works as a vehicle for ideological expression, constituency engagement, and policy framing. These results have implications for methodological innovation, the dynamics of political communication, and the broader understanding of legislative behavior. This research project includes three papers and a research note. The first paper (Chapter 2) conducts a systematic assessment of the capabilities of 12 machine learning models and model variations in detecting economic ideology. As an evaluation benchmark, I use manifesto data spanning six elections in the United Kingdom and pre-annotated by expert and crowd coders. The analysis assesses the performance of several generative, fine-tuned, and zero-shot models at the granular and aggregate levels. The results show that generative models such as GPT-4o and Gemini 1.5 Flash consistently outperform other models against all benchmarks. However, they pose issues of accessibility and resource availability. Fine-tuning yielded competitive performance and offers a reliable alternative through domain-specific optimization. But its dependency on training data severely limits scalability. Zero-shot models consistently face difficulties with identifying signals of economic ideology, often resulting in negative associations with human coding. Using general knowledge for the domain-specific task of ideology scaling proved to be unreliable. Other key findings include considerable within-party variation, fine-tuning benefiting from larger training data, and zero-shot’s sensitivity to prompt content. The assessments include the strengths and limitations of each model and derive best-practices for automated analyses of political content. The second paper (Chapter 3) investigates whether and why legislators express different economic ideologies on social media compared to floor speeches. It focuses on legislators in Ireland, the United States, and the United Kingdom from 2011 to 2019. The analysis relies on 16+ million texts from Twitter posts and speeches by over 2,400 lawmakers. I detect economic content using a fine-tuned transformer model and classify it based on labels of economic ideology using a generative language model. The findings reveal that electoral systems resulting in high party discipline enforce more cross-venue ideological coherence among legislators. In addition, lawmakers’ behavior changes over their tenure and adapts to different contexts. Senior legislators exhibit more consistent economic messaging. Campaign periods reduce cross-platform inconsistency in the United States but amplify it in Ireland and the UK. Reactions to national unemployment show similar cross-country differences. Higher unemployment rates promote more economic messaging coherence in the United Kingdom but lead to increased divergence in the United States. Being in the opposition was consistently associated with lower degrees of cross-venue inconsistency. However, several robustness tests suggest a heavy dependence of this effect on partisan affiliation. These findings highlight the strategic nature of legislators’ economic communication, its adaptability to different contexts, and development over time. The third paper (Chapter 4) examines how members of the House of Commons adapt their economic communication to changing economic conditions in their constituencies. Using a corpus of 970,072 speeches delivered between 2005 and 2019, I detect economic and local content using transformer models fine-tuned on hand-coded data and measure sentiment using a zero-shot language model. The computational approach to detecting constituency focus improves upon prior studies based on mentions of geographic localities alone. Relying on constituency-level jobless claims as a proxy for economic conditions, the analysis evaluates the impact of institutional and individual-level predictors. Higher unemployment is associated with increased locally-oriented speeches, more focus on economic policy within these speeches, and a negative shift in sentiment. Electoral security and seniority discourage responsiveness to constituents’ economic conditions. Government lawmakers are more likely to react to changes in jobless claims. The research note (Chapter 6) examines how US House Members adapt their economic sentiment in social media posts and congressional speeches in response to financial, monetary, and macroeconomic indicators. The analysis focuses on standardized measures of stock performance, treasury yields, oil prices, industrial production, and consumer confidence while controlling for demographic and political attributes. It also includes analyses of potential temporally delayed responses. The findings reveal that stock performance and treasury yields are the most influential drivers of sentiment with cross-platform differences in duration and intensity. Tweets tend to show quicker and more sustained reactions. Speeches show a shorter and sometimes weaker responsiveness. Oil prices, industrial production, and consumer confidence also shape sentiment but with varied timing and strength. These findings highlight legislators’ strategic use of different communication channels for immediate versus measured responses. They underscore how economic conditions inform political discourse and suggest further research on platform-specific messaging and policy impacts.