Tracing Policy Ideas in Coalition Agreements: A Computational Text Analysis Approach to Measuring Party Influence

Coalition agreements are central instruments of joint governance, yet systematic evidence on how individual parties shape their content remains limited. Which parties succeed in advancing their policy priorities across issue areas? In which domains do parties reach compromise, and where do they trade policy concessions? And which commitments cannot be attributed to the manifesto of any coalition partner?
This talk presents a computational approach for tracing policy ideas from party manifestos into coalition agreements using transformer models and large language models (LLMs). Drawing on the electoral manifestos of coalition parties and their joint agreements, it introduces and evaluates a paragraph-level matching framework that compares documents across multiple dimensions of textual and semantic similarity. The approach combines computational matching, LLM-based coding, and human validation to distinguish between commitments reflecting unilateral party influence, joint influence, and no identifiable manifesto origin. The resulting framework provides scalable and transparent indicators of party influence and creates new opportunities to study compromise, logrolling, and agenda formation in coalition governments.