At the Centre for AI in Government (CAIG), Christine is engaged in a diverse range of projects. For example, her research on computational methods in constituency service explores both the dynamics of polarization in direct political communication and biases in constituency responses. Drawing on a novel dataset of 250,000 citizen questions and nearly 200,000 political replies from Germany, she examines polarization trends over two decades. She also investigates bias in responses to constituency queries facilitated by Large Language Models (LLMs). Although automation reduces human disparities in response rates and quality across demographic groups, her experiments demonstrate that LLM-generated responses still exhibit subtler forms of bias, albeit at lower levels than human responses.
In turn, together with her colleagues at CAIG, she is contributing to the development of Large Language Model (LLM)-based digital twins to analyse governance, focusing on EU policy decision-making. Using LLM-based Multi-Agent Systems (MAS), the project aims to simulate structured negotiation processes. The European Parliament serves as the initial case study, leveraging annotated EU negotiation transcripts for high-fidelity simulations. This adaptable framework will lay groundwork for scalable governance analysis applications, bridging computational innovation with policymaking.
Finally, her PhD research examined coalition governance by analysing the dynamic behaviour of political parties in coalition cabinets, with a focus on "coalition differentiation"—when parties prioritize distinct goals over coalition unity. Through supervised text classification of parliamentary speeches, she introduced a novel, monthly measure of differentiation across 10 European countries, enabling unparalleled comparative analysis. Her findings reveal that powerful parties often differentiate to influence policy negotiations and threaten coalition stability. The contributions of her thesis are currently being refined for publication.