This research investigates the factors influencing viewer perceptions of debate winners during the 2021 German chancellor discussion. Utilizing survey and real-time response data from 4613 participants, the study employs machine learning (random forest and decision tree models) to identify determinants of opinion changes. The analysis reveals the impact of pre-debate preferences, candidate images, and specific speech moments on post-debate winner perception shifts. Pre-debate chancellor preference and candidate images significantly influence perception change, while party identification plays a less crucial role. The study also pinpoints specific speech moments that shifted viewer perceptions, providing a novel approach to evaluating political debate performance.

Felix Ettensperger, Thomas Waldvogel, Uwe Wagschal, Samuel Weishaupt