Biased reporting by the German media?

The dynamics of online news and political outcomes have been of high interest in various research fields in recent years. This paper provides a new method to estimate media bias using a structural topic model and cosine similarity to test slanting toward different political actors. For the empirical analysis, the content of German online newspapers and press releases of German parties during the election campaign before the federal election in 2017 is analyzed. Following the assumption that a) potential media bias is demand-driven and b) election results can be used as a proxy for reader beliefs, the results show that news articles of most newspapers slant towards AfD topics. Furthermore, we find evidence for the hypothesis that the election day results in changes in news coverage since newspapers can observe the true beliefs of readers.