Measuring Agendas, and Positions on Agendas

Variation in agendas and positions on agendas are independently important. The former explains the kinds of issues people think are important, and the latter potentially explains people's positions on the issues. In this paper, we estimate variation on both these dimensions. To do this, we first estimate a supervised topic model using the bill labels from the Policy Agendas Project. Then within each of the topics, we estimate a model of slant using congressional speech as training data. Using this method, we scale 255 news programs across 50 television channels using original, and closed captions transcripts from these shows. We find large systematic variation in broad agendas of news media, and positions on those agendas.