Dictionaries, Supervised Learning, and Media Coverage of Public Policy

Abstract

There are many different approaches to automated content analysis. This paper focuses on dictionaries and supervised learning; in addition to comparing the effectiveness of the two, we argue for the advantages of using them in combination. We do so in a research area in which we have an independent objective referent: government spending. With an eye toward capturing the accuracy of media coverage on public policy, we apply both hierarchical dictionary counts and supervised learning to measure mass media coverage of change in US defense spending. Both approaches appear to do well at capturing a media “policy signal” in the area, which provides an important test of convergent validity. While the results highlight the value of both dictionary and machine learning methods used independently, they also illustrate ways in which the two can be used in combination.