Patrick Chester (New York University)
Abstract: There is substantial empirical evidence that indicates that democracy can spread between countries through observational learning. But do autocracies try to bias learning against democratic institutions through the use of targeted media framing? In this paper, I argue autocracies portray democratic institutions as being corrupt and chaotic in their news media relative to non-democracies and relative to non-autocratic media’s coverage of those same countries. To empirically test this hypothesis, I use word embeddings to identify the average cosine distance between words associated with political institutions and adjectives that carry the meaning of “chaos” or “corruption”. I do so by fitting word2vec word embedding models on large news corpora from the Gigaword project; separate word2vec will be fit on news associated with countries covered by Xinhua News, a major Chinese state media outlet, and those covered by two control media outlets that I do not expect to use this strategy: Taiwan’s Central News Agency and France’s Agence France Press. The average distance between sets of words describing political institutions and chaos and corruption will be the dependent variable in a fixed effects OLS regression where the units of analysis are the country and news publication. Findings consistent with my hypothesis would see smaller distances between political words and chaos and corruption adjectives for Xinhua’s news coverage of democratic countries relative their coverage of autocracies and to the control media outlets coverage of those same countries. This paper advances methodological research by Garg et al. (2018) by utilizing similar conceptual distance metrics to compare systematic differences in news coverage across units of interest (countries, in this case) and corpora; moreover, this is the first paper to use lexical distance identified using word embeddings to identify whether framing strategies using specific concepts are used to achieve political goals.