Open lkcao opened 8 months ago
Gentzkow & Shapiro (2007) use text analysis and supply-demand models to provide a strong analysis of how political "slant" of consumer demand correlates with the slant of newspapers.
Throughout my reading of this article as a whole, I had this broad question (not related to any part of the reading in particular): Is the question of “What drives media slant” perhaps an unanswered Q only for people not working at/owning these media companies? Answering this question would then be a kind of leveling of information. What does this tell us about the nature of social science inquiry (or am I potentially challenging the inevitable/unchangeable)?
This paper compares phrase frequencies in the newspaper with phrase frequencies in the 2005 congressional record to identify whether the newspaper's language is more similar to that of a congressional Republican or a congressional Democrat. Using this as the media slant index, the authors smartly illustrate that media firms respond strongly to consumer preferences rather than their own identity. Based on this, I think there could be deeper investigations on why there's a disparity in the explanatory power of firm identity and consumer preference. If the media companies cater to consumers' slant, why don't they expose the tendency explicitly?
I am particularly impressed by the authors' appendices. Most of my questions about their methodology were well answered in that section of the paper. Though I'm not an economist by training, Gentzkow and Shapiro (2010) do a good job to argue that more than a newspaper owner's identity, consumer preference (aka demand) drives media slant. However, I am not fully convinced by their measures to elucidate causality. They take into consideration one variable: religiosity. They mention that this variable is highly predictive of political preferences and less likely to be affected by newspaper content, but to me (as well as the authors) it remains unclear the causal direction. In a different direction, If the authors could redo this research in today's political climate in the US, how would their results differ? Would their findings still hold?
Part 3 (p. 42) and the appendix with table explicitly illustrate what it means to dig out phrases unique to certain sources. That's very helpful to visualize what I might want to look for across the corpus of my interests. A good example to extrapolate in case you want to use the different use of terminology among contents --- This paper focused more on economics and politically related slants; what resonated with me is the use and evolution of academic terminologies in different fields/journals.
I'm actually curious about the cleaning process --- as a reader, I take for granted the tables on p.44 and p.45, but as I think about how to obtain such a cleaned table with only related terms, it almost feels like a miracle: why there are exactly such reasonable amount of phrase usage differences? In the real research scenario, would it be common to experience a situation where you find two corpus have 1) too few differences or 2) too much/noisy differences such that you have to modify the research strategy and goals?
I think this is an interesting paper. I actually did something similar before, where I conducted a NER on the headlines of online news media to extract key political terms, and found media from both ends of the political spectrum have the similar preference of political news topic : for example, if you see an increase discussion of Trump in left media at a given time, you likely would also observe an increase in the right media. This suggests that the author could expand the method to identify news bias by considering the words used by both parties but potentially with different setting and sentiment, in order to improve the accuracy of the model to calculate slant of news media.
Post questions here for this week's exemplary readings: