2. Counting Words & Phrases - [E1] 1. Gentzkow, Matthew & Jesse M. Shapiro. 2007.

[JunsolKim]

Post questions here for this week's exemplary readings: Gentzkow, Matthew & Jesse M. Shapiro. 2007. “What Drives Media Slant? Evidence from U.S. Daily Newspapers.” Econometrica 78(1): 35–71.

[pranathiiyer]

This was an interesting read! " Using zip code-level data on newspaper circulation, we show that right-wing newspapers circulate relatively more in zip codes with a higher proportion of Republicans, even within a narrowly defined geographic market". I wonder how things might have changed more than a decade later. With online prints of newspapers and articles becoming so widely accessible, and the prevalence of news of twitter and other platforms becoming so ubiquitous, would it be possible to find such a correlation? (taking into account how content of newspapers itself might have changed)

[melody1126]

On page 56, the authors explain a new instrumental variable: "the share of the newspaper's market attending church monthly or more during 1972-1998." According to the authors, this instrument was well-correlated with X (market % republican). Ideally, instruments are uncorrelated with the error term of Y (media slant in newspapers). Since the frequency of church attendance is an attribute of the residents of a zip code area, it would probably be uncorrelated to the media slant of newspapers except through the fact that they circulate in that area.

But why did the authors introduce this statistical instrument at all? What do we gain from introducing instrumental variables?

[Sirius2713]

Very interesting paper! Using the frequencies of diagnostic phrases, the authors show that they can infer the political propensity of a newspaper. Therefore, is it possible to use this method to detect the change in the political propensity of some newspapers circulating in battleground states? Furthermore, is it possible to use this method to infer the slant of those battleground states in election years? For example, is it possible to predict the change in the political propensity of Georgia in 2020 or Virgina in 2021 by analyzing diagnostic phrases from newspapers circulating in these areas?

[Qiuyu-Li]

This is a very creative, sophisticated, and well-designed paper to me, and I can hardly think of any flaws that would render its effectiveness questionable. The only point that might be a little bit weak is their way of measuring leanings. What they did was basically select "representative phrases" featured in each party's speeches, and then calculate how frequently these phrases have occurred in the newspaper. However, it's not likely that they can exhaust every politically leaned phrase, and thus the index calculated out of it may not be accurate for quantifying leanings. But I think they did an excellent job overall and I can't think of a better way of measuring the leanings.

[konratp]

This was a very interesting paper to read! Reading this in 2021, I wonder how much has changed in American politics since 2005, and how these changes would affect an attempt at reproducing the study? There are several levels of analysis that come to mind that this paper does not really address. For example, how would politicians not represented in Congress affect the analysis? Donald Trump immediately comes to mind, but many other "outsiders" as well who ended up running for office in the US. Trump (pre-2017), Andrew Yang, and others have significantly affected political discourse without holding political office. In the case of Trump one could argue that his language in the 2016 campaign matched his future constituency's interests much better than any Republican congressperson running for president that year. How do such phenomena complicate the results of the study?

Another potential limitation that comes to mind is the assumption that the congressperson's ideology is directly related to the electoral outcomes in a previous election. How would the authors explain phenomena like Joe Crowley consistently winning against his Republican competitors in general elections by wide margins, yet being considered one of the more conservative Democrats in the House? Further, a centrist politician of either party might win by wide margins precisely because of his/her centrist attitudes, yet in this model a wide margin would be interpreted as a strong ideological slant.

[NaiyuJ]

1) I was looking at their tables about the phrases used more often by Democrats and the phrases used more often by Republicans comparatively. I think there is a difference between what to cover and how to cover in regard to the newspaper coverage as well as the bills. For example, if two newspapers reported the same news using different words but conveying the same content, we can say that they are biased in how to cover the news. However, if two newspapers were reporting totally different news that maybe a particular party usually gave special attention to, then is this kind of comparison meaningful? Would this influence how we interpret the results? I guess the best way to test the hypothesis is that we control the content of the news and see their differences in the use of language. But now it seems that they also compare what to cover.

2) Another thought is that this paper has focused primarily on text-based measures, how about comparing the images that are mostly used in different newspapers (like Trump vs. Biden)?

[ZacharyHinds]

I am also interested in how the landscape change of American politics since this article has changed this study. I would also wonder then, how exactly the massive growth in Social media (and its related growth of importance within the realm of politics) has affected the reliability of their index? Additionally, could it be possible or relevant to attempt to create a similar index to measure political slants that arise within social media, especially as it pertains to decisions made by the platform itself (such as through moderation, advertising, algorithms, etc.)?

[facundosuenzo]

Super interesting paper! They tackled many alternative hypotheses throughout the manuscript and provided different sources of robustness for the measures. The only theoretical assumption that may be worth revisiting is that circulation of newspapers != to consumption nor actual reading. The scholarship on this has documented the fact that people are very used to receiving the physical copies at home, but not necessarily for their content but as a ritual, they want to sustain. Second, deriving policy implications from this could be risky since, as they cited, newspapers are likely to be consumed by white, older, and more affluent males.

[sizhenf]

A very interesting topic! And I particularly liked how the relationship between owners and readers of the media are explained in the paper using the simple economic model of demand and supply. Using rigorous statistical analysis, the paper finds that variation of media slant is largely due to consumer preferences but not ownership of the media. I wonder if a similar research can be done to study the variation of user values/ideology of social media. As we observe in the real life, users of difference social media platform tend show dramatically different viewpoints. I wonder how this variation has been evolved and if the companies have intentionally chose to promote content to cater its target users that can further increase the slant.