Both of these papers examine the trend towards “selective exposure” and away from diverse ideological exposure when viewing news electronically. At a high level, both papers are touching on the same big idea – that users seem to be creating insular “echo chambers” of polarized news sources based on their ideals, ignoring the viewpoints of the opposing ideology either by their own conscious choice or algorithmically by their past behavior. The Garrett paper looks at general web browsing for news sources and focuses on the area of opinion reinforcement. The study details a web-administrated behavioral study in which participants were shown a list of articles (and their summaries) and were given the choice of which ones they wanted to view. The study findings supported the author’s hypotheses, including that users prefer to view news articles that are more opinion-reinforcing and that users will spend more time viewing those opinion-reinforcing articles. The Bakshy et al. study was Facebook-centered, examining how users interact with shared news articles on that platform. Among their findings were that ideologically cross-cutting depended on both the spectrum of friend ideologies and how often those friends shared, but that there was some evidence of ideological isolation in both liberal and conservative groups.
Both of these studies had notable limitations that were discussed by the authors, but I felt that each was addressed insufficiently. The Garrett study made use of both a liberal and a conservative online news outlet to obtain participants, which obviously will not ensure that the sample is representative of the population. Garrett justifies this by supposing that if selective reinforcement is common in these groups, then it is likely the same among mainstream news readers; however, (1) no attempt is made to justify that statement (the brief mention in the Limitations section even contradicts this assertion), and (2) my intuition is that the opposite is true: that if selective reinforcement is common among centrists, then it almost certainly will be true at the ideological extremes. In my opinion, the results from this study do not generalize, and this is a killer limitation of the paper.
Bakshy’s study has a similar limitation that the authors point out: that they are limited to recording engagement based on clicks to interact with articles. As a result, individuals might spend some time reading the displayed summaries of some articles but never click to open the source, and such interactions are not logged. To use the authors’ phrasing, “our distinction between exposure and consumption is imperfect.” This surprised me – there was no way to record the amount of time that a summary was displayed in the browser, to measure the amount of time a viewer may have thought about that summary and decided whether or not to engage? I know in my experience, my newsfeed is so full and my time is so limited that I purposefully limit the number of articles that I open, though I often pause to read summaries in making that decision. I do occasionally read the summaries of ideologically-opposing articles, but I rarely if ever engage by clicking to read the full article. Tracking exposures based on all forms of interaction would be an interesting follow-up study.
Despite the limitations, I thought that both studies were well-performed and well-reported with the data that the authors had gathered. Garrett’s hypotheses were clearly stated, and the results were presented clearly to back up those hypotheses. I wish the Bakshy paper had been longer so that more of their results could be presented and discussed, especially with such a large set of users and exposures under study.
