Five Things We Know About The “Filter Bubble Thesis”

Will Rinehart
6 min readApr 27, 2016


Source: Flickr User carterse

In a recently penned piece, Megan McArdle argued that even as the American political conversation has placed a higher priority on diversity, people are “increasingly dividing themselves into like-minded bubbles where other people, with other experiences and viewpoints, almost never penetrate.” This is a view that has been gaining steam among journalists involved in politics and privacy advocates who would like to see further regulations on Internet companies. Yet, the research doesn’t allow us to make a statement as clean as that.

As McArdle explains,

Social media, of course, makes this problem worse. Even if we are not deliberately blocking people who disagree with us, Facebook curates our feeds so that we get more of the stuff we “like.” What do we “like”? People and posts that agree with us. Given that Facebook seems to be the top news source for millennials, and an increasingly important one even for folks who grew up skimming dead trees for information, that matters quite a lot.

But a closer look at the third report cited pokes some holes in this neat chain of logic:

Contrary to the idea that social media creates a polarizing “filter bubble,” exposing people to only a narrow range of opinions, 70 percent of Millennials say that their social media feeds are comprised of diverse viewpoints evenly mixed between those similar to and different from their own. An additional 16 percent say their feeds contain mostly viewpoints different from their own. And nearly three-quarters of those exposed to different views (73 percent) report they investigate others’ opinions at least some of the time — with a quarter saying they do it always or often.

For the most part, echo chambers and the filter bubble thesis are interchangeable terms that refer to the narrowing of topics and views that occurs among people of the same background while online.

However, this has long occurred offline. Homophily, as it is known, is the term to describe the homogenous nature of personal networks. One person’s sociodemographic, behavioral, and intrapersonal characteristics in the real world tend to be fairly narrow. So, segregating the effects of the medium with the already existing social patterns is difficult. Do social networks narrow or expand the range of characteristics? And how, exactly, would we measure this change? Researchers are perplexed by these problems.

There is a lot here to unpack, so let’s step back for a second and actually review five things we know about this broad topic.

1. If filter bubbles exists, it might be due to consumption of diverse viewpoints.

Decades of work under the banner of motivated reasoning shows that new and contradictory information actually entrenches beliefs. In an effort to maintain specific beliefs and stave off cognitive dissonance, individuals will find supporting reasons for their views and will discount information that is contrary to that belief. As one study found, those that occupy either extreme of the political spectrum, be it liberal or conservative, tend to be less influenced by outside information even in simple tasks. Partisans also tend to think the media is against them even if information is neutrally presented. The idea has been dubbed the hostile media effect.

In short, exposure to disparate viewpoints tends to make you more sure in your own. So, if we accept that filter bubbles caused by social media actually do exist, it might be because people are getting to see what is out there, which is a paradoxical outcome. In other words, what might be perceived and measured as an echo chamber might be due to the exposure of more diverse voices.

2. Polarization does exist, but mainly for political topics.

In an analysis of Twitter data, Pew found that conversations can fall into six basic archetypes of conversation. While polarization does exist, it is contained largely of one kind of conversational topic, politics. Hobbies, conferences, brands, public events, and the news all display different organizational structures, detailed here:

Source: Pew analysis of Twitter data

So even if there is polarization for political topics online, isn’t the case that this partisanship spills over into non-political conversations. Using a more dynamic method of analysis, an empirical analysis of the 2012 presidential election, 2013 government shutdown, the 2013 Boston Marathon bombing, and the 2014 Super Bowl found that ideological preferences in the case of the first two political issues did not occur with the last two non-political events. The researchers concluded saying, “previous work may have overestimated the degree of ideological segregation in social-media usage.”

3. Most people aren’t going online to talk about politics.

While many in the academy and in the press are heavily entrenched in political discussions, most Americans aren’t actively involved in the dialogue, even if they consume political news.

Pew research finds that among all people, only 13 percent are talking about politics every day, while another 29 percent talk about politics a few times a week. The vast majority, about 58 percent of people, are discussing politics a few times a month or less.

The same kinds of dynamics seem to exist online as well. About a month back, Matt Bruenig shared this chart he made from Census data showing how many people use the Internet to express political opinions:

Source: Matt Bruenig

A vast majority of people aren’t going online to express political opinions, much as they are likely to do offline.

What people actually do online is engage in “pointless babble,” the “I am eating a sandwich now” posts as one analysis described it. Linguists however have a different term, they call it phatic communication. Phatic communication is meant to express sociability and maintain bonds, not convey information. And this is the power of social media. Social media allows us to share our feelings, needs, and simple statements. They allow us to be social and to connect, and to share life moments. And guess what? Most of life isn’t drawn down party lines.

4. The more partisan you are, the more likely you are to discuss politics.

Discussing politics is heavily dependent on how strongly you identify with party politics. Those who identify as either consistently conservative or consistently liberal are far more likely to discuss politics a few times a week or nearly every day than those who hold more centrist beliefs even within their party.

Source: Pew

5. Actual consumption of online media seems to be fairly balanced.

Few have analyzed actual news consumption, because it poses difficult data collection issues, but one study by economists Matthew Gentzkow and Jesse M. Shapiro found that people tend to visit a relatively centrist set of websites. Isolation indices have long been used to track racial segregation in housing. Gentzkow and Shapiro adopted the concept for online news consumption, where 100 shows ideological isolation and 0 shows none. While it is true that the news consumption online (7.5) shows more segregation as compared to broadcast television (1.8) and local news (4.8), it is still lower than national newspaper consumption (10.4). Moreover, Internet news consumption was found to be significantly lower than “actual networks formed through voluntary associations (14.5), work (16.8), neighborhoods (18.7), or family (24.3).”



Will Rinehart

Senior Research Fellow | Center for Growth and Opportunity | @WillRinehart