The Filter Bubble Revisited

A new study suggests online media aren’t to blame for political polarization—yet.

Photo illustration by Natalie Matthews-Ramo. Photos by Bryan R. Smith/AFP/Getty Images, Sara D. Davis/Getty Images, John Sommers II/Getty Images and Kyle Grillot/AFP/Getty Images
We have a filter bubble problem—it just might not involve online media.

Photo illustration by Natalie Matthews-Ramo. Photos by Bryan R. Smith/AFP/Getty Images, Sara D. Davis/Getty Images, John Sommers II/Getty Images and Kyle Grillot/AFP/Getty Images.

In 2011, with Facebook and Google growing in influence, liberal activist and entrepreneur Eli Pariser wrote a best-selling book that coined a term: The Filter Bubble. Personalized news feeds and search results, he warned, would undermine civic discourse by steering people toward information that appeals to their preconceptions. We would search for, like, and retweet the ideas we already agreed with, and algorithms optimized for engagement would serve us more of the same—crowding out anything that might trouble our worldview.

Five years later, Pariser’s work was hailed anew. The electoral triumphs of Brexit and Donald Trump, which germinated in the online warrens where like-minded declinists congregate, seemed to validate his theory, at least to the establishment pundits who failed to see them coming. The Guardian profiled him as a sort of social media prophet; Bill Gates weighed in; a Wired headline chided, “Your Filter Bubble Is Destroying Democracy.” Liberals and moderates, suddenly more aware of their own cozy echo chambers, resolved to better understand what the other side was thinking.

But if filter bubbles are destroying democracy, a new study suggests they aren’t doing it in quite the ways you might expect—or to the extent you might assume.

We know that the American electorate has become more polarized in recent decades in multiple ways. But a new working paper from economists at Brown University and Stanford University, which studies the relationship between polarization and the use of the online media in American adults from 1996 to 2012, suggests the self-refining contours of your Facebook feed are not to blame. There are lots of ways to measure polarization, so the authors split the difference by combining nine different plausible metrics from the academic literature into a single index. Their counterintuitive finding: Polarization has been driven primarily by the demographic groups that spend the least time online.

Specifically, the researchers find that Americans older than 75 experienced by far the greatest ideological divergence of any age group over the time period studied. Yet just 20 percent of this group reported using social media as of 2012. In contrast, the vast majority (80 percent) of Americans aged 18 to 39 used social media. Yet according to the study’s findings, this younger group was hardly any more polarized in 2012 than it had been in 1996, when online media barely existed.

“These facts argue against the hypothesis that the internet is a primary driver of rising political polarization,” conclude authors Levi Boxell and Matthew Gentzkow of Stanford, and Jesse Shapiro of Brown.

That doesn’t mean there’s zero filtering at work in online media, of course. The phenomenon Pariser described is real, and polarization is just one of its putative effects. There are also some important limitations in this paper’s data and methodology that might help to explain why it found so little correlation between polarization and reading the news online.

For instance, while the authors’ macroscopic approach captures differences between age groups, it may obscure trends within age groups along lines such as party affiliation, income and education level, or intensity of engagement with social media. And, importantly, while the study covers the period in which online media grew from infancy to near-ubiquity, the data capture only the first few years of the algorithmic personalization trend that Pariser described. Facebook introduced the “like” button in 2009; Google began personalizing search results the same year. Twitter launched in 2006 and introduced a ranking algorithm just last year.

Even so, if the authors’ polarization index is to be trusted, it’s noteworthy that so much of our country’s rise in partisanship transpired offline among people who didn’t even have Facebook accounts. The finding is consistent with a previous body of work by Gentzkow, Shapiro, and others, who have found little evidence for the hypothesis—advanced by legal scholar Cass Sunstein—that the internet is tearing us apart. In 2011, for instance, they reported that people were actually more likely to encounter opposing views in online media than they are in their day-to-day interactions with neighbors, co-workers, and family.

A different group of researchers reported somewhat analogous findings in a 2015 Science study funded by Facebook: The news feed algorithm did filter ideologically “cross-cutting” news to some extent but less so than users’ own choices of what to read. (That study was widely criticized for both its methodology and Facebook’s self-serving interpretation of the data.)

Less controversially, this newest study is a useful reminder of a simple fact that media pundits are prone to overlook. As central as Facebook and Twitter are to the news consumption habits of certain groups—like, cough, the media—most Americans still get their news through more traditional channels. Even as nearly half of Americans now read at least some news on Facebook, the majority still cite television as their primary news source. And while Facebook took the brunt of media criticism in the wake of the 2016 election, Pew found that voters relied more heavily on Fox News and CNN.

Pariser himself is well aware of that, as he told me this week when we discussed Boxell, Gentzkow, and Shapiro’s paper. “I’m not surprised,” he said of their findings on polarization and social media, though he added that finer-grained research would be needed to properly interpret their results. “I’ve always been worried about overclaiming what social media is doing to us now based on the way that people actually consume media.”

Yet Pariser noted that the influence of social media on the news continues to grow, and he believes the filter bubble problem is growing with it. Facebook and Google today are already quite different from Facebook and Google in 2012, when the Boxell et al. study ends, and the Facebook and Google of 2020—or whatever platforms complement or displace them—are likely to look different still. “We haven’t yet reached the event horizon where social media is the primary driver of how we consume things,” he said. “I still believe that across that event horizon are strange and scary phenomena.”

Meanwhile, Pariser highlighted a different way in which filter bubbles may be contributing to polarization. Some of the people who rely most heavily on social media for news, he pointed out, are members of the media itself. If journalists’ own filter bubbles are influencing what news they cover, and how they cover it, that could help explain the vast and still-growing partisan divide between the audiences of more traditional outlets such as Fox News and CNN. Those cable news networks are especially popular among older Americans—the same ones who are leading the polarization trend, according to the study.

And then, of course, there’s the reality that the political media have only so much power, and they tend to reflect public opinion at least as much as they shape it. As Trump’s rallies reminded us, a lot of politics still happens at the local and interpersonal levels: if not at a strip mall, then in a small arena.

But if social media isn’t driving polarization as much as we might assume, then what is? I put that question to Shapiro, one of the study’s authors, and he demurred: “I honestly don’t know. I would think, given how demographically broad we find the polarization trend to be, whatever is driving it is also likely to be something that’s very demographically broad.”

Shapiro acknowledged that more research is needed to fully understand the relationship between social media and polarization. “I think part of it is, traditional media remain very important,” he said. “And part of it is that, although the internet does provide some scope for people to have really segregated news diets, they’re not as segregated as people imagine them to be. The notion that people are out there reading only really extreme sources and no mainstream sources just didn’t wash with the data we looked at” in 2011. “That’s not to say the opportunities for it aren’t changing. It’s just not as powerful a factor as it might seem.”

In other words, whatever future our online media habits will bring has yet to fully arrive. And it’s not a foregone conclusion that it will be a hyperpartisan dystopia—at least, not any more than the one that’s already reflected on cable news.