How did echo chambers influence the 2020 election?

Research from the Annenberg School for Communication shows that people are consuming news from more diverse sources, but many don’t consume any news at all. It’s too soon to tell what role that played in the recent race for president.

A close-up image of a hand holding a mobile phone, with the words "News," "Business," Politics" and "Sports" visible on the screen. Next to the phone is a cup of coffee. In the background are blurred open books.

Four years ago, Donald Trump’s victory disproved many polls that, in the run up to Election Day, put Hillary Clinton in the White House. Clinton supporters likely also heard from few or no outliers in their social media feeds, causing something of an echo chamber effect, where the same idea bounced around, reinforcing itself within that subgroup.

The 2020 election will go down in record books for the historic number of people who voted—about two-thirds of the eligible population—and because so many of those votes were cast by mail in the midst of a worldwide pandemic. But did people experience the same news bubble they did in 2016? Research from the Annenberg School for Communication’s Sandra González-Bailón and doctoral candidate Tian Yang offers some insight, in particular about the types of news sources people now consume.

Close-up headshots of two people. The person on the left wears a black jacket and red shirt and sits facing the camera. The person on the right wears a blue button-down shirt.
Annenberg School for Communication associate professor Sandra González-Bailón (left) and doctoral candidate Tian Yang. (Images: Courtesy of Annenberg School for Communication)

“Far from encouraging people to remain enclosed in echo chambers of similar others, we’re finding that those bubbles are breaking up,” says González-Bailón, an Annenberg associate professor and affiliated faculty at the Warren Center for Network and Data Sciences. “People are actually exposed to and consuming news from a wide range of sources.” The researchers published these findings in the Proceedings of the National Academy of Sciences.

Penn Today discussed with González-Bailón and Yang what this means in the context of the recent election, a win for President-elect Joseph R. Biden Jr., as well as what else we need to understand about how people process the news they read and watch.

How did this line of research emerge?

González-Bailón: This is part of a project in which we analyze how people consume news online, partly to test claims that digital technologies are fragmenting the public domain, allowing us to increasingly enclose ourselves into echo chambers or listen only to those opinions that are aligned with our own. We wanted to analyze what people actually do when they go online to gain exposure to political news from news sources.

What made this project interesting for us is that the data we have access to is the most comprehensive dataset that’s been used to analyze exposure to news. We’re tracking web-browsing behavior from 2014 to 2018, including mobile access and a wide range of news sources.

Your dataset included more than 250,000 people and more than 400 news outlets. What did your analysis show?

González-Bailón: Our main finding is that people are actually exposed to and consume news from very diverse sources. In other words, co-exposure to news domains is going up, especially when you incorporate into the analysis browsing behavior through mobile devices.

Our research looks at which news domains you access—The New York Times or CNN, for example—but also which other sources you’re consuming. If you go to CNN, do you also go to Fox News? If you go to The New York Times, do you also go to the Financial Times? This is how we analyze exposure to different sources and identify growing levels of co-exposure over time.

However, we also find that half of the people who go online do not access any of the news domains we were tracking. And our list of news domains is pretty comprehensive.

What does that latter finding tell you? Where do those people get their news?

González-Bailón: We’re only tracking online activity, not TV or printed media, so some fraction of the people not accessing news on the web are probably consuming it through those other channels. We are particularly interested in the role that social media play in informing people who opt out of voluntary news exposure. People who are not going to news domains are probably viewing news on social media like Facebook.

When you’re scrolling through your feed, you see news stories or other forms of political content that your friends or the groups you follow are posting. Our research is only tentative in this domain, but we think that the half of the online population opting out of news might be potentially more vulnerable to information manipulation on social media platforms. This is something that requires more research to substantiate.

Yang: We also tried to identify selective exposure to news content along ideological lines. Ideological selective exposure is the idea that people will select news sources that align with their own political inclination. So, we tried to test whether this could account for the co-exposure patterns we identify in our data. We found that evidence of selective exposure is not very strong.

Choosing news sources based on their political slant seems important within the framework of a presidential election. Can you explain how it worked in the context of your study?

González-Bailón: For some of the panelists whose online behavior we were tracking, we also had information about how they identified in terms of partisanship and political outlook, so conservative versus liberal. We know what they do online. We also know how they position themselves in those two ideological scales.

On the basis of that, we assigned ideological scores to news outlets. Breitbart, for example, is on the conservative side of the scale, and BuzzFeed is on the liberal side. We did this with all the news outlets and then tested whether audience overlap is stronger between outlets closer on that scale. We found that ideology explains only a small percentage of the co-exposure that we analyze in our data. This means that ideological self-selection is not the main driver of news consumption. In other words, people are systematically consuming sources that are quite far apart on the ideological scale.

In this research, you measure exposure, but that doesn’t indicate necessarily how people process the information they read, right?

González-Bailón: The million-dollar question here is, does exposure explain behavior? Meaning, if your exposure is diversified, if you consume Fox News and CNN, is that going to change the way you think about political issues? We don’t have an answer to that. Exposure alone doesn’t tell us too much about the effects of the exposure, but it’s a first step to understanding how the information we consume shapes our beliefs or behaviors.

With all of this in mind, how did you expect it to play out in the election? What effect might you expect news consumption to have had on how people voted?

González-Bailón: Thinking about this election, to me, the crucial question is what happens with the 50% of users who are not consuming news on the web. What type of political information are they exposed to? A lot has been said about the impact that Facebook and Twitter are having on the election. To me, the key in that discussion is, what happens with users whose only source of political information is Facebook? Are they more vulnerable to misinformation campaigns than politically motivated users engaging with the news elsewhere?

Yang: In our research, we showed that from 2014 to 2018, audience fragmentation is decreasing. If this trend were to continue into 2020 then that might mean the current information environment is more promising in terms of bridging different views compared with the last election. It is good for the democratic process that people are consuming news from a wider range of sources. What is not so good is that only half of the online population is engaging with news.

González-Bailón: I agree. Also, the way many of these platforms operate has changed. Facebook and Twitter have been way more aggressive in terms of labeling content that’s misleading, including the president’s tweets. My expectation is that the trends will continue as they’ve been evolving the past four years. But exposure to news on the web or on social media are only two pieces in a larger puzzle; there are others we have to put together to understand the complex news ecosystem in which we live today.

Like what?

González-Bailón: One important factor is how platforms regulate content. As I mentioned, they have been more aggressive this year, labeling misleading posts but also introducing measures aimed at curtailing the viral diffusion of that content. These changes have an impact on how people behave on the platforms. 

Another important factor is the divide between people who are highly interested in politics—which is not the majority—and people who are not. This affects the data we collect and analyze. By analyzing news source exposure, we miss the online behavior of those who aren’t consuming news but who may be exposed to other forms of political information relevant to understanding their behavior, opinions, or choices. We have to get better at meeting people where they are, not where we would like them to be.

What can your work tell us about these difficult-to-capture groups and where we go from here?

González-Bailón: We are starting to realize how important it is to pay attention to people who are not that interested in politics and therefore do not engage in the type of behavior we measure. The question is, how can we learn more about that section of the population? It’s reassuring to find evidence of decreasing fragmentation in news audiences, but it’s far less reassuring to also identify such a large divide between those who consume news and those who don’t. The next step is to try to understand better the fraction of the online population that opts out.

Sandra González-Bailón is an associate professor in the Annenberg School for Communication and affiliated faculty at the Warren Center for Network and Data Sciences at the University of Pennsylvania.

Tian Yang is a fourth-year doctoral student in the Annenberg School for Communication at the University of Pennsylvania.