Sharing on Facebook reveals 2 very different news environments

COLUMBUS, Ohio – A first-of-its-kind study examined 2.2 million news stories shared on Facebook and found that publishers create two very different news environments.

These distinct ecosystems involve low-credibility publishers – those that publish what is sometimes referred to as fake news – versus high-credibility publishers.

Findings showed that while these two types of publishers often pushed out bursts of coverage at the same time – a common feature of news coverage – they were often about different topics, said Kelly Garrett, senior author of the study and professor of communication at The Ohio State University.

“These aren’t just random differences in what is being covered – there seems to be some systematic bias happening here,” Garrett said.

“Low- and high-credibility publishers often give their readers distinct versions of what are the most important topics at a particular time.”

The results may have troubling implications for news consumers and for those trying to fight the spread of misinformation on social media, Garrett said.

The study was published this week in the journal New Media & Society.

This research comes from a team that has access to a unique collection of anonymous data from Facebook on the sharing of online news content. Their goal was to examine the supply of low-credibility information, which may pose harmful influences on what people learn about science, politics and their community.

The researchers examined more than 2 million online news stories that were shared at least 100 times by Facebook users from February 2017 to April 2019.

A publisher was marked as low-credibility if it was included in one of six lists of low-credibility news domains created by reputable fact checkers and scholars, or if it produced more than five articles labeled as problematic by Facebook fact checkers. Some examples include Breitbart News Network and Daily Kos.

High-credibility sites were identified by two academic sources of traditional news sites.  Examples included the New York Times and the Wall Street Journal.

The researchers examined news stories shared on Facebook to identify bursts of coverage that occurred when many outlets ran stories on the same topic at the same time. For example, the study found multiple bursts of coverage concerning the family of Barack Obama, the Trump organization, Medicaid, NASA and the Emmy Awards, among many others.

Findings showed that low-credibility publishers more often converged on stories about politics and the government than did high-credibility publishers, whose topics ranged more widely.

Researchers found that after periods of relative quiet in terms of the number of stories run, publishers put out bursts of coverage during certain periods.  It was notable that both high- and low-credibility publishers tended to put out bursts of coverage at the same time – but they often were about different topics.

When high-credibility publishers put out a burst of coverage on a particular topic – say Trump or NASA – only 40% of the time did low-credibility publishers have a burst of coverage on the same topic at the same time.

“That’s not trivial, but it is lower than what we would expect if credibility played no role in what these outlets covered,” Garrett said.

The study can’t tell why the two types of publishers put out bursts of coverage on different issues at the same time, Garrett said.  But it is consistent with the belief that some partisan news outlets try to counteract negative news about their ideological side by focusing their news coverage on another topic.   

However, Garrett noted that ideology is not the biggest divide between the two types of publishers in this study.

While the researchers found, as expected, differences in coverage based on political ideology, the biggest difference in what publishers covered was based on their credibility, not on their partisan stance.

“We would expect conservative- and liberal-leaning publications to behave like other outlets that share their ideology, and they do. But we see the credibility divide is even bigger than the ideological divide in determining what topics they cover,” Garrett said.

“In other words, publishers’ preferences for covering topics covered by outlets of similar credibility is much stronger than their preference for covering topics of outlets sharing similar ideology.”

Garrett noted that low-credibility publishers were originally distinguished from high-quality publishers by how they report the news.  But these results suggest the difference runs deeper to include what they cover and when they cover it.

High-credibility publishers have long been seen as following “pack journalism” in which news outlets look to see what stories other similar outlets are covering to determine their story selection.

Although this study can’t say for sure, the same may be happening among low-credibility publishers – although their definition of the most important stories of the day is often distinct from their high-credibility peers, Garrett said.

“Low-credibility publishers may have a different set of news values than the high-credibility publishers,” he said.

These results suggest it may be more difficult than often thought to counter “fake news” put out by low-credibility publishers, according to Garrett.

Fact checkers often work at the story level, trying to correct misinformation put out in one article by one publisher.  But this study suggests bursts of coverage in low-credibility publications may be putting out misleading information on the same topic in a variety of forms and in a variety of news outlets.

“We tell people that they should get their news from a variety of sources to reduce the risk of being misled. But that doesn’t work as well if many low-credibility outlets are talking about the same stuff,” Garrett said.

“There’s a little more nuance to what it takes to be a well-informed person.  You can’t just be reading multiple news stories.  You have to be reading sources that are well known to be credible.”

The lead author of the study was Ceren Budak, associate professor of information at the University of Michigan.  Other co-authors were Drew Margolin from Cornell University, Lia Bozarth of the University of Michigan, Jason Jones of Stony Brook University and Robert Bond of Ohio State.