It’s not exactly a secret that extreme, less-than-accurate content finds a big audience on Facebook. In the months before last year’s election, the list of most-engaged-with pages on the site was almost always dominated by far-right figures like Dan Bongino and Dinesh D’Souza, who are not known for their fealty to fact-based journalism. An anonymous Facebook executive told Politico last September, “Right-wing populism is always more engaging.” New research released today, however, appears to be the first to show empirically that the relationship between accuracy and engagement varies dramatically based on where the source aligns on the partisan spectrum.
According to researchers at the Cybersecurity for Democracy project at New York University, far-right purveyors of misinformation have by far the highest levels of engagement per follower compared to any other category of news source. Indeed, the researchers found that while left-leaning and centrist publications get much less engagement if they publish misinformation, the relationship is reversed on the far right, where news organizations that regularly publish false material get up to 65 percent more engagement than ones that don’t. The study provides perhaps the most substantial evidence yet about what types of news—and fake news—perform best according to this metric on Facebook.
“What we find is that among the far right in particular, misinformation is more engaging than non-misinformation,” said Laura Edelson, a doctoral candidate and the lead researcher. “I think this is something that a lot of people thought might be the case, but now we can really quantify it, we can specifically identify that this is really true on the far right, but not true in the center or on the left.”
The analysis is an excerpt from an academic working paper. The team looked at 2,973 Facebook pages of US news sources that had been analyzed for partisanship and accuracy by two independent organizations, NewsGuard and Media Bias/Fact Check. This allowed the team to categorize each source both by ideological positioning—far right, slightly right, center, slightly left, far left—and by whether or not it had been flagged for regularly publishing false content. Of course, these rankings are an inexact science, but Edelson said the two databases were generally consistent with each other and with her own spot-checks of individual news sources.
Next, using CrowdTangle, a Facebook-owned tool that analyzes activity on the platform, Edelson and her team downloaded every public post from every one of the news organizations’ Facebook and Instagram pages for a five-month period between August and January, tallying how many likes, comments, or other interactions each page accumulated. This allowed them to rank each publication by engagement per follower. Finally, they plotted that engagement score against each category of publication.
The results were striking. In the far left, slightly left, and center categories, publications rated credible by NewsGuard and MBFC saw between two and five times as much engagement as ones that were not. (Fake news published by centrist organizations, the study notes, tends to be of the medical quackery variety.) In the slightly right category, accurate sources held only a slim edge. It’s in the far-right category that things get strange: Sources designated as purveyors of misinformation saw 426 interactions per thousand followers in an average week, compared to only 259 for far-right sources without the misinformation label. Both those engagement numbers dwarf any other category; the next highest is “far left, not misinformation,” at only about 145 interactions per thousand followers per week.
One implication of this finding, Edelson said, is that Facebook is more likely to steer users toward pages that include right-wing misinformation, since its algorithms are generally optimized for engagement. “Among the menu of options, we know that misinformation on the far right is even more engaging,” she said. “If something is attempting to blindly just promote content that is engaging, that is not going to be neutral. It will have an effect of promoting content from sources that spread misleading and conspiracy theories and fake news.”
In an email, a Facebook spokesperson said, “This report looks mostly at how people engage with content, which should not be confused with how many people actually see it on Facebook. When you look at the content that gets the most reach across Facebook, it’s not at all as partisan as this study suggests.”
After hearing a description of the new research, Sinan Aral, the director of the Initiative on the Digital Economy at MIT, said it aligns with past work on how false content spreads online. In a 2018 study published in Science, Aral and colleagues found that falsehoods spread on Twitter “farther, faster, deeper, and more broadly than the truth.” But, he said, the fact that certain pages get more engagement per user doesn’t necessarily mean Facebook promotes that material more heavily. “Facebook is way more sophisticated about measuring engagement,” he said. “They would have a different estimate of its ‘engagement,’ because they would have the causal estimates. They also have a multi-objective function algorithm that isn’t solely looking at engagement.”
That’s a complicated way of saying that we can’t know with scientific certainty whether Facebook amplifies the pages of unreliable right-wing news sources, because Facebook keeps that information to itself. In her paper, Edelson notes that CrowdTangle doesn’t provide information about impressions—how many people saw a given piece of content or how much time they spent looking at it. Facebook guards the workings of its recommendation algorithms even more closely. Without that data, it’s impossible to know precisely how the per-user engagement statistics uncovered by Edelson’s study translate into algorithmic recommendations. (Few people are more familiar with the issue of Facebook’s transparency, or lack thereof, than Edelson. The company recently served her organization, until recently called the Online Political Transparency Project, a cease-and-desist letter regarding a crowdsourced effort to get more information on how political ads are targeted.)
The research also opens up questions about the economics of right-wing fake news. Among the news sources flagged as spreading misinformation or conspiracy theories, 148 were categorized as slightly or far-right, compared to just nine left-of-center organizations. Does this mean disreputable publications are flooding the zone with shit, making their Facebook followers more misinformed than they otherwise would be? Or are they responding to demand for favorable coverage of Donald Trump and his fictional worldview, in the same way that Fox News pivoted toward encouraging stolen-election conspiracizing under ratings pressure from upstart networks like Newsmax and OANN?
Facebook has promised to crack down on certain categories of misinformation, including about the election and Covid-19. But it’s well established that Facebook’s algorithms favor more engaging content, and as news reports have repeatedly documented, if anything, it seems to be in the habit of bending its standard operating procedures in favor of right-wing news sources—not the other way around.