It was the social-science equivalent of Barbenheimer weekend: four blockbuster academic papers, published in two of the world’s leading journals on the same day. Written by elite researchers from universities across the United States, the papers in Nature and Science each examined different aspects of one of the most compelling public-policy issues of our time: how social media is shaping our knowledge, beliefs and behaviors.
Relying on data collected from hundreds of millions of Facebook users over several months, the researchers found that, unsurprisingly, the platform and its algorithms wielded considerable influence over what information people saw, how much time they spent scrolling and tapping online, and their knowledge about news events. Facebook also tended to show users information from sources they already agreed with, creating political “filter bubbles” that reinforced people’s worldviews, and was a vector for misinformation, primarily for politically conservative users.
But the biggest news came from what the studies didn’t find: despite Facebook’s influence on the spread of information, there was no evidence that the platform had a significant effect on people’s underlying beliefs, or on levels of political polarization.
These are just the latest findings to suggest that the relationship between the information we consume and the beliefs we hold is far more complex than is commonly understood.
‘Filter bubbles’ and democracy
Sometimes the dangerous effects of social media are clear. In 2018, when I went to Sri Lanka to report on anti-Muslim pogroms, I found that Facebook’s newsfeed had been a vector for the rumors that formed a pretext for vigilante violence, and that WhatsApp groups had become platforms for organizing and carrying out the actual attacks. In Brazil last January, supporters of former President Jair Bolsonaro used social media to spread false claims that fraud had cost him the election, and then turned to WhatsApp and Telegram groups to plan a mob attack on federal buildings in the capital, Brasília. It was a similar playbook to that used in the United States on Jan. 6, 2021, when supporters of Donald Trump stormed the Capitol.
But aside from discrete events like these, there have also been concerns that social media, and particularly the algorithms used to suggest content to users, might be contributing to the more general spread of misinformation and polarization.
The theory, roughly, goes something like this: unlike in the past, when most people got their information from the same few mainstream sources, social media now makes it possible for people to filter news around their own interests and biases. As a result, they mostly share and see stories from people on their own side of the political spectrum. That “filter bubble” of information supposedly exposes users to increasingly skewed versions of reality, undermining consensus and reducing their understanding of people on the opposing side.
The theory gained mainstream attention after Trump was elected in 2016. “The ‘Filter Bubble’ Explains Why Trump Won and You Didn’t See It Coming,” announced a New York Magazine article a few days after the election. “Your Echo Chamber is Destroying Democracy,” Wired Magazine claimed a few weeks later.
Changing information doesn’t change minds
But without rigorous testing, it’s been hard to figure out whether the filter bubble effect was real. The four new studies are the first in a series of 16 peer-reviewed papers that arose from a collaboration between Meta, the company that owns Facebook and Instagram, and a group of researchers from universities including Princeton, Dartmouth, the University of Pennsylvania, Stanford and others.
Meta gave unprecedented access to the researchers during the three-month period before the 2020 U.S. election, allowing them to analyze data from more than 200 million users and also conduct randomized controlled experiments on large groups of users who agreed to participate. It’s worth noting that the social media giant spent $20 million on work from NORC at the University of Chicago (previously the National Opinion Research Center), a nonpartisan research organization that helped collect some of the data. And while Meta did not pay the researchers itself, some of its employees worked with the academics, and a few of the authors had received funding from the company in the past. But the researchers took steps to protect the independence of their work, including pre-registering their research questions in advance, and Meta was only able to veto requests that would violate users’ privacy.
The studies, taken together, suggest that there is evidence for the first part of the “filter bubble” theory: Facebook users did tend to see posts from like-minded sources, and there were high degrees of “ideological segregation” with little overlap between what liberal and conservative users saw, clicked and shared. Most misinformation was concentrated in a conservative corner of the social network, making right-wing users far more likely to encounter political lies on the platform.
“I think it’s a matter of supply and demand,” said Sandra González-Bailón, the lead author on the paper that studied misinformation. Facebook users skew conservative, making the potential market for partisan misinformation larger on the right. And online curation, amplified by algorithms that prioritize the most emotive content, could reinforce those market effects, she added.
When it came to the second part of the theory — that this filtered content would shape people’s beliefs and worldviews, often in harmful ways — the papers found little support. One experiment deliberately reduced content from like-minded sources, so that users saw more varied information, but found no effect on polarization or political attitudes. Removing the algorithm’s influence on people’s feeds, so that they just saw content in chronological order, “did not significantly alter levels of issue polarization, affective polarization, political knowledge, or other key attitudes,” the researchers found. Nor did removing content shared by other users.
Algorithms have been in lawmakers’ cross hairs for years, but many of the arguments for regulating them have presumed that they have real-world influence. This research complicates that narrative.
But it also has implications that are far broader than social media itself, reaching some of the core assumptions around how we form our beliefs and political views. Brendan Nyhan, who researches political misperceptions and was a lead author of one of the studies, said the results were striking because they suggested an even looser link between information and beliefs than had been shown in previous research. “From the area that I do my research in, the finding that has emerged as the field has developed is that factual information often changes people’s factual views, but those changes don’t always translate into different attitudes,” he said. But the new studies suggested an even weaker relationship. “We’re seeing null effects on both factual views and attitudes.”
As a journalist, I confess a certain personal investment in the idea that presenting people with information will affect their beliefs and decisions. But if that is not true, then the potential effects would reach beyond my own profession. If new information does not change beliefs or political support, for instance, then that will affect not just voters’ view of the world, but their ability to hold democratic leaders to account.
Thank you for being a subscriber
Read past editions of the newsletter here.
If you’re enjoying what you’re reading, please consider recommending it to others. They can sign up here. Browse all of our subscriber-only newsletters here.
I’d love your feedback on this newsletter. Please email thoughts and suggestions to interpreter@nytimes.com. You can also follow me on Twitter.