Worried that your feed is quietly steering elections? On social media and democracy, the major 2020 US experiments found big changes in what people saw, but little change in political attitudes over weeks to months.
This guide is part of our wider coverage of technology and society.
| Item | Value |
|---|---|
| People who deactivated Facebook and Instagram before the 2020 election | 35,000 people, for six weeks |
| Length of the feed-algorithm experiments | Three months |
| Untrustworthy-source content on Facebook with a chronological feed | Up 68.8% |
| Uncivil posts on Facebook with a chronological feed | Down 43% |
| Untrustworthy-source content on Facebook after removing reshares | Down 30.6% |
| Frequently shared false-rated news links seen by more conservatives than liberals | 97% |
The figures come from the Center for Social Media and Politics summary of the first papers and an NBC News report on the deactivation study. Figures can change as papers are updated or new ones appear, so check the source before you rely on them.
What the 2020 research on social media tested
The best-documented evidence comes from a research partnership between Meta and outside academics around the 2020 US election. The University of Texas at Austin called it the most comprehensive research project to date on social media in American democracy, as of its 27 July 2023 release. Talia Stroud of UT Austin and Joshua Tucker of New York University co-led the academic team.
The academics worked with Meta researchers to design the experiments. Platform-wide data reached them only in aggregated form to protect user privacy, according to the Center for Social Media and Politics.
The researchers changed three things in participants’ feeds for three months, all with consenting users:
- Removing reshared content.
- Deprioritising content from like-minded sources.
- Switching from an algorithmically ranked feed to a chronological one.
Independence matters when you weigh the results. The project’s design gave the lead academics control of write-up decisions, but its independent observer Michael Wagner said that Meta set the agenda in ways that affected the overall independence of the researchers, a point the Tech Policy Press primer records.
Does the feed algorithm change what you see?
Yes, and by a lot. Replacing the personalised feed with a chronological feed cut the time participants spent on the platforms and how much they engaged with posts, according to the Center for Social Media and Politics.
The content mix shifted in ways that cut both directions:
- Untrustworthy-source content rose by 68.8% on Facebook and 22.1% on Instagram.
- Uncivil posts on Facebook fell by 43%.
- On Facebook, the share of content from like-minded sources fell from 53.7% to 48.1%, per the Tech Policy Press primer.
Removing reshares also reshaped feeds. It reduced the proportion of political content by nearly 20% and cut content from untrustworthy sources by 30.6%, as the same summary reports. The lesson is that a feed’s design is a powerful lever over exposure, even when attitudes barely move.
What happened to voting and beliefs
Across the three experiments, none of the changes produced significant effects on outcomes such as political polarization, ideological extremity or how people evaluated candidates, the University of Texas at Austin reported. The chronological feed did not significantly alter issue polarization, affective polarization or political knowledge. Removing reshares actually lowered news knowledge among participants.
A separate deactivation experiment asked what happens if you leave altogether. The NBC News report says 35,000 people took a six-week break from Facebook and Instagram before the 2020 election, with results published in PNAS on 13 May 2024. The break had nearly zero effect, at least in the short term, on how people viewed the candidates, whether they voted and how they saw the election’s legitimacy.
The results were not all neutral:
- People who deactivated Facebook were less likely to believe election misinformation, but also less knowledgeable about general news.
- Those who deactivated were slightly more likely to vote for Joe Biden, the equivalent of 1.3% of Trump voters swinging to Biden, which the researchers described as small and unverified.
- On issues such as immigration, mask requirements and policing, deactivators stayed as polarized as before.
Echo chambers and ideological segregation
The same project found that the news people see is ideologically segregated. Many political news links were seen and engaged with mainly by conservatives or mainly by liberals, but not both. Segregation was higher for links posted by Pages and in Groups than for content posted by individual users.
The audience was also lopsided. Of the political news links posted at least 100 times and rated false by Meta’s third-party fact-checkers, 97% were seen by more conservatives than liberals, although false-rated links were a very small share of the total, according to the Center for Social Media and Politics.
Meta read the results differently from many critics. Its announcement said there is little evidence that key features of its platforms alone cause harmful “affective” polarization, and described the segregation as a complex interaction between algorithmic and social factors. Treat that as the company’s interpretation, not a neutral verdict.
What the studies cannot settle
Gentzkow, a Stanford economist and co-author, was blunt about the limits. He said the deactivation study “cannot say one way or the other – in a decadelong sense – whether social media is causing polarization or not.” Stroud and Tucker said they know how influential the algorithm is, but that what they do not know is why attitudes stayed put.
Several caveats apply, as the Tech Policy Press primer lays out:
- Three months may be too short for a robust intervention.
- The US context may not generalise to other countries.
- Meta was closely involved in shaping the research questions and design.
- Participants in the chronological group may still have seen posts shaped by algorithmic ranking from other users.
Ranking systems can also fail in other ways, and our explainer on what is algorithmic bias covers how that happens. Automation is reshaping work as well as politics, a topic we cover in ai impact on jobs.
Frequently asked questions
Does social media make voters more polarized?
Show answer
In these experiments, changing feeds for three months did not significantly change polarization. The researchers say that does not settle the long-term question, because the studies were short and limited to Facebook and Instagram in the US.
Did Meta control the research?
Show answer
Meta could not restrict or censor findings, and the academic leads had final say over writing. Independent observer Michael Wagner nonetheless said Meta set the agenda in ways that affected independence, so the design reflects company choices.
Would a chronological feed fix the problem?
Show answer
Not on this evidence. It cut uncivil posts on Facebook but raised untrustworthy-source content, and it did not significantly change polarization.
Do these findings apply outside the United States?
Show answer
That is unclear. The study authors themselves note that the US context may not generalise to other countries.
