Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 70 of the 70 references it cites were found in a registry or an index; 4 of the 20 sources its claims cite were checked for support.

How do social media feed algorithms affect attitudes and behavior in an election campaign?

Open access

Abstract

We investigated the effects of Facebook's and Instagram's feed algorithms during the 2020 US election. We assigned a sample of consenting users to reverse-chronologically-ordered feeds instead of the default algorithms. Moving users out of algorithmic feeds substantially decreased the time they spent on the platforms and their activity. The chronological feed also affected exposure to content: The amount of political and untrustworthy content they saw increased on both platforms, the amount of content classified as uncivil or containing slur words they saw decreased on Facebook, and the amount of content from moderate friends and sources with ideologically mixed audiences they saw increased on Facebook. Despite these substantial changes in users' on-platform experience, the chronological feed did not significantly alter levels of issue polarization, affective polarization, political knowledge, or other key attitudes during the 3-month study period.

Overview

CoChat Summary

TL;DR

A large randomized experiment on Facebook and Instagram during the 2020 US election found that switching users from algorithmic to reverse-chronological feeds substantially changed on-platform experience and reduced time spent and online political engagement. However, the chronological feed did not significantly alter issue polarization, affective polarization, political knowledge, or self-reported offline participation/turnout over the 3-month study.

Key findings

  • Chronological feeds reduced platform use: Facebook users in the Algorithmic Feed group spent 73% more time daily than US monthly active users, versus 37% more in the Chronological Feed group (p < 0.005); Instagram values were 107% versus 84% more (p < 0.005). Likes fell from 6.7% to 3.1% of exposures on Facebook and from 1.4% to 0.9% on Instagram (p < 0.005).
  • Content exposure shifted: political content increased; Facebook untrustworthy-source content rose from 2.6% to 4.4% (+68.8%, p < 0.005), uncivil content fell from 3.2% to 1.8% (−43.0%, p < 0.005), and moderate/mixed-audience content rose from 22.6% to 30.9% (+36.7%, p < 0.005). On Instagram, untrustworthy-source content rose from 1.3% to 1.6% (+22.1%, p < 0.005).
  • Primary political attitudes showed no significant treatment effects: issue polarization, affective polarization, election knowledge, and news knowledge did not differ significantly (p ≈ 1 for polarization; p > 0.63 for knowledge). Self-reported political participation and turnout also showed no significant differences (Facebook p = 1.0 for both; Instagram p = 1.0 and p = 0.64).
  • On-platform political engagement declined strongly (Facebook −0.117 SD, p < 0.005; Instagram −0.090 SD, p < 0.005), while Facebook users in the Chronological Feed clicked more on political news from likely partisan sources (0.107 SD, p < 0.01).

Methods

  • Randomized controlled experiments within Facebook and Instagram during the 2020 US presidential election. Consenting US users aged ≥18 were recruited via survey invitations atop their feeds in August 2020 and randomly assigned to status quo Algorithmic Feed or reverse-chronological Chronological Feed, active 24 September to 23 December 2020. Ads were not affected; about 80% of material presented was manipulated.
  • Analytic samples: Facebook n = 23,391; Instagram n = 21,373, comprising participants who completed the first two surveys and at least one of the subsequent three waves. Participants completed five surveys and consented to on-platform activity and passive off-platform tracking.
  • Preregistered on OSF. Main estimand was the population average treatment effect (PATE), weighted by predicted ideology, friend count, number of political pages followed, and days active; unweighted sample average treatment effects (SATE) were also reported. Effects were estimated for feed content, attitudes, knowledge, and participation, with multiple-comparison adjustment for primary hypotheses.

Limitations

  • The paper states the design estimates direct individual effects only and cannot speak to “general equilibrium” effects: it cannot capture whether algorithms shape societal incentives, content producers, campaigns, or feedback from users outside the sample whose behavior was shaped by algorithmic rankings. Scaling the intervention to all users might produce different results.
  • The paper states possible explanations for null downstream effects include the approximately 3-month intervention being too short, the specific polarized US election context, and the particular alternative ranking system used; results may not generalize to other political systems or ranking systems. Mechanisms involving knowledge, cross-cutting exposure, and coordination/information costs could not be directly tested because mediators were post-treatment.
  • The paper notes the sample was more active than average monthly active users, and the unweighted sample overrepresented users aged 30–44, white, female, liberal, higher-income, and college-educated; PATE weighting was designed to address this. Attrition was 19.5% for posttreatment waves but did not differ significantly by treatment group. Instagram also lacked ideology and political-news classifications for some measures.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1126/science.abp9364

Cites

70 of 70 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 Valid · 0 Not valid · 9 Not checked · 74 Own results and interpretations

57 more claims

Create a free account to see all 86 claims and to verify the 2 references still open.

See all claims — free account
  • Probably not a substantial claim. 10
  • It may state two separate claims as one. 16

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 66 references

Research works printed but not linked: 4

  1. E. Pariser The Filter Bubble: How the New Personalized Web is Changing What We Read and How We Think (Penguin 2011).
  2. P. Barbera Social Media and Democracy: The State of the Field and Prospects for Reform N. Persily J. A. Tucker Eds. (Cambridge Univ. Press 2020) pp. 34–55.
  3. M. J. van der Laan J. M. Robins Unified Methods for Censored Longitudinal Data and Causality (Springer 2012).
  4. A. A. Tsiatis Semiparametric Theory and Missing Data (Springer 2006).

Cited by

See all 363

Metadata

Authors
Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Kennedy, E., Kim, Y. M., Lazer, D., Moehler, D., Nyhan, B., Rivera, C. V., Settle, J., Thomas, D. R., Thorson, E., Tromble, R., Wilkins, A., Wojcieszak, M., Xiong, B., de Jonge, C. K., Franco, A., Mason, W., Stroud, N. J., & Tucker, J. A.
Affiliations
  • Princeton University (US)
  • Stanford University (US)
  • Meta (United States) (US)
  • University of Vienna (AT)
  • University of North Carolina at Chapel Hill (US)
  • Annenberg School for Communication, University of Pennsylvania, Philadelphia, PA, USA.
  • Carnegie Mellon University (US)
  • University of Wisconsin–Madison (US)
  • Northeastern University (US)
  • Dartmouth College (US)
  • William & Mary (US)
  • Syracuse University (US)
  • George Washington University (US)
  • University of California, Davis (US)
  • Meta, Menlo Park, CA, USA.
  • Center for Media Engagement (US)
  • New York University (US)
Corresponding author
Guess, Andrew M.
Venue
Science
Published
2023-07-28
Volume
381
Issue
6656
Pages
398-404
ISSN
  • 0036-8075 (ISSN-L)
  • 1095-9203
OpenAlex
W4385298751
Cited by
363
References
66
Citations by year
2026: 124 · 2025: 113 · 2024: 89 · 2023: 33
FWCI
169.45
Type
journal article
Language
en
Open access
Locations
  • Science (publishedVersion)
  • PubMed (publishedVersion)
  • Carolina Digital Repository (University of North Carolina at Chapel Hill) (submittedVersion)
  • UvA-DARE (University of Amsterdam) (submittedVersion)
  • UNC Libraries
Funders
  • Alfred P. Sloan Foundation
  • American Association for the Advancement of Science
  • William and Flora Hewlett Foundation
  • John Simon Guggenheim Memorial Foundation
  • John S. and James L. Knight Foundation
  • Google
  • University of Wisconsin-Madison
  • Charles Koch Foundation
  • York University
  • Nutrition Obesity Research Center, University of North Carolina
  • Stanford Institute for Economic Policy Research
  • National Opinion Research Center
Topics
  • Social Sciences › Communication › Social Media and Politics
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
  • Social Sciences › Sociology and Political Science › Media Influence and Politics
Keywords
Facebook, Instagram, 2020 US presidential election, political campaigns, political polarization, affective polarization, political knowledge, user engagement, content exposure
MeSH terms
Algorithms, Attitude, Humans, Politics, Friends, Social Media

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: How do social media feed algorithms affect attitudes and behavior in an election campaign? https://app.cochat.ai/papers/10.1126/science.abp9364
The paper
Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Kennedy, E., Kim, Y. M., Lazer, D., Moehler, D., Nyhan, B., Rivera, C. V., Settle, J., Thomas, D. R., . . . Tucker, J. A. (2023). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science. https://doi.org/10.1126/science.abp9364

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.