Guess, A., Aslett, K., Tucker, J., Bonneau, R., & Nagler, J. (2021). Cracking open the news feed. Journal of Quantitative Description: Digital Media, 1. https://doi.org/10.51685/jqd.2021.006
Summary
This paper offers a large-scale descriptive portrait of how U.S. Facebook users encounter and share news, drawing on the Social Science One “Condor” dataset of URL-level engagement data covering millions of publicly shared links. Uniquely, the dataset lets the authors distinguish views (exposure) from shares, and combine platform-scale counts with supervised classifiers to categorize content as credible vs. low-credibility, political vs. non-political, and clickbait vs. non-clickbait. The central argument is descriptive rather than causal: low-credibility news is comparatively rare relative to credible news, but its circulation is far from trivial and is markedly concentrated among older and very conservative users, with clear signs of a preference for ideologically congenial misinformation. The paper also functions as a methodological proof of concept for working with differentially-private, aggregated platform data.
Key Contributions
- First analysis of newly released, differentially-private Facebook engagement data that measures both exposure (views) and sharing of news, allowing the two behaviors to be separated.
- Fine-grained descriptive disaggregation of news into credible/low-credibility, political, and clickbait categories across age and ideological groups.
- A methodological template combining platform-scale URL data with supervised classifiers and misclassification bias correction (Hopkins–King plus bootstrap).
- An empirical descriptive foundation for subsequent research and policy debate on misinformation exposure and consumption.
Methods
The authors analyze 466,591 URLs first posted in 2018 by U.S. users from the Condor dataset (URLs shared publicly more than 100 times, aggregated by URL-year-month-age-gender-political-page-affinity with Gaussian differential-privacy noise). They construct three binary URL-level measures: credibility (NewsGuard, threshold 60), political vs. non-political (a random forest trained on ~8,552 labeled headlines, ~90% accuracy, F1=0.90), and clickbait (a pre-trained SVM). Classifier misclassification is corrected via the Hopkins–King method with 100-sample bootstrap resampling to produce corrected proportions and confidence intervals. Ideological slant of sources comes from media partisanship scores and manual coding; user ideology comes from Facebook’s five-point political page-affinity measure. An updated version corrects for inadvertent Facebook filtering of users lacking page-affinity scores, which shifted estimates slightly without changing the patterns.
Findings
- Roughly 84% of news shares and 89% of views came from credible domains; about 15% of shares and views were from low-quality domains — roughly one in eight views of at least moderately popular news.
- 27% of news URLs shared by very conservative users were low-quality vs. 9% for very liberal users; for views, 19% vs. 7% — evidence of ideologically congenial misinformation.
- 20% of URLs shared by users 65+ were low-quality vs. 11% for the 24–35 bracket; the age gradient is steepest within the two most conservative groups.
- Older users don’t view much more low-credibility news in absolute counts, but it forms a larger share of their news (18% vs. 8% for the second-youngest), suggesting sharing differences aren’t merely a function of feed exposure.
- Low-quality content is disproportionately political and clickbait: 53.4% of low-quality URLs were political vs. 31.3% of credible; 12.3% were both political and clickbait vs. 6.3% of credible.
- No support that older users share more clickbait (26% for 25–34 vs. 24% for 65+), but strong support that they share more political news (22% vs. 56%).
- Of ~280 billion U.S. URL views in 2018, over 44% were news domains; of ~2.1 billion shares, over 48% were news.
Connections
This paper sits alongside foundational descriptive work on the prevalence and skewed distribution of fake news exposure, connecting to Guess2020-rr, Guess2019-ym, and Grinberg2019-ua on the concentration of misinformation among older and conservative users, and to Allen2020-nj on how small misinformation is relative to the overall news diet. Its emphasis on ideologically congenial sharing links to Osmundsen2021-et and Vosoughi2018-at, while its reliance on privacy-preserving platform data speaks to ongoing debates on platform-data-access-governance represented here by Gonzalez-Bailon2024-rq and Freelon2018-ao. Its exposure-versus-behavior framing also resonates with recalibration work in Budak2024-ef and Allen2024-av.