Allcott, H., Gentzkow, M., & Yu, C. (2019). Trends in the diffusion of misinformation on social media. Research & Politics, 6, 2053168019848554. https://doi.org/10.1177/2053168019848554
Summary
This paper offers an early empirical tracking of how misinformation diffusion on social media changed in the immediate aftermath of the 2016 US election, when Facebook and other platforms began rolling out content-moderation interventions. Using BuzzSumo data on 569 fake news websites and 9,540 verified false-story URLs from January 2015 to July 2018, Allcott, Gentzkow, and Yu measure Facebook engagements and Twitter shares over time. Their central finding is a divergence: interactions with false content rose on both platforms through late 2016, then fell sharply on Facebook while continuing to climb on Twitter — the Facebook-to-Twitter engagement ratio for fake news declined by roughly 60%. Because this divergence appears for fake news but not for comparison content categories, the authors read it as suggestive (though not causally established) evidence that Facebook-specific factors, plausibly its post-election interventions, slowed the relative spread of misinformation.
Key Contributions
- One of the first longitudinal, cross-platform measures of how the scale of social media misinformation evolved around the 2016 election.
- A comparative Facebook-vs-Twitter ratio design that helps net out sample-selection and demand-side confounds affecting both platforms similarly.
- A documented, replicable combined dataset of fake news sites and verified false-story URLs.
- Suggestive evidence relevant to evaluating platform content-moderation interventions.
Methods
The authors assembled a list of 672 fake news sites (569 usable in BuzzSumo) by combining five prior lists and studies, and built three comparison groups from Alexa rankings: major news sites, small news sites, and business/culture sites. Monthly Facebook engagements (shares + comments + reactions) and Twitter shares were gathered via BuzzSumo and averaged by quarter. A URL-level sample of 9,540 verified false stories was constructed by scraping Snopes “false”/“mostly false” claims, extracting keywords, matching in BuzzSumo, and manually screening. Extensive robustness checks varied list-inclusion thresholds, excluded outlier sites, used alternative comparison groups, and substituted Facebook shares for total engagements.
Findings
- Fake news interactions rose on both platforms from early 2015 through just after the 2016 election.
- Post-election, Facebook engagements with fake news fell more than 50%, while Twitter shares kept rising.
- The Facebook-to-Twitter ratio for fake news fell from about 45:1 (stable through late 2016) to roughly 15:1 by mid-2018 — about a 60% decline.
- Comparison categories stayed relatively stable across both platforms, showing no comparable divergence.
- Facebook fake news engagements peaked near 160 million/month in late 2016 and fell to about 60 million/month by the sample’s end (vs. 200–250 million for major news).
- On Twitter, fake news shares stayed in the 3–5 million/month range post-2016 (vs. ~20 million for major news).
- The URL-level analysis of false stories showed a similar post-2016 halving of the Facebook-to-Twitter ratio.
Connections
This paper is a foundational empirical benchmark in the misinformation-measurement literature and is directly extended by Allcott2025-jb, as well as by later work quantifying the prevalence and reach of low-quality news such as Budak2024-ef, Allen2024-av, and Gonzalez-Bailon2024-rq. Its focus on how misinformation exposure shifts over time and across platforms speaks to the recalibration debate about the true scale of the problem, alongside Guess2020-rr, Grinberg2019-ua, and Vosoughi2018-at. The comparative-platform design and reliance on curated fake news lists connect it to methodological discussions in Lazer2018-mm and Bruns2019-nr.