Allen, J., Howland, B., Mobius, M. M., Rothschild, D. M., & Watts, D. (2020). Evaluating the fake news problem at the scale of the information ecosystem. Sci. Adv., 6, eaay3539. https://doi.org/10.2139/ssrn.3502581
Summary
This paper reframes the “fake news problem” by measuring the prevalence of fake news relative to all media consumption, rather than in isolation on social media. Using a nationally representative multimode dataset spanning television, desktop, and mobile consumption in the U.S. from 2016 to 2018, the authors show that news of any kind is a small slice of Americans’ media diet, that television dominates news consumption over online sources, and that overt fake news accounts for only about 0.15% of the daily media diet — roughly one-tenth of one percent. The central argument is that placing fake news in the denominator of total consumption reveals it to be a negligible share, and that the roots of misinformedness and polarization more plausibly lie in ordinary news bias and news avoidance than in deliberate online fakery. The paper further notes a striking mismatch between the volume of academic attention devoted to online fake news and its empirical scale.
Key Contributions
- One of the first estimates of fake news prevalence at the scale of the entire information ecosystem, spanning TV, desktop, and mobile.
- Integration of three siloed data sources (Nielsen TV, Comscore online, Nielsen web panel) to address representativeness and comprehensiveness gaps in prior work.
- Analytical reframing that places fake news in the denominator of total media consumption, exposing its tiny relative share.
- A critique of the research literature’s disproportionate focus on social-media fakery relative to its actual footprint, urging attention to ordinary news bias and news avoidance.
Methods
- Assembled a multimode dataset categorizing content by mode of consumption rather than production.
- TV: Nielsen’s nationally representative panel (~100,000, plus a ~50,000 local-market subset), with news defined across ~400 Nielsen-classified programs (hard news, magazine, morning shows, entertainment news, late-night comedy).
- Online: Comscore’s desktop and mobile panel; online news defined via 800+ hard-news sites (adapted from Athey et al.) and fake news defined via 98 researcher/fact-checker-identified sites.
- Passive exposure: Nielsen’s desktop web panel (declining from 90,000 to 60,000) using referral URLs to impute news exposure from Facebook, YouTube, Twitter, Reddit, and search engines; 360,000 YouTube “news and politics” videos sampled.
- A ~15,000-person overlap subset (present in both Nielsen web and TV panels) linked desktop and TV consumption. Robustness tested with stricter definitions of news and fake news, including hyperpartisan and fraudulent categories.
Findings
- Americans consume ~460 minutes of media per day, ~86% of which is not news; news is at most 14.2% of the media diet.
- Online news (including passive social/search/portal exposure) is only 4.2% of total online consumption; TV news is ~23% of TV consumption.
- The TV-to-online news ratio exceeds 5:1 (54 vs 9.7 minutes), ranging from ~2:1 among 18–24 year-olds to over 7:1 among those 55+.
- 44% of the overlap panel consumed no online news at all; nearly three-quarters spent under 30 seconds/day reading news online.
- No age group spent more than ~1 minute/day on fake news or more than ~1% of news consumption on it; older, more conservative users consumed more.
- Only ~1.97% of desktop panelists consumed more fake than mainstream news, dropping to 0.7% among those reading at least a minute of fake news daily.
- Local news dominates TV news for most age groups; late-night comedy is under 5% overall.
- Google Scholar shows 2,210 “fake news” titled publications since 2017 (vs. 73 before), vastly outstripping attention to the far larger TV news channel.
Connections
This paper is a companion to broader consumption-scale work on the same team’s agenda, closely echoing Allen2021-ai’s finding that fake news is a small fraction of exposure, and complements Allen2024-av on the aggregate effects of misinformation. It directly engages the social-media-centric prevalence estimates it aims to contextualize — including Allcott2017-yz, Grinberg2019-ua, Guess2019-ym, Guess2020-rr, and Guess2021-ym — and its ecosystem framing aligns with exposure and diet studies such as Gonzalez-Bailon2024-rq, Budak2024-ef, and Eady2023-xg, while its emphasis on ordinary news bias and avoidance over engineered falsehood connects to critiques like Nyhan2023-gb and the definitional debates in Lazer2018-mm.