Less Than You Think: Prevalence and Predictors of Fake News Dissemination on Facebook
Science Advances, 2019 · paper · Media studies
Links a survey of 3,500 people to their Facebook sharing records to study who shared fake news during the 2016 U.S. campaign. Sharing was rare; conservatives shared more, and users over 65 shared nearly seven times as many fake news articles as the youngest group.
Checked: PDF from europepmc.org, fetched 5 Oct 2026. Page numbers are those of that PDF.
3 findings
Substance Table 1 prints "8 (0.01%)", "5 (<0.01%)", "9 (0.01%)" and "4 (<0.01%)" for 3, 4, 5–10 and 11–50 shares; of 1191 respondents these counts are about 0.7%, 0.4%, 0.8% and 0.3%. The cells are proportions printed with a percent sign, while the first three cells (91.5%, 5.3%, 1.0%) are true percentages.
- p. 2 Table 1, column '3'
8 (0.01%)
- p. 2 Table 1, column '4'
5 (<0.01%)
- p. 2 Table 1, column '5–10'
9 (0.01%)
- p. 2 Table 1, column '11–50'
4 (<0.01%)
Judge's note: Read Table 1 on the page 2 image: values are as quoted. The seven counts sum to 1191, the linked sample, and the first three percentages are correct for that denominator. The last four are understated by a factor of 100 (0.0067 rounds to 0.01; 0.0042 and 0.0034 are below 0.01). The text's "8.5% of respondents ... shared at least one such article" equals 101/1191, which confirms the denominator and that the small cells should be about 0.3-0.8% each. Nothing in the article explains a different scale for these cells.
Second check: another model agreed
Report a problem with this finding
Reference or label Table 3 row label "% Knowledge (0–2)" gives the scale as 0–2, but every value in the row (2.05, 2.04, 2.13, 2.13) exceeds 2 and the text says "Knowledge ranged from 0 to 4"; the row is also prefixed "%" although the values are mean scores.
- p. 5 Table 3, row label
% Knowledge (0–2)
- p. 5 Table 3, Knowledge row values
2.05
- p. 7 Sample details, first paragraph
Knowledge ranged from 0 to 4
Judge's note: Read Table 3 on the page 5 image and Sample details on the page 7 image. A mean above 2 is impossible on a 0–2 scale, and the text describes four items (House majority, Senate majority, uninsured rate, earnings) with a range of 0 to 4, which fits the values. The range in the row label is wrong, and the "%" prefix does not apply to a mean score. No rescaling is described anywhere in the article.
Second check: another model agreed
Report a problem with this finding
Wording Fig. 2 caption: "Proportions adjusted to account for sample-matching weights", but all four panels plot the mean number of fake news stories shared (y-axis "Mean number of fake news stories shared", values up to 1.0 on an axis to 1.2), not proportions.
- p. 3 Fig. 2 caption
Proportions adjusted to account for sample-matching weights
- p. 3 Fig. 2, y-axis label
Mean number of fake news stories shared
Judge's note: Viewed Fig. 2 and its caption on the page 3 image. The caption title says "Average number of fake news shares (and 95% CIs)" and the y-axis label says mean number; the very-conservative bar is at 1.0 with an upper interval above 1.2. No proportion is plotted in any panel, so "Proportions" should read "Means" or "Averages".
Second check: another model agreed
Report a problem with this finding
10 minor (typos and the like; most checked by the judge only)
Reference or label Table 3 row "% Voted in 2016 general" gives 0.59, 0.59, 0.63, 0.63, which are proportions, under a "%" label, while the other "%" rows in the table are whole-number percentages (31, 32, 40; 36, 37, 47; 23, 20, 22).
- p. 5 Table 3, row '% Voted in 2016 general'
% Voted in 2016 general
- p. 5 Table 3, same row, Sample 3 value
0.63
- p. 5 Table 3, same row
0.59
- p. 5 Table 3, row 'Knowledge'
% Knowledge (0–2)
Judge's note: Read Table 3 on the page 5 image; values are as quoted. The text (page 7) says the table "reports raw proportions", which accounts for the 0.59 format but not for the "%" label or for the other rows being in percentage points. The meaning is recoverable (59% and 63%), so this is a unit label inconsistency. The "%" on the knowledge row is covered under C011.
Second check: another model agreed
Report a problem with this finding
Reference or label Methods say "we report (below) analyses using three supplementary collections of fake news articles" and "Below, we included negative binomial and Ordinary Least Squares (OLS) regressions", but neither appears below in the article; they can only be in the Supplementary Materials (Tables S1-S13).
- p. 7 Additional lists
we report (below) analyses
- p. 7 Modeling strategy and covariates
Below, we included negative binomial and Ordinary Least
- p. 7 Modeling strategy
Below, we included negative binomial and Ordinary Least Squares (OLS) regressions
Judge's note: Read page 7 and page 8 in full. After these sentences the article contains only the domain list, sample details, the supplementary materials list, references and acknowledgments. No analysis with the additional lists and no negative binomial or OLS model appears in the article (Table 2 is quasi-Poisson). Elsewhere the article names 'the Supplementary Materials' explicitly, so 'below' is a leftover pointer. The supplement is not among the files, so I could not confirm which S table holds these analyses. Minor: a reader can guess the destination.
Second check: another model agreed
Report a problem with this finding
Wording Text says "The right panel of Fig. 1 reveals a large spike at 0, with a long tail that goes as far as 50 shares for a single Facebook user", but the right panel's x-axis is ticked 0 to 25 and ends just past 25, with no visible bars beyond about 11.
- p. 1 Results, first paragraph
that goes as far as 50 shares for a single Facebook user
- p. 7 Fig. 1, right panel, x-axis
Number of fake news stories shared
- p. 1 Results, first paragraph
The right panel of Fig. 1 reveals a large spike at 0, with a long tail that goes as far as 50 shares for a single Facebook user
- p. 2 Fig. 1, right panel, x axis
25
Only the authors can say what is right here.
Judge's note: Viewed Fig. 1 on PDF page 2 (the figure is on page 2, not page 7). The right panel's axis runs 0, 5, 10, 15, 20, 25 and stops there; nothing is drawn near 50. Table 1's top bin is 11-50, so a maximum of 50 may be right for the full linked sample, and the caption restricts the panel to Democrats, Republicans and independents, which could exclude the 50-share user. The paper does not say so, and the sentence attributes the 50-share tail to this panel, which does not display it.
Second check: another model agreed
Report a problem with this finding
Wording "310 (26.1%) respondents shared 10 to 100 links ... and 729 (61.3%) respondents shared 100 to 1000 links": with the stated linked sample of 1191 the counts give 26.0% and 61.2%; the printed percentages fit a denominator of about 1189, which the paper does not state.
- p. 1 Results, first paragraph
310 (26.1%)
- p. 1 Results, first paragraph
729 (61.3%)
- p. 6 Facebook profile data, third paragraph
1191 of survey respondents
- p. 1 Results, first paragraph
310 (26.1%) respondents shared 10 to 100 links
Judge's note: Read the sentence on the page 1 image; the population is "respondents for whom we have Facebook profile data", which is 1191 (page 6 text, Table 3 Sample 4, and the Table 1 counts sum to 1191). Both percentages are 0.1 point above what 1191 gives and both are reproduced with 1189. Not a rounding tie (26.03 and 61.21 do not round to the printed values). No statement in the article explains a denominator of 1189; two respondents with missing link data, or weighting, would explain it. Minor.
Second check: another model agreed
Report a problem with this finding
Wording Table 3: the P values printed for "% Democrat" (32 vs 40, P = 0.17) and "% Vote intention (Clinton)" (37 vs 47, P = 0.07) do not correspond to the printed percentages and sample sizes (N = 2711 and 1331) under the declared test ("t tests of the difference in means" between Sample 2 and Sample 3), whereas the other percentage rows do.
- p. 5 Table 3, P column, rows % Democrat and % Vote intention (Clinton)
0.17
- p. 5 Table 3, row % Voted in 2016 general
0.01
Only the authors can say what is right here.
Judge's note: Read Table 3 and footnote ‡ on the page 5 image. For a binary variable the variance follows from the proportion, so the declared t test can be recomputed from the printed values. It reproduces the other rows: voted 0.59 vs 0.63 gives z-squared of about 6.1 (P about 0.01, printed 0.01); high school or less 20 vs 22 gives z-squared of about 2.1 (P about 0.15, printed 0.17). For % Democrat the same calculation gives z-squared of about 24.6 and for vote intention about 36, both far beyond what P = 0.17 or 0.07 requires, and rounding of the percentages cannot close the gap. The same P (0.17) is printed for an 8-point and a 2-point difference. Caveats: Sample 3 is largely nested in Sample 2, item non-response may change the Ns for single rows, and the test details are not given; the text says the table is unweighted ("raw proportions"). The text on page 7 also calls the samples "comparable on ... vote intention". Question for the authors: which values and which test produced the P values in these two rows, and are they aligned with the right rows?
Second check: another model agreed
Report a problem with this finding
Typo Table 2 caption: "the number of Facebook posts by each respondent that includes a link to an external URL"; the verb should be "include" (the posts include the link).
- p. 4 Table 2 caption
the number of Facebook posts by each respondent that includes a link to an external URL
Judge's note: Read the caption on the page 4 image; the text is as quoted. Subject-verb disagreement; cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Two entries in the list of 21 domains end in a stray period: "(14) donaldtrumpnews.co." and "(19) departed.co."; the other 19 entries have no terminal punctuation.
- p. 7 Details on main BuzzFeed-based list, item 14
(14) donaldtrumpnews.co.
- p. 7 Details on main BuzzFeed-based list, item 19
(19) departed.co.
- p. 7 Details on main BuzzFeed-based list, items (14) and (19)
donaldtrumpnews.co.
- p. 7 item (19)
departed.co.
Judge's note: Read the list on the page 7 image. Items 1-13, 15-18, 20 and 21 have no trailing period, including the last item; only 14 and 19 do, and both are the entries ending in '.co'. The period is not list punctuation and reads as part of the domain name.
Second check: none (judge only)
Report a problem with this finding
Typo Competing interests: "participated in other unpaid meetings of academics researchers and Facebook product managers" should read "academic researchers".
- p. 8 Competing interests statement
participated in other unpaid meetings of academics researchers
Judge's note: Read on the page 8 image. The same statement has 'approximately 30 academic researchers' a few lines earlier.
Second check: none (judge only)
Report a problem with this finding
Typo Reference 14: "How the elderly of lisbon use and perceive ICT" has the place name Lisbon in lower case.
- p. 8 Reference 14
How the elderly of lisbon use and perceive ICT
Judge's note: Read on the page 8 image. The reference list uses sentence case but keeps other proper nouns capitalised ("Russian" in ref. 5, "U.S." in ref. 3), so lower-case 'lisbon' is a slip, probably from automatic case conversion.
Second check: none (judge only)
Report a problem with this finding
Typo "facebook" is lower-cased in three reference titles: ref. 7 "outperformed real news on facebook", ref. 23 "on facebook. Science 348, 1130–1132 (2015)" and ref. 26 "biggest fake news hits on facebook from 2016".
- p. 8 Reference 23
on facebook. Science 348, 1130–1132 (2015).
- p. 8 Reference 26
biggest fake news hits on facebook from 2016
- p. 8 Reference 7
outperformed real news on facebook
Judge's note: Read all three entries on the page 8 image. The article body and the acknowledgments write 'Facebook' throughout, and other proper nouns in the list keep their capitals ("Russian", "U.S.", "BuzzFeed"). Same kind of case-conversion slip as 'lisbon' in ref. 14, but a different word and different entries.
Second check: none (judge only)