Social Media and Fake News in the 2016 Election
2017 · paper · Media studies
Discusses the economics of fake news and presents data on its consumption before the 2016 U.S. election. Social media was important but not dominant; pro-Trump false stories were shared more than pro-Clinton ones; the average adult saw few fake stories, and people believed ones favoring their candidate.
Checked: PDF from nber.org, fetched 5 Oct 2026. Page numbers are those of that PDF.
8 findings
Substance Section 3.2 says the 30 headlines are "six from each of five categories", but then describes the second category as "the four most recent pre-election headlines from each of Snopes and PolitiFact" (8) and the fourth as "the two most recent pre-election headlines from each of Snopes and PolitiFact" (4); the introduction's "14 major pre-election fake news headlines" fits 6 + 8, not 6 + 6.
- p. 11 Section 3.2, third paragraph
six from each of five categories
- p. 11 Section 3.2, second category
The second category contains the four most
- p. 11 Section 3.2, fourth category
contains the two most recent pre-election headlines from each of Snopes and PolitiFact
- p. 4 Introduction
recalled seeing each of 14
Only the authors can say what is right here.
Judge's note: Read page 11 and page 4 images. Itemised sizes are 6, 8, 6, 4 and 6 (three placebo headlines in two versions), which still total 30 and give 14 Fake headlines as the introduction says, so "six from each of five categories" is the statement that does not fit. Section 6 (page 16 text) repeats the equal-category design: "three headlines randomly selected from each of the five categories". No statement in the article reconciles the two; the survey instrument and online appendix are not among the files. Only the authors can say whether the categories were equal (three per site and one per site) or unequal.
Second check: another model agreed
Report a problem with this finding
Substance Section 6 says each respondent saw "three headlines randomly selected from each of the five categories" (and Section 3.2 says the list has "six from each of five categories"), but Table 1 reports N = 12,080 false and 6,040 true observations, i.e. 10 false and 5 true headlines per respondent for 1208 respondents, not 9 and 6.
- p. 16 Section 6, third benchmark
three headlines randomly
- p. 28 Table 1, row N, columns 1 and 2
12,080
- p. 28 Table 1, row N, column 2
6,040
- p. 11 Section 3.2
an online survey of 1208 US adults
Only the authors can say what is right here.
Judge's note: Viewed Table 1 (PDF p.28), Section 3.2 (p.11) and Section 6 (p.16). Three per category with three false categories (Big Fake, Small Fake, Placebo) and two true ones gives 9 false and 6 true per respondent: 10,872 and 7,248. Table 1 prints 12,080 and 6,040, exact multiples of 1208 (10 and 5). Searched for an explanation: Section 3.2 itself describes Small Fake as four headlines from each of Snopes and PolitiFact (8) and Small True as two from each (4), so the category sizes are 6, 8, 6, 4, 6 (sum 30), and half of each gives 3, 4, 3, 2, 3 = 10 false and 5 true, which matches the table. So the table agrees with the detailed category descriptions, and the statements "six from each of five categories" and "three headlines ... from each of the five categories" are what do not fit. The paper does not reconcile them; the authors should say which is right. The 'equal balance of true versus false' point is handled in C025.
Second check: another model agreed
Report a problem with this finding
Substance Section 4 and the Figure 3 notes give "the top 690 US news sites" / "top 690 US news websites", but Section 6 says "the 3 billion impressions on the 665 top news websites over the same period".
- p. 13 Section 4, paragraph on Figure 3
referral sources for the top 690 US news sites
- p. 32 Figure 3 notes
for the top 690 US news websites and for 65
- p. 15 Section 6, second benchmark
on the 665 top news websites over the same period
- p. 32 Figure 3 notes
the top 690 US news websites and for 65
Only the authors can say what is right here.
Judge's note: Read page 13, page 15 and the Figure 3 notes on page 32. Section 6 refers to "the 65 websites in the bottom part of figure 3" and then to "the 665 top news websites over the same period", so it points at the same Alexa comparison set that is counted as 690 elsewhere. Searched the text for 665 and 690: no statement of a subset or exclusion. Which count is right is for the authors.
Second check: another model agreed
Report a problem with this finding
Reference or label Text cites "Martin and Yurukoglu 2014" but the only reference-list entry is "Martin, Gregory J., and Ali Yurukoglu. 2016. Bias in cable news: persuasion and polarization. NBER Working Paper 20798."
- p. 5 Introduction, literature paragraph
Martin and Yurukoglu 2014
- p. 25 References
Martin, Gregory J., and Ali Yurukoglu. 2016. Bias in cable news
- p. 25 References
Martin, Gregory J., and Ali Yurukoglu. 2016. Bias in cable news: persuasion and polarization.
- p. 25 reference list
Martin, Gregory J., and Ali Yurukoglu. 2016.
Judge's note: Page 5 image shows the citation year 2014; page 25 image shows the single Martin and Yurukoglu entry dated 2016. The mismatch is internal to the paper and needs no outside fact. Which year is intended is for the authors (the reviewers' claim that the working paper dates from 2014 was not verified here).
Second check: another model agreed
Report a problem with this finding
Reference or label Table 1 notes say "This table presents estimates of equation (1)" and Table 2 notes say "This table presents estimates of equation (2)", but the two displayed regression equations in Section 7 carry no numbers.
- p. 28 Table 1 notes
estimates of equation (1)
- p. 29 Table 2 notes
estimates of equation (2)
- p. 18 Section 7, displayed equation
in a linear regression:
- p. 28 Table 1 notes
This table presents estimates of equation (1)
Judge's note: Viewed the page images for 18, 19, 28 and 29. The equations for C_ia (page 18) and B_ia (page 19) are displayed without equation numbers, and a search of the text found no "(1)" or "(2)" equation label anywhere other than table column headers. The notes point to labels that do not exist; a reader can infer the target from the order, so the effect is small.
Second check: another model agreed
Report a problem with this finding
Reference or label "Column 3 then allows β to vary by the same vector of Xi variables as reported in Table 2, except excluding Di to avoid collinearity": the sentence describes Table 2 column 3 itself; the Xi vector including the Democrat indicator is reported in Table 1.
- p. 20 Section 7, paragraph on Table 2
variables as reported in Table 2, except excluding
Judge's note: Viewed p.20, Table 1 (p.28) and Table 2 (p.29). Table 2 column 3 lists Aligned interacted with Republican, media time, social media variables, education, undecided and age: the Table 1 regressors minus Democrat. 'Same ... as reported in Table 2' is circular; Table 1 is evidently meant.
Second check: another model agreed
Report a problem with this finding
Reference or label Figure 2 notes cite "American National Election Studies (2012)", but the only ANES entry in the reference list is "American National Election Studies. 2010. Times Series Cumulative Data File [dataset]."
- p. 31 Figure 2 notes
American National Election Studies (2012)
- p. 23 References
American National Election Studies. 2010. Times Series Cumulative Data File
- p. 31 Figure 2 notes
using data from the American National Election Studies (2012)
- p. 23 American National Election Studies entry
Times Series Cumulative Data File
Only the authors can say what is right here.
Judge's note: Viewed the page 31 and page 23 images. The figure note uses the author-year form with 2012; the reference list has a single ANES entry dated 2010, and no other ANES citation appears in the text. Panel B's x-axis runs to 2012, so one of the two years is likely wrong, but only the authors can say which release was used. The related point that the entry prints "Times Series" where the URL has "timeseries" is a cosmetic slip in the same entry and rests partly on outside knowledge of the product name.
Second check: another model agreed
Report a problem with this finding
Wording Introduction summary reads "the average US adult might have seen perhaps one or several news stories in the months before the election"; "fake" is missing before "news stories".
- p. 4 Introduction, 'Third' paragraph, last sentence
average US adult might have seen perhaps one or several news stories in the months before the
- p. 17 Section 6
election-period fake news exposure was on the order of one
Judge's note: Page 4 image shows the sentence as quoted, following "saw and remembered 1.14 fake stories". The abstract ("one or perhaps several fake news stories") and Section 6 ("election-period fake news exposure was on the order of one or perhaps several articles") include "fake". As printed the sentence is literally about news stories of any kind.
Second check: another model agreed
Report a problem with this finding
21 minor (typos and the like; most checked by the judge only)
Wording Introduction says "The upper end of previously reported statistics for the ratio of page visits to shares" yields 760 million; Section 6 reports prior ratios of 14, 14 and 3 to 8 and says "we consider a ratio of 20 page visits per share as an upper bound on the plausible range".
- p. 4 Introduction, 'Third' paragraph
The upper end of previously reported statistics for the ratio of page visits to shares of stories on
- p. 15 Section 6, first benchmark
we consider a ratio of 20 page visits per share as an upper bound on the
- p. 4 Introduction, 'Third' paragraph
The upper end of previously reported statistics for the ratio of page visits to shares
- p. 15 Section 6, 'First' paragraph
we consider a ratio of 20 page visits per share as an upper bound
Judge's note: Read both passages on the page 4 and page 15 images. 760 million corresponds to a ratio of 20, which Section 6 presents as the authors' own upper bound chosen from "very rough reference points" whose highest reported value is 14. The introduction's wording attributes the 20 to previously reported statistics. A loose summary rather than a wrong number; the 760 million and "about three" are consistent with Section 6.
Second check: another model agreed
Report a problem with this finding
Wording Introduction says "Democrats and Republicans are both about 15 percent more likely to believe ideologically aligned headlines"; Section 7 reports "17.2 and 14.7 percentage points more likely".
- p. 5 Introduction, 'Fourth' paragraph
both about 15 percent more
- p. 20 Section 7, Table 2 results
17.2 and 14.7 percentage points
- p. 5 Introduction, top of page
Democrats and Republicans are both about 15 percent more
- p. 20 Section 7
17.2 and 14.7 percentage points more likely
Judge's note: Read on the page 5 and page 20 images. The introduction uses "percent" for what the results section gives as percentage points, and rounds 17.2 and 14.7 both to "about 15". Loose summary wording; the meaning is recoverable from Section 7.
Second check: another model agreed
Report a problem with this finding
Wording "Since headlines are randomly assigned to respondents, with equal balance of true versus false and pro-Trump versus pro-Clinton", but Table 1 has 12,080 false-headline and 6,040 true-headline observations, and p.18 says "our survey included a large proportion of fake articles".
- p. 19 Section 7, last paragraph of page
with equal balance of true versus false and pro-Trump versus
- p. 18 Section 7, discussion of column 5
Given that our survey included a large proportion of fake articles
- p. 28 Table 1, row N, columns 1 and 2
12,080
- p. 16 Section 6
three headlines randomly selected from each of the five categories
Only the authors can say what is right here.
Judge's note: Viewed p.19, p.18, Table 1 (p.28) and Section 3.2 (p.11). False headlines outnumber true ones 2 to 1 in Table 1, and three of the five categories are false. Section 3.2 does state an equal pro-Clinton/pro-Trump split within each category, so that half of the sentence holds. 'Equal balance of true versus false' may be meant as 'the true/false mix is the same for pro-Clinton and pro-Trump headlines', but as written it reads as equal numbers, which the table does not show. The second reviewer note (N against three per category) is the same point as C018 and is dealt with there.
Second check: another model agreed
Report a problem with this finding
Wording The same Alexa traffic measure is called "visits" in the Figure 3 title ("Share of Visits to US News Websites by Source") and the introduction ("159 million visits"), and "impressions" in the Figure 3 notes and Section 6 ("159 million impressions on the 65 websites").
- p. 32 Figure 3 title and notes
Share of Visits to US News Websites by Source
- p. 15 Section 6
159 million impressions on the 65 websites
Judge's note: Viewed the page 32 image and searched the text. The figure title and the weighting sentence use "visits", the notes define the categories in terms of "impressions", and the 159 million figure is given as visits on page 3 and as impressions in Section 6. No definition distinguishing the two terms was found; the text also has "page visits or impressions". The authors appear to use them as synonyms, so this is a terminology inconsistency and not a data error.
Second check: another model agreed
Report a problem with this finding
Typo Abstract page prints "A online appendix is available at http://www.nber.org/data-appendix/w23089"; should read "An online appendix".
- p. 2 Abstract page, last line
A online appendix is available at
Judge's note: Read on the page 2 image; the last line begins "A online appendix". Plain article slip on the NBER abstract page.
Second check: none (judge only)
Report a problem with this finding
Typo Introduction reads "we discuss the importance of social media relative to sources of political news and information"; a word such as "other" appears to be missing before "sources".
- p. 4 Introduction, 'First' paragraph
importance of social media relative to sources of political news
- p. 4 Introduction, 'First' paragraph
the importance of social media relative to sources of political news and in-
Judge's note: Page 4 image shows the sentence as quoted. Social media is itself a source of political news, so the comparison does not parse without "other". Cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Section 3.1 reads "4 of the 13 articles in the PolitiFact appear in Snopes"; "list" is missing after "PolitiFact".
- p. 10 Section 3.1, second paragraph
4 of the 13 articles in the PolitiFact appear in Snopes
Judge's note: Page 10 image shows the sentence as quoted, directly after "12 of the 21 articles in the BuzzFeed list appear in Snopes". Missing word.
Second check: none (judge only)
Report a problem with this finding
Typo Opening quotation marks are set as closing marks in "”Big Fake” and ”Small Fake,” respectively, or collectively as ”Fake.”"
- p. 11 Section 3.2, third paragraph
”Big Fake” and ”Small Fake,” respectively
- p. 11 second category paragraph
”Big Fake” and ”Small Fake,” respectively, or collectively as ”Fake.”
Judge's note: Visible in the crop and the page 11 image: the three labels open with a closing-style mark, while “Big True”, “Small True” and “Placebo” on the same page open correctly. Not an extraction artefact.
Second check: none (judge only)
Report a problem with this finding
Typo "SurveyMonkey’s Audience Panel" in the first paragraph of Section 3.2 becomes "surveys on the Audience panel" in the second.
- p. 11 Section 3.2 first paragraph
SurveyMonkey’s Audience Panel
- p. 11 Section 3.2 second paragraph
surveys on the Audience panel
Judge's note: Both forms are on the page 11 image. Capitalisation of the same named panel differs within one page; trivial.
Second check: none (judge only)
Report a problem with this finding
Typo The survey question is quoted as "Do you recall seeing this reported or discussed before the election?" in the Figure 5 notes but as "Do you recall seeing this reported or discussed prior to the election?" in Section 3.2 and Section 6.
- p. 33 Figure 5 notes
Do you recall seeing this reported or discussed before the election?
- p. 11 Section 3.2
Do you recall seeing this reported or discussed prior to the election?
Judge's note: Viewed the page 33 image and the crop of page 11. Both are inside quotation marks as the verbatim question, and the wording differs ("before" versus "prior to"); the text uses "prior to" twice. The meaning is unchanged. The survey instrument is not among the files, so which wording is the original cannot be checked.
Second check: none (judge only)
Report a problem with this finding
Typo Text cites "Lewandowsky, Oberauer, and Gignac (2013)"; the reference list has "Lewandowsky, Stephan, Gilles E. Gignac, and Klaus Oberauer. 2013." The entry also ends "PloS One 8(10):" with no article number.
- p. 17 Section 7, first paragraph
Lewandowsky, Oberauer, and Gignac (2013)
- p. 25 References
Lewandowsky, Stephan, Gilles E. Gignac, and Klaus Oberauer. 2013.
- p. 25 References
Lewandowsky, Stephan, Gilles E. Gignac, and Klaus Oberauer. 2013
Only the authors can say what is right here.
Judge's note: Viewed p.17 and the reference-list crop (p.25). Author order differs between citation and entry, and the entry is cut off after the issue number. Both are visible in the PDF. Which order is right, or whether a different 2013 paper was meant, cannot be settled from the files.
Second check: none (judge only)
Report a problem with this finding
Typo "1.2 percentage points more likely ... and the 95 percent confidence interval allows us to exclude differences greater than 2.9 percent": the bound on a percentage-point difference is given in "percent".
- p. 17 Section 6, survey estimate
exclude differences greater than 2.9 percent
Judge's note: Viewed p.17. Same sentence gives the estimate in percentage points and its bound in percent. Meaning is clear from context; a unit-wording slip.
Second check: none (judge only)
Report a problem with this finding
Typo "We construct a variable Cia, that takes value 1 if survey respondent i correctly identifies ...": stray comma before restrictive "that".
- p. 17 Section 7, first paragraph
We construct a variable Cia, that takes value 1
Judge's note: Viewed p.17; the comma after the subscripted variable is in the page image, not an extraction artefact. Trivial punctuation.
Second check: none (judge only)
Report a problem with this finding
Typo "how likely Democrats or Republicans are as a group are to believe all stories": duplicated "are".
- p. 19 Section 7, paragraph after the second equation
how likely Democrats or Republicans are as a group are to believe all stories
Judge's note: Viewed p.19; the doubled verb is in the page image.
Second check: none (judge only)
Report a problem with this finding
Typo Reference: "American National Election Studies. 2010. Times Series Cumulative Data File [dataset]." The URL in the same entry reads "anes_timeseries_cdf".
- p. 23 References, American National Election Studies
Times Series Cumulative Data File
Judge's note: Viewed the crop of the reference list (p.23). 'Times Series' is printed; the entry's own URL gives 'timeseries', so 'Time Series' is the evident reading.
Second check: none (judge only)
Report a problem with this finding
Typo Reference (Gentzkow, Shapiro, and Stone 2016): "edited by Simon Anderson, Joel Waldofgel, and David Stromberg."
- p. 24 References, Gentzkow, Shapiro, and Stone 2016
Joel Waldofgel
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the Handbook of Media Economics editor's surname is spelled Waldfogel. The page prints 'Waldofgel' (viewed crop of p.24); the name appears nowhere else in the paper to check against.
Second check: none (judge only)
Report a problem with this finding
Typo Reference entry reads "Mosseri Adam. 2016." with no comma after the surname; every other personal-author entry is "Surname, Given name".
- p. 25 References, Mosseri 2016
Mosseri Adam. 2016.
Judge's note: Viewed the page 25 image. The entry is printed "Mosseri Adam. 2016. News Feed FYI: ..."; neighbouring entries ("McClosky, Herbert", "Mullainathan, Sendhil") have the comma. Cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Lewandowsky, Gignac and Oberauer 2013 entry ends "PloS One 8(10):" with a colon and no page or article number.
- p. 25 Lewandowsky et al. 2013 entry
PloS One 8(10):
Judge's note: Viewed the page 25 image. The entry stops after the colon, while other journal entries give a page range after the colon, so the entry is visibly incomplete. The capitalisation "PloS" versus the journal's own styling is a side point that rests on outside knowledge and is not needed for the finding.
Second check: none (judge only)
Report a problem with this finding
Typo Subramanian 2017 entry opens a quotation mark that is never closed: "“Inside the Macedonian Fake-News Complex, Wired, February 15."
- p. 26 References, Subramanian 2017
“Inside the Macedonian Fake-News Complex, Wired, February 15.
- p. 26 Subramanian 2017 entry
“Inside the Macedonian Fake-News Complex, Wired
Judge's note: Viewed the page 26 image. The title has an opening double quotation mark and no closing one, and no other entry on the page puts its title in quotation marks.
Second check: none (judge only)
Report a problem with this finding
Typo Zhao et al. 2013 entry prints "ariXiv:1305.4018"; the repository name is misspelt (arXiv).
- p. 27 References, Zhao et al. 2013
ariXiv:1305.4018
- p. 27
A Peep on the Interplays between Online Video Websites and Online Social Networks. ariXiv:1305.4018.
Judge's note: Viewed the crop of page 27. The text reads "ariXiv:1305.4018." with an extra "i". Plain misspelling.
Second check: none (judge only)
Report a problem with this finding
Typo Figure 5 title reads "Percent of US Adult Population that Recall Seeing or Believed Election News", mixing present "Recall" with past "Believed"; the legend has "Recalled seeing" and "Recalled seeing and believed".
- p. 33 Figure 5 title
Percent of US Adult Population that Recall Seeing or Believed Election News
Judge's note: Viewed the page 33 image. The title's verbs are not parallel, and the legend and notes describe the second series as recalled seeing and believed, where the title has "or". Wording only; the figure is readable as it stands.
Second check: none (judge only)