The Spread of True and False News Online
Science, 2018 · paper · Media studies
Examines how verified true and false news stories spread on Twitter from 2006 to 2017, covering about 126,000 stories. False news diffused farther, faster, deeper and more broadly than true news, especially in politics, and humans rather than bots were more likely to spread it.
Checked: PDF from politics.media.mit.edu, fetched 5 Oct 2026. Page numbers are those of that PDF.
5 findings
Substance Text: "were verified significantly less often (K-S test = 0.004, P < 0.001)" (Fig. 4A), but the Fig. 4A 'verified' row prints "D=0.005, p<0.001".
- p. 4 Main text, user characteristics paragraph
K-S test = 0.004, P < 0.001
- p. 4 Fig. 4A, 'verified' row
D=0.005, p<0.001
- p. 4 Main text, user characteristics paragraph
were verified significantly less often (K-S test = 0.004, P < 0.001)
- p. 4 Results text, user characteristics paragraph
(K-S test = 0.004, P < 0.001)
Only the authors can say what is right here.
Judge's note: Both appear on p. 4 (text crop and Fig. 4A table on the full page). The other four K-S values in the sentence (0.104, 0.136, 0.054, 0.125) match Fig. 4A exactly. For a binary variable, D equals the difference in proportions. The rounded means are 0.002 and 0.006, which allows any true difference between about 0.003 and 0.005, so either printed value is possible. That doesn't settle which one is correct, so only the authors can say.
Second check: another model agreed
Report a problem with this finding
Reference or label The same ~126,000 count is labelled "~126,000 stories" in the abstract, "~126,000 rumor cascades" in the main text, and "About 126,000 rumors" in the editor's summary.
- p. 1 Abstract
The data comprise ~126,000 stories tweeted by
- p. 1 Main text, last paragraph
The data include ~126,000 rumor cascades
- p. 7 Editor's summary
About 126,000 rumors were spread by
Judge's note: The paper itself defines a rumor (story) as made up of one or more cascades, so stories and cascades are different units. The body text and the Fig. 1F axis count cascades. The abstract's unit label conflicts with the paper's own definitions.
Second check: another model agreed
Report a problem with this finding
Reference or label Fig. 4 caption (D) and (F): "across seven dimensions categorized by the NRC", but panels D and F each show eight emotions and the text says "associations with eight emotions".
- p. 4 Fig. 4 caption (D) and (F)
seven dimensions categorized by the NRC
- p. 5 Main text
their associations with eight emotions
- p. 5 Main text, emotion paragraph
associations with eight emotions
- p. 4 Fig. 4 caption, (D) and (F)
across seven dimensions categorized by the NRC
Judge's note: Fig. 4D plots Surprise, Disgust, Fear, Anger, Sadness, Anticipation, Joy, Trust, and Fig. 4F tabulates the same eight rows. On p. 5 the text lists eight Plutchik emotions and refers to 'each of the eight emotions'. The caption's 'seven' appears twice and is a count/label error.
Second check: another model agreed
Report a problem with this finding
Reference or label Ref. 9: "Proceedings of the 22nd International Conference on World Wide Web (ACM, 2010), pp. 729–736." The year conflicts with the edition number.
- p. 5 Reference 9
of the 22nd International Conference on World Wide Web (ACM, 2010), pp. 729–736.
- p. 5 References, item 9
(ACM, 2010), pp. 729–736.
- p. 5 Reference 9
Proceedings of the 22nd International Conference on World Wide Web (ACM, 2010), pp. 729–736.
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the 22nd WWW conference was held in 2013. Internally, the paper's own reference list dates the 24th conference to 2015 (refs 15, 16) and the 25th to 2016 (ref 35), which for an annual series implies 2013 for the 22nd. The p. 5 image prints 2010, the same year as ref. 8 just above it.
Second check: another model agreed
Report a problem with this finding
Wording Abstract: "Whereas false stories inspired fear, disgust, and surprise in replies", but Fig. 4F gives fear 0.108 vs 0.102 with "D=0.021, p~0.164", and the results text (p. 5) names only surprise and disgust as greater for false rumors.
- p. 1 Abstract
false stories inspired fear, disgust, and surprise in replies
- p. 4 Fig. 4F, fear row
D=0.021, p~0.164
- p. 5 Results text on emotions
greater surprise (K-S test = 0.205, P ~ 0.0)
- p. 4 Fig. 4F, fear row
D=0.021, p=0.164
Only the authors can say what is right here.
Judge's note: Checked the abstract crop and the Fig. 4F table and the p. 5 emotion paragraph on the page images. Fig. 4F fear row: means 0.108 (false) vs 0.102 (true), D=0.021, p~0.164. Every other emotion the text discusses has p~0.0. The results paragraph lists surprise (0.205) and disgust (0.102) for false rumors and sadness, anticipation, joy and trust for true ones; fear does not appear. So the abstract's attribution of fear has no support in the paper's own test or in its results text. Only the authors can say whether fear was meant to be listed.
Second check: another model agreed
Report a problem with this finding
6 minor (typos and the like; most checked by the judge only)
Reference or label The Fig. 2H y-axis ("Mean Breadth") ticks run 0.20, 0.40, 0.60, 0.80 and then "10" with equal spacing; the top tick evidently should be 1.0.
- p. 3 Fig. 2H y-axis
Mean Breadth
Only the authors can say what is right here.
Judge's note: Checked the page 3 image. Fig. 3H shows a similar sequence (0.80, 10, 15, 25), which suggests decimal points were lost from the tick labels. Only the authors can confirm the intended values.
Second check: another model agreed
Report a problem with this finding
Reference or label "Falsehood reached more people at every depth of a cascade than the truth ... (Fig. 2C)" cites the maximum-breadth CCDF; the per-depth quantities are in Fig. 2G (unique users at every depth) and 2H (mean breadth at every depth).
- p. 3 Results paragraph 1
Falsehood reached more people at every depth of a cascade than the truth, meaning that many more people retweeted falsehood than they did the truth (Fig. 2C)
- p. 3 Results text
meaning that many more people retweeted false-hood than they did the truth (Fig. 2C)
Judge's note: The Fig. 2 caption on p. 3 defines (C) as 'Maximum breadth', a CCDF over cascades with no breakdown by depth. (G) is 'The number of unique Twitter users reached at every depth' and (H) is 'the mean breadth ... at every depth'. The same column then cites Fig. 2G for 'more unique users than the truth at every cascade depth'. Because maximum breadth is related to people reached per depth, the pointer may be intentional, but the 'at every depth' wording matches 2G/2H, not 2C.
Second check: another model agreed
Report a problem with this finding
Wording The worked example says a tweet "retweeted 100 times" forms a cascade "of size 100", while size is defined as the number of users involved and an un-retweeted tweet has size one.
- p. 1 Introduction, rumor B example
each of those two tweets is retweeted 100 times, the rumor would consist of two cascades, each of size 100
- p. 2 Definition of size
size (the number of users involved in the cascade over time)
Judge's note: Under the paper's own definitions, the originator plus 100 retweeters is 101 users. This is an illustrative example only and does not affect any result.
Second check: another model agreed
Report a problem with this finding
Wording "the top 0.01% of false cascades diffused eight hops deeper into the Twittersphere than the truth, diffusing to depths greater than 19 hops", while p. 3 states "the truth never diffused beyond a depth of 10"; a depth beyond 19 against a maximum of 10 is a gap of at least 9 hops, not 8.
- p. 2 Results, p. 2 to 3
diffused eight hops deeper into the Twittersphere than the truth, diffusing to depths
- p. 3 continuation, Fig. 2A text
greater than 19 hops from the origin tweet
- p. 2 Results, para on depth
diffused eight hops deeper into the
- p. 3 timing paragraph
the truth never diffused beyond a depth of 10
Only the authors can say what is right here.
Judge's note: Read the p. 2 crop ("eight hops deeper"), the p. 3 crop ("greater than 19 hops from the origin tweet") and the p. 3 timing paragraph ("truth never diffused beyond a depth of 10 ... falsehood reached a depth of 19"). Fig. 2A/2E/2G show true curves ending near depth 10 and false curves near 19. The stated figures imply a difference of 9 or more. The 'eight' may come from comparing at a percentile the text does not state. The authors need to say which depths the 'eight hops' compares. I did not rely on the claim about what the published version says, since it cannot be checked here.
Second check: another model agreed
Report a problem with this finding
Wording The text states "we do find that false news is more novel and that novel information is more likely to be retweeted", but no result shown links novelty to retweeting.
- p. 5 Emotion/novelty paragraph
we do find that false news is more novel and that novel information is more likely to be retweeted
- p. 4 Fig. 4B regression variables
account age
Only the authors can say what is right here.
Judge's note: Fig. 4B's retweet model includes no novelty variable, and Fig. 4C/E only compare novelty between true and false rumors. The supporting analysis may be in the supplement, but no supplement is among the files and none is cited at this sentence.
Second check: another model agreed
Report a problem with this finding
Typo "a rumor as the social phenomena" should read "phenomenon".
- p. 1 Introduction, definition of rumor
a rumor as the social phenomena
Judge's note: Checked the crop. The plural noun is used with a singular article.
Second check: none (judge only)