Thread by @davidlazer

Happy to announce our paper, just out this minute in @sciencemagazine, which examines the prevalence of Fake News on Twitter during the 2016 election
w/ @grinbergnir @Ldfriedl @_kenny_joseph @Briony_Swire science.sciencemag.org/content/363/64…
1/N

Some of our key findings:

(1) Fake news was moderately prevalent during the election. About 5% of election related content people were exposed to was fake news.
2/N

(2) Exposure was highly concentrated among a few people. The large majority of the fake news content was in the feeds of 1% of our sample.
3/N

(3) Sharing was even more concentrated- where the large bulk of fake news came from .1% (!) of Twitter accounts.
4/N

(4) Similar to prior research, we found that fake news was concentrated more on the right than the left, and more consumed by older voters.
5/N

(5) Fake news did not seem to be more viral than nonfake news- contingent on possible exposure, fake news was not more likely to be retweeted. Ideological congruence of content was a key driver of sharing.
6/N

(6) Looking at the coexposure network (domains A and B are linked if many people are exposed to both), we see that there are two clear clusters of news sites, main stream, and fake news, with the large bulk of exposure to the former.
7/N

(7) So, bottom line: fake news on Twitter is more a phenomenon of a seedy neighborhood than systemic, and is driven in significant part by a strategy of flooding the platform by a tiny fraction of accounts. (& this is not even accounting for bots)
8/fin