Misinformation is not a new problem, but there are plenty of signs that the advent of social media has made it worse. Academic researchers are trying to understand the scope of the problem, identify the social media networks where misinformation is most prevalent, and examine government coordinated efforts to spread false information, as well as the impact of cyberattacks. Celebrity Who is the source of the misinformation?
These are all potentially valuable data points. But there’s another big contribution missing: ordinary people who, for one reason or another, feel motivated to spread misinformation. A study published today looks at a large panel of Twitter accounts connected to voters living in the US (the study was conducted back when X was still Twitter). The study identifies a small group of misinformation super-spreaders who make up just 0.3% of the accounts, but share 80% of the links to fake news sites.
While one might expect these to be younger, internet-savvy people who will automate sharing, we found that this demographic is actually older, female, and highly likely to click the “retweet” button.
Finding super-spreaders
The study, by Sahar Valibi-Bartov, Briony Swire-Thompson, and Neil Greenberg, relied on a panel of over 650,000 Twitter accounts linked to US voter registrations, using names and location information. These voting records provide information about individuals as well as location information that can be linked to the average demographics of their voting district. All of these users were active on the platform before the 2020 election, but the study was stopped before the post-election surge in misinformation.
The researchers first used machine learning classifiers, pre-validated by human callouts, to identify tweets containing political content posted by these users. They focused on tweets that contained links to news sites. They then matched those links against a list of “news” websites known to spread misinformation about the election.
There are some caveats to this approach. Researchers cannot be sure whether the voters in question had full (or even any) control over their accounts during the election period. Also, the accuracy of the individual articles behind the shared links is not tested. Thus, while these sites may have been consistent sources of misinformation, it is still possible that they published accurate articles that were shared. Still, these are unlikely to be a major consideration given the size of the population studied and the corresponding number of tweets.
From this population, Baribi-Bartov, Swire-Thompson, and Greenberg identified 2,107 accounts that were responsible for 80% of the tweets linking to misinformation sources. They call these misinformation superspreaders. In their analysis, they compare the superspreaders to a random sample of the overall population and to people who most frequently share links to trusted news sources.