Brazilian politicians rarely publish fake news on their social media accounts, a recently published study shows. Of every 100 publications made by 945 politicians between January 2018 and December 2020 on X (formerly Twitter), Facebook and Instagram, only one (1%) contained misinformation. Posts with fake news, however, generate much more engagement — they have 10 times more likes or shares than factually correct posts.
The data comes from the study “Detecting misinformation: Identifying false news spread by political leaders in the Global South”, published in the Journal of Quantitative Description: Digital Media.
“The result is counterintuitive, because most people think that politicians spread a lot of misinformation,” says one of the work’s authors, Nara Pavão, professor of political science at the Federal University of Pernambuco (UFPE).
“In reality, what they share most is not actually false information, it is biased, hyper-partisan publications,” she continues. Another of the study’s six authors is political scientist Felipe Nunes, from the Federal University of Minas Gerais (UFMG).
And what kind of politician shares false information most often? According to the study, age, gender or geography are not determining factors — but the party coalition is.
Politicians who were part of the alliance that elected Jair Bolsonaro (then in the PSL, now in the PL) in 2018 were 21% more likely to make false posts, and those who were part of the coalition of the defeated candidate, Fernando Haddad (PT), were 10% more likely to make false posts. % more likely, compared to those not aligned with either.
The study evaluated 4.03 million posts from 945 politicians. Only 421, about 1%, contained false information. Among politicians, 15% made at least one publication with fake news.
The data also shows that politicians are very active on social media. They post, on average, six times a day. The study analyzed publications by the president, vice-president, ministers, governors, vice-governors, federal deputies, senators and mayoral candidates in all Brazilian capitals in 2020.
The researchers used a less common method in analyzing misinformation. Most research evaluates domains (posts with links to sites that are considered uninformative or that have a lot of news classified as false by checking agencies).
In Nara’s study, the content of the text, video or image was considered. According to her, the domain method is not suitable for the Brazilian context — 75% of posts made by politicians had no link or reference to any website or news.
“In Brazil, the majority of posts are texts, images or videos that later circulate in WhatsApp and Telegram groups, and not publications with links”, says the professor.
According to the researchers, the domain method ends up overestimating the amount of fake news in circulation. “This approach ends up classifying as false several publications that are not untrue, are biased or hyperpartisan,” says the political scientist. “This does not mean that they are better, it is also a problem of quality of information, but they are not fake news.”
According to her, the method overestimates the occurrence of fake news because linked sites are often uninformative, but not necessarily everything on them is false, there is a lot of biased content.
An example cited in the study is from deputy Filipe Barros (PL-PR), who published a video with the caption “Haddad the father of the gay kit and the crack bag”. This publication is uninformative, since neither of those things exist.
Another publication classified as fake news using the domain method was, in reality, only partisan. The same deputy posted a news item published on the website O Antagonista stating that a judge had barred the visit of former president Dilma Rousseff and deputies to President Lula (PT), then imprisoned in Curitiba.
According to Nara, one hypothesis is that politicians often do not cross the line of factually incorrect due to the cost that this may have — posts removed from the social network and negative repercussions.
To determine which posts were false, the researchers used artificial intelligence to identify content similar to publications considered false by the six main checking agencies (Rumors, E-farsas, Fato ou Fake, Lupa, Aos Fatos and Projeto Comprova). Afterwards, the materials underwent human review, a more laborious and time-consuming method than that which analyzes domains.
The research was funded by Meta and Fapemig.