This content originally appeared on COSSPP’s Wicked Problems, Wicked Solutions Blog and is the work of the individual authors sharing their research, expertise, and experience and do not necessarily reflect the opinions or positions of the College of Social Sciences and Public Policy, Florida State University, or any other agency, institution, or entity.
This article originally appeared in the Washington Post.
There’s no shortage of hand-wringing over Elon Musk’s acquisition of Twitter.Journalists,progressivesand theWorld Health Organizationhave sketched out disinformation dystopias that could emerge once Musk takes Twitter’s reins. One tool that Twitter uses to stop the spread of disinformation is banning users who tweet misleading information. Many fear that if Musk removes this policy, Twitter will become more toxic.
But all this anxiety assumes that Twitter was effectively managing disinformation to begin with. With my research team — Kyle Rose, Sarah Warren and Rob Lytle — I find that disinformation continues to spread across the platform. Musk’s takeover is not the only problem that Twitter has in managing its information reliability.
Twitter bans don’t improve online discussion
In itsRules and Policies, Twitter writes that the platform is intended to facilitate human connection and to help users find reliable information. Twitter warns that it may permanently suspend an account “that misleads others and/or disrupts their experience.”
The policy assumes that bans affect conversation and information and that removing influential accounts spreading disinformation must improve the quality of information and discussion. Someresearchdoes bear this out. But our study analyzes the influence of bans on one political issue and finds otherwise.
We did a deep dive into tweets about Arizona’sindependent auditof Maricopa County ballots cast in the 2020 presidential election, backed by pro-Trump politicians who argued that Democrat Joe Biden had not won the state. On July 27, 2021, Twitter permanently suspended eighthigh-profile accountsthat were associated with the audit and were known to spread falsehoods about the presidential election. What happened?
How we did our research
We usedDiscoverTextand academic credentials provided by Twitter to gather a sample of 245,020 tweets that included the words “Arizona audit” or the hashtag “#Arizonaaudit” and were sent between July 17 and Aug. 5, 2021, several days before and after Twitter banned the accounts. These dates seemed sufficient to assess the conversation beforehand and whether the ban changed it. We were particularly interested in the most influential accounts before, during and after the ban.
We started by analyzing the 148 accounts of Twitter users whose tweets were retweeted 200 times or more in the total sample. These 148 accounts represent 55.4 percent of the total 245,020 relevant tweets, meaning that these 148 tweets about the audit, which didn’t include the banned accounts, were shared 135,797 times.
We used a technique called qualitative content analysis to detect categories and themes in data, carefully examining the 148 accounts for dozens of pieces of information so that we could systematically see what kinds of accounts were most retweeted, whether they supported or opposed the audit and why, whether they shared additional content with their tweet, and, if they shared a news story, the quality of information they shared. If a news story was shared, we categorized it usingAd Fontes Media Bias Chart, which rates sources according to news value, reliability and bias. The chart is useful because the company hasa robust systemfor coding articles and works to mitigate bias within its coding and sample. We also noted whether these accounts shared pictures, memes, images, quotes or other materials with their post.
The ban boosted the retweeting of right-wing personalities, outlets and content
Before Twitter suspended the Arizona audit accounts, the accounts retweeted most often weremainstreamandleft-leaningjournalists, who were largely critical of the audit. After the ban, we found fewer tweets citing sources opposing the audit; right-wing personalities peddling conspiracy theories and partisan right outlets dominated.
To better understand this shift, we analyzed the 10 accounts with the most retweets on each of the days included in the sample, excluding the day of the ban. Before the ban, right-leaning personalities and outlets represented only 9.6 percent of the 51,164 tweets sent by these accounts. After the ban, right-leaning personalities and outlets accounted for 40.1 percent of the 37,773 most retweeted posts by these accounts.
The ban had no effect on the most prolific accounts
We also analyzed the 10 top-tweeting accounts — meaning which accounts sent the most tweets — on each of the days, a total of 200 accounts, to better assess what accounts were trying to affect the debate. We usedbot sentinel, a tool that uses machine learning and artificial intelligence to categorize account behavior in accordance with Twitter’s policies, as well as analysis to determine the kind of content they typically shared.
Most of the 200 accounts leaned to the right, politically. In addition, when we focused on the top-tweeting account for each day in the sample, a total of 20 accounts, we found that these accounts were always right-leaning. Further, 18 of these 20 accounts spread right-leaning conspiracy theories about the 2020 presidential election and the left more generally. The two most prolific accounts sent 1,729 conspiracy theory tweets during the 20-day window, including one of the most retweeted comments before the ban and two after the ban.
In short, the ban did not hurt users’ ability or willingness to spread disinformation.
The ban does not diminish the audience for disinformation
We explored the relationships among the accounts whose audit-related tweets were retweeted most often, the accounts that sent the most audit-related tweets, and their followers. To understand whether communities form and share information, we included the different accounts in a social network analysis, which uses graph theory and networks to explore structures in data.
We found two communities of right-leaning accounts tweeting and retweeting with one another during this time. While we do not know whether all the tweets were about the audit, the communities are different. One used hashtags to express support for Donald Trump and traditional conservative causes, such as opposition toabortionand support for gun rights. The other used hashtags to express support for Trump and to advocate for more right-wing and conspiracy-minded causes, such as a ban on sharia law, proclamation that pedophilia is evil, and claims that the country suffers from mass psychosis.
The Twitter ban had virtually no effect on the conspiracy-minded community.
Does the takeover matter?
Many worry that Musk will let disinformation spread freely. But that assumes Twitter is currently stanching the spread. That’s not what we find. Bans don’t necessarily improve discussion or the quality of information shared on important political issues. Moreover, bans don’t prevent radical accounts from finding one another or slow their sharing of disinformation.
Deana A. Rohlinger is an associate professor in the Department of Sociology and a research associate in the Pepper Institute of Aging and Public Policy at Florida State University. Her work examines politics, social movements, and political participation, typically using a comparative perspective.
