In the months leading up to the presidential election, both sides of the political divide expressed concern about potential impediments to the democratic process. Democrats worried about Trump’s efforts to undermine the USPS and Republican-led state governments disenfranchising voters in various ways. Republicans proclaimed that mail-in voting was inherently vulnerable to fraud by individual voters.
After the election, the relative volume of these two concerns shifted. All major media networks have declared Joe Biden the winner. Democrats focus on recording all outstanding mail-in ballots. Republicans are up in arms about state elections officials miscounting or producing pro-Biden ballots out of thin air — all while Trump himself refuses to acknowledge even the possibility of defeat, supported by senior leadership of the GOP.
Unsurprisingly, much of this conversation is playing out on social media. But who is doing the talking? MarvelousAI, a startup specializing in media narrative analysis, has been tracking relevant content on Twitter. The data suggest that Democrats’ concern is driven by real people (activists and media figures), while trolls and bots are the primary amplifiers of the Republican accusations of fraud.
We compared the content and participants of the left-leaning mentions of “voter suppression” starting in October 2020 vs. right-leaning mentions of “#stopthesteal” starting November 3, 2020.
Voter Suppression v. Election Fraud
Voter suppression received on average 2000 tweets/day throughout the time examined. “StopTheSteal” gained steam after Nov. 3 and is averaging around 4000 tweets/day. MarvelousAI uses the link-sharing behavior of users to infer their political bias, based on ratings of news websites produced by the Media Bias Fact Check project. In addition to political bias classification, users can be categorized into groups based on how bot-like their behavior is. We rely on third-party lists of known bots (such as botsentinel, Hamilton 68 etc.) and a handful of common bot identification heuristics such as account creation date, long sequences of numbers in the account name, etc. The chart below shows a breakdown of left-leaning vs. right-leaning tweets by user type.
The contrast is striking. While neither side of the conversation is bot-free, the right-leaning discussion is at least 40% inauthentic, compared to about 20% for the left-leaning conversation.
Some right-leaning accounts are especially brazen. For example, @Christi68901055 was created in November 2020, has no last name or profile photo, follows only four people (all related to Trump and his campaign), but has produced over 4000 tweets in the past week:
This account seems entirely dedicated to promoting the “Stop the Steal” narrative. Occasionally it gets fooled into retweeting satire:
While the “Stop the Steal” narrative took off after Election Day, seeds were planted earlier. In our data, there are mentions of this hashtag in late September, e.g.
The @Cuds_1246 account follows a common pattern in right-wing Twitter: it was created in 2017, participates in ‘follow-back’ behavior (i.e., has almost the same number of followers and accounts it follows, in the many thousands), and otherwise has no personally-identifying information.
The @Infinite_Ennui account, although not a bot by our measure, is particularly active in the conversation, tweeting as much as 60 times/day. If even briefly got suspended by Twitter for violating its policies (but was back in action within a day):
Trump’s refusal to recognize the outcome of a legitimate election is disturbing and dangerous. At the same time, one should not extrapolate the amount of genuine support he has based on social media mentions. As always, we will continue to monitor the situation; get in touch if you are interested in a deeper dive into the data or want to collaborate.