The researchers analyzed 4.4 million tweets sent in the wake of 21 U.S. mass shootings — defined as four or more people shot (not including the shooter) — from 2015 to 2018 and selected the incidents that offered the most Twitter data. They used natural language processing, a field of artificial intelligence that seeks to understand language.
As candidates gear up for a presidential race in which gun control could be a top issue, the study joins a host of research examining the political polarization often blamed for government gridlock. It builds on the methods of another project from 2016 — conducted by some of the same researchers — that found language from members of Congress grew more distinctively partisan over the years, especially after the mid-1990s.
“We live in a very polarized time,” Dan Jurafsky, a Stanford professor and co-author of the study on responses to mass shootings, told Stanford News. “Understanding what different groups of people say and why is the first step in determining how we can help bring people together.”
The differences go beyond a focus on victims versus perpetrators, according to the new study, which was presented earlier this month at a conference with peer-reviewed submissions. Democrats are also more likely to emphasize policy change, the study concluded, while Republicans tend to zero in on the facts of the shooting.
Democrats’ language also conveys more sadness or trust than that of Republicans, the researchers say. Republicans’ language, meanwhile, projects more feelings of fear and disgust, especially when the shooter is nonwhite.
And Democrats are 25 percent more likely than Republicans to use the word “terrorist” when a shooter is white, the study found. Republicans, on the other hand, are 25 percent more likely to use “terrorist” when the shooter is African American, Middle Eastern or Hispanic or Latino.
Some of the findings reinforce other studies’ conclusions, Dora Demszky, a Stanford doctoral candidate and lead author of the study, told The Washington Post. But others took the researchers by surprise.
The team expected Republicans to use the word “terrorist” more frequently when a shooter was Middle Eastern, because of common perceptions of terrorists, Demszky said. But they were taken aback to find a similar trend for incidents involving African American and Hispanic or Latino shooters.
“We didn’t really expect that the terrorism frame would extend to these other ethnicities for Republicans,” Demszky said.
The researchers ignored retweets and categorized Twitter users by party based on whether they followed more accounts from Democratic or Republican politicians.
Many analyses illustrate how Republicans and Democrats are moving further apart in their views: Data from Voteview and Gallup, for example, shows that Congress has become more polarized and that public approval rates of the president have pulled further apart along party lines.
Gun policy is certainly a divisive issue, especially as the United States mourns the victims of deadly attacks with firearms everywhere from schools to movie theaters to places of worship. With each tragedy, Democrats call for stricter gun control, while Republicans tend to argue weapons aren’t the issue.
A Vox analysis in June counted more than 2,000 mass shootings, leaving nearly 2,500 dead, since the 2012 shooting at Sandy Hook Elementary School that left 20 children and six adults dead. The archive Vox used to compile these counts defines mass shootings as incidents in which at least four people besides the perpetrator are shot.
A Washington Post analysis that looks more narrowly at public mass shootings in which four or more people were killed tallied more than 1,100 deaths from 163 incidents since 1966.
With shootings and the ensuing clashes over gun control perpetually in the news, the Stanford and Brown researchers’ work opens up questions for future research. Sharing their work at a computational linguistics conference, Demszky said, the researchers encountered colleagues with all sorts of questions for further investigation: How has the polarization they observed developed over the years? How do “bots” — the fake social media accounts like those Russia created in an effort to influence the 2016 presidential election — factor in?
For now, Demszky will be moving on to other projects. But the code and data for the project are public, and she said she looks forward to what others will discover.
The findings, Demszky qualified, are “purely descriptive” — the team has not examined why Americans of different political persuasions have such distinct reactions on Twitter. But she thinks capturing the disparities is crucial to better explaining political polarization.
“Maybe we don’t need to do something about it or may we cannot do something about it, but first we definitely need to understand that to figure out the next steps,” she said.