ROCHESTER, N.Y. — Could Twitter have predicted the outcome of the general election? A 14-month study of Hillary Clinton’s and Donald Trump’s tweets may provide insight into how and why the Republican won the presidency.
Jiebo Luo and Yu Wang, two Ph.D students at the University of Rochester, examined both candidates’ feeds, making a number of astute observations over a series of eight papers. “We wanted to understand how each of the candidate’s campaigns evolved, and be able to explain why someone won or lost,” Luo, an associate professor of computer science, says in a university release.
One substantial finding was that Trump’s following was positively correlated with the amount he used Twitter. As it turned out, his popularity continued to grow as he tweeted, regardless of how much controversy he created in real life.
Of course, his tweets also seemed to help Clinton at time. The researchers discovered that accusing the Democrat of playing the “woman card” led to women becoming more supportive of her on social media. Still, Trump’s attacks had no impact on his own base of support.
Perhaps more deeply explaining the outcome, Luo and Wang found that Clinton’s base of female supporters didn’t exceed Trump’s. A general phenomenon observed in many past elections has been a “gender affinity effect,” in which females predominantly vote for a female candidate.
The study also points to Bernie Sanders playing a role in Trump’s victory. Sanders’ supporters changing sides after he dropped out of the race may have led to an unexpected outcome: an influx of Trump supporters at the expense of Clinton supporters.
The researchers were interested in examining Twitter data because of the burgeoning use of the platform and other social media platforms by politicians, along with its ease of use for data mining and analysis.
In order to determine the demographics of supporters, Luo and Wang used a customized neural network program. Usernames, tweets, and other data were also looked at.
Although their methods may not have been 100% accurate, they were able to draw from a dramatically larger sample size than other methods, such as polls.
“In the end, even though we chose not to make any predictions, we were not surprised at all that Donald Trump won,” says Luo.
Luo has suggested that their approach, which can peer into the opinions of millions, has broader applications; it could also be used to gauge opinion on “the next generation of iPhones, or a new model of car,” for instance.