Rude Awakening

Who sparks more UNcivil conversations on Twitter — Trump or Clinton?

by Andrew Heyward and Uzra Khan

Just how crude and insulting is the social conversation around Election 2016? And how does it vary by candidate?

Electome, a set of algorithms created by data scientists here at MIT’s Laboratory for Social Machines (part of the Media Lab), has been collecting and classifying election-related tweets since early last year. One of our tools is an Incivility Index called Tonar (more about Electome here and Tonar here).

Drawing on Electome’s access to the full output of Twitter, Tonar tracks the percentage of tweets about the campaign that contain profanity; insults; ethnic and sexual slurs; and violent expressions.

Last month, when both parties held their national conventions and political passions naturally ran high, the election-related conversation on Twitter was consistently more civil than conventional wisdom might predict. Even on days when volume spiked, like July 21st (Trump’s acceptance speech) and July 28th (Clinton’s), only one in 20 tweets were uncivil by our metrics.

We wondered which candidate’s supporters were more likely to use uncivil terms. It turns out there’s no difference. Using people who follow either Trump or Clinton exclusively as a proxy for their supporters, Tonar shows that both groups hovered around the 6% incivility mark in July. Within the pool of uncivil tweets, 15% came from Trump followers, and 8% from Clinton followers, but this closely mirrors the proportion of overall election-related tweets coming from those two groups.

But there is an interesting difference in tweets about the two candidates. Tweets that mention Trump are more likely to be rude, crude or cruel than tweets that mention Clinton, even tweets by Trump’s own supporters.

Among Clinton supporters last month, 59% of the uncivil tweets mentioned Trump by name, and 20% mentioned Clinton. No surprise there. Here are a few of the more printable ones.

But among Trump supporters, while 37% of uncivil tweets referred specifically to Clinton, like this one…

…46% mentioned Trump.

The takeaway: according to Tonar, conversations that involve Trump are more likely to be — what’s the civil way to say it — colorful. No matter who’s having them.

Andrew Heyward is a visiting researcher at the MIT Media Lab’s Laboratory for Social Machines . Uzra Khan, a recent graduate of Harvard’s Kennedy School of Government, is spending the summer as a project manager there. Soroush Vosoughi and Prashanth Vijayaraghavan, researchers at the Laboratory for Social Machines, developed the analytics for this post.