Penn scholars distinguish presidential candidates by analyzing their words

The lead-up to U.S. presidential elections can be drawn out and jargon-filled, challenging voters to distinguish between what politicians say and what they actually mean. A team from the World Well-Being Project (WWBP) at Penn’s Positive Psychology Center aims to cut through the confusion by analyzing individual words 2016 presidential hopefuls used during debates.

The researchers, who included postdoctoral fellows Anneke Buffone, Jordan Carpenter, and Daniel Preotiuc, along with WWBP co-founder Johannes Eichstaedt, also looked at which unique terms were tweeted by supporters of several candidates.

“We’re all extremely politically interested,” Eichstaedt says. “We thought it would be natural to see if we could try to distill insight out of that political language.”

Using two different algorithms, the group analyzed candidate statements from the 11 Republican and eight Democratic debates, as well as tweets from likely voters. They disregarded words used by multiple candidates to get at what distinguishes each individually, then, using their algorithm, built word clouds showcasing frequently used terms with the largest, darkest lettering.

“The [results] in many ways confirm what we already know,” Eichstaedt says, “but there are also some interesting, even surprising elements that tell you a little bit about the priorities of these candidates.”

Businessman Donald Trump’s diagram overflows with words like “tremendous” and “nasty,” “eminent domain,” and “currency.

“[U.S. Sen.] Ted Cruz has a more fear-based rhetoric, and compared to the other candidates, is really obsessed with Texas and political attacks, Washington, and the Second Amendment,” Buffone says. “[Ohio Gov.] John Kasich looks like a friendly accountant.”

On the Democratic side, former Secretary of State Hillary Clinton prioritizes health care, prompting “affordable,” “coverage,” and “prescription” to take top billing in her word cloud. For U.S. Sen. Bernie Sanders, “campaign finance” and “infrastructure” jump out, along with the word “handful,” usually within phrases like “handful of billionaires” or “handful of wealthy,” and terms like “crumbling” and “corrupt.

Unlike the word clouds based on the debates, the Twitter language of several candidates’ followers surprised the researchers.

The team asked a non-representative sample of people online if the election took place tomorrow, whom they would choose for president. Of 3,152 respondents, 1,000 or so “voted for” Sanders and about 750 went for Clinton. Trump took fourth place, after the group who said they wouldn’t elect any of the current candidates.

Sanders supporters tend to be self-effacing and anxious, according to Carpenter.

“They’re conversational, casual, and foul-mouthed,” he says. “They post about everyday irritations, discuss plot twists they didn’t like on TV shows, and will swear at the drop of a hat. When they talk about politics, though, they do get earnestly mad.”

Conversely, the researchers say Trump supporters are surprisingly positive.

“They’re just more excited,” Preotiuc says, “and they actually post about victory.” They’re also more likely than Sanders supporters to make broad pronouncements, and rarely note their political views, except to retweet their candidate or troll those whose allegiance falls elsewhere. 

Such unanticipated results “contradict media caricatures that support for Sanders comes from blind, starry-eyed optimism, while support for Trump is accompanied by inchoate rage,” Carpenter writes in a blog post about this work.

Preotiuc says the researchers do not currently have plans to publish this study. Instead, they wanted to use the language-analysis tools at their disposal to take a data-driven approach to the election rather than one that relies on theory. They do, however, plan to conduct similar analyses during the general election.

“Data-driven snapshots using impartial methods of the candidates’ rhetoric and the language of their supporters is helpful to track the dynamics of this race,” Carpenter says, “and may be another data point when deciding on a candidate.”

Word Cloud