The 2016 election has been a strange and surprising one, and the rise of two highly publicized candidates symbolize this unexpectedness: Donald Trump and Bernie Sanders. Few experts would have predicted that a reality TV star and an avowed democratic socialist would inspire strong bases of support in a U.S. presidential election.
The two candidates are not mirror-images in most ways. Donald Trump is currently leading in national polls over his Republican rivals, while Sanders is trailing Clinton. Sanders has been a senator for a decade, while Trump has never held any political office. Trump’s fame and extreme personality make it practically unfair to compare him to any other human being. But both candidates share a narrative of serving as a potential spoiler to their respective party establishments, and pundits are frequently nonplussed by the success of each.
There remains uncertainty about what candidate will gain their party’s nomination. It is not unusual, at this point in primary season, for a highly touted and favored candidate to later flounder (Howard Dean in 2004, Clinton herself in 2008, Newt Gingrich in 2012, to name three). However, the media attention focused on both candidates and the caucus victories of Trump in particular make this year’s race unusual and unpredictable.
We know much about the demographic groups inspired by each of these unusual candidates.
Sanders supporters skew young, white, and more extremely left-wing. Trump fans are older, less educated, and less religious than supporters of other Republican candidates. But further insight can be provided by looking deeper into people’s daily language use. What do Trump and Sanders supporters post that distinguish them from everyone else?
To answer this question, we collected data from 3,152 American participants recruited online. Each participant provided his or her private Twitter handle (to qualify, they had to tweet regularly), and answered the question “If the U.S. election were held today, which candidate would you vote for?”
Surprisingly, Sanders won our imaginary election by a landslide: nearly a third of participants indicated their intention to vote for him. Trump came in fourth (after Clinton and “I would not vote for any candidate”) with about 15% of the vote. These results likely differ from nationwide polling because of our survey’s requirement that respondents use Twitter (its users skew younger, less Caucasian, and more liberal).
Nonetheless, we now had the ability to delve into the mindsets of people involved in these political movements. We ran automatic, data-driven Differential Language Analysis to determine the social media language that distinguishes Sanders supporters and Trump supporters. That is, we correlated the likelihood that people used a word or phrase with their status as a Trump or Sanders supporter or not.
First, the results for Trump supporters:
Viewed qualitatively, the main pattern is the obvious positivity. Trump supporters appear hopeful and sanguine, using words like “fantastic” and “awesome.”
Next, the results for Sanders:
Contrasted to the Trump supporters, there is a fairly obvious negative valence in the tweets of Sanders fans. Much of this is a result of the highly salient profanity… Sanders supporters swear more than everyone else. They also describe negative experiences and feelings: “screaming,” “my head,” “crying.”
They also were more likely to talk about themselves, using phrases that contain “I” “I’m” or “my.” Finally, they used qualifying or hedging words, packing their statements with meaningless words like “actually,” “honestly,” and “basically.”
Let’s turn to another way of analyzing language: the Linguistic Inquiry and Word Count (LIWC; Tausczik & Pennebaker, 2010 ). LIWC is a process of automatically sorting words into a host of premade dictionaries that have psychological meaning. For instance, if you used LIWC with the sentence “The only thing we have to fear is fear itself,” it would reveal that you had one article (“the”), two references to anxiety (“fear” and “fear”), and one pronoun (“itself”) among others. This technique can give you a sense of the categories of language people use.
Looking at Sanders supporters first, there were a number of LIWC categories significantly associated with membership in this group. The five strongest correlations were:
- Conjunctions (and, but, because)
- Negative emotion (agonizing, hates, tedious)
- First-person singular pronouns (I, mine, myself)
- Adverbs (basically, thoughtfully, quickly)
- Sexual references (breasts, bdsm, gay)
Taken together, most of these categories imply a general wordiness that distinguishes Sanders supporters from everyone else. They write more, and they write more completely. They are also distinguished by their negativity and references to sex (although, bear in mind that LIWC counts profanity as negative emotion and “fuck” as a reference to sex. While anger, despair, or sexuality might be a common reason people swear, the youth and liberalism of Sanders supporters may also lead them to be less concerned with traditional social propriety.)
The top five categories which distinguished Trump supports were:
- Achievement (obtains, loser, victory)
- Reward (jackpot, benefits, success)
- Money (budget, earnings, payment)
- Leisure (baseball, fishing, xbox)
- Power (conquer, chairman, menial)
These results largely speak for themselves. We will add that the “power” category measures an orientation toward hierarchies and power structures generally.
To summarize, we were able to find several verbal behaviors reliably distinguishing members of these two notable political movements. Sanders fans are profane and wordy, while Trump fans are victory-oriented and talk about money. One piece of insight is the general negativity of Sanders supporters and the overwhelming optimism of Trump supporters; these results contradict media caricatures that support for Sanders comes from blind, starry-eyed optimism, while support for Trump is accompanied by inchoate rage.
The 2016 election seems to grow more confusing and complicated by the day. Our results at least provide some degree of relief by demystifying the behaviors and psychological traits of the supporters of two unexpectedly popular candidates.