As the Democratic presidential primary contest between Hillary Clinton and Bernie Sanders has grown much closer, some commentators suggest that Clinton’s campaign is being bombarded with sexist invective — especially from Sanders supporters.
But based on systematic analysis of recent Twitter data, we find that little of the attacks directed at Clinton can be attributed to the left in general or Sanders supporters in particular. And a remarkably small number of tweets mentioning Clinton contain the most egregious and overt forms of sexism: gendered slurs.
Quite a few observers have suggested that a hostile and misogynistic ethos has built up among some men who #FeelTheBern — the so-called “Bernie Bros.” Journalist Joan Walsh, for example, complains of “vile” online trolling, harassment and sexism coming from “Berniebot keyboard warriors.”
On the other hand, writer Glenn Greenwald calls the Bernie Bro narrative an “all-purpose, handy pro-Clinton smear” used to undermine legitimate criticism of Clinton. He argues that Sanders supporters are not “uniquely abusive and misogynistic in their online behavior” and points out that there are plenty of women among die-hard Sanders supporters.
So far, the case presented by each side has been based on a series of (admittedly often disturbing) anecdotes, rather than systematic analysis.
We, therefore, decided to investigate how much hostility and sexism Clinton faces, as well as who seems to be behind such attacks, on a social media platform particularly notorious for abusive behavior: Twitter. We captured 101,021 tweets containing mentions of @HillaryClinton, @BernieSanders, or both in real time during the New Hampshire primary and analyzed both the content and senders of these tweets.
Here’s how we did our analysis
The first step of our analysis examined what types of words were associated with tweets mentioning each candidate. We fit a series of statistical models with words that occur at least 50 times as the independent variables, and with whom the tweet mentions (i.e., @HillaryClinton, @BernieSanders, or both) as the dependent variable. In other words, we examined how well the presence of a word in a tweet predicts who the tweet is about.
Naturally, we were not interested in the actual prediction (we knew who the tweet was talking about), but in the weight assigned to each word. The larger the weight, the stronger the word’s association with a given candidate.
We then took a closer look at the tone of the 100 words most strongly associated each candidate. To assess tone, we coded a random sample of tweets containing each of these words. Rather than let a word “speak for itself,” we wanted to place it in context to more accurately understand whether the term was being used in a positive, negative, or neutral way.
For example, on its own, the word “email” is neutral, but when used in tweets mentioning @HillaryClinton, “email” relates to the ongoing investigation of Clinton’s use of private servers while Secretary of State and actually carries a negative connotation. Combining this information with the weight the model assigned to each word provides us with an intriguing picture of how Twitter users address the Democratic candidates.
As these two figures show, the words associated with @HillaryClinton are far more negative than those associated with @BernieSanders. Only nine words associated with Sanders are negative in tone, and 40 are positive. Just eight words associated with Clinton are positive, and 53 are negative.
However, most of the negative words directed at Clinton — such as “Benghazi,” “injustice,” “jail,” “emails” and “unborn” — are associated with longstanding rightwing claims and do not point to specific critiques coming from the left. Indeed, just two of the words associated with @HillaryClinton – “Goldman” and “donors” — are tied to complaints commonly offered by progressives (in this case the claim that Clinton is cozy with Wall Street elite). And both of these words fall near the bottom of the top 100 rankings (at 97th and 98th, respectively).
So did we find sexist rhetoric?
Drawing on the literature addressing mass media coverage of female politicians, we next evaluated how many of these words carried gendered connotations. Gendered terms include those related to emotions, family life, policy areas often considered more masculine or feminine (e.g., defense and the military vs. education), physical appearance and so on.
As the following figure demonstrates, only 12 words associated with @BernieSanders carry gendered meaning. None were negative, and three of the four positive words refer to Sanders’s prowess on the basketball court.
In contrast, among the 29 gendered words associated with Clinton, 13 carry a negative connotation, including several related to her husband. Indeed, the vast majority of the tweets mentioning Bill Clinton either blame Hillary Clinton for, or suggest her complicity in, her husband’s sexual misdeeds. Interestingly, three of the four positively toned gendered words are stereotypically masculine. All three suggest Clinton is “fighting” for someone or something.
Most of the negative words in this group are consistent with misogynistic claims, and several are particularly disturbing (e.g., “vagina,” “b*tch”). To understand these dynamics even further, we searched all tweets mentioning @HillaryClinton for any term from a set of 30 common gendered slurs, such as “bimbo,” “slut,” “whore,” and “shrill.” (A number of these slurs are too crude to mention.) We then hand-coded all of the tweets in which these slurs appeared, assessing whether the insult was indeed directed at Hillary Clinton.
In the end, we found that 23 of the 30 gendered slurs were directed at Clinton. However, out of a total 52,181 tweets mentioning @HillaryClinton, just 606, or 1.16 percent, contained these insults. While these slurs only represent one particularly overt form of sexism, the fact that so few were present in these tweets is remarkable.
Are Bernie Bros behind the slurs?
This is quite a small number. But any such slur is troubling. And we still need to know who is responsible for the invective.
Therefore, in the final stage of analysis, we coded whether each slur originated from a Bernie Sanders supporter (as determined by their Twitter bios or corpus of tweets) and, among verifiable Sanders supporters, whether the sender was male, female, or unknown.
The vast majority of the slurs were associated with Twitter users on the right — particularly self-identified Trump supporters. But 14.7 percent came from those backing Sanders. Among Sanders supporters, 60.6 percent tweeting gendered slurs were men, 29.2 percent women, and 10.1 percent unknown. Most slurs are used by both genders, but some seem more specific: in the data, “whore” was used as an insult mostly by female Sanders supporters.
Thus, while we do find some evidence of Bernie Bros’ bad behavior, abuse against Clinton by Sanders supporters — both male and female —seems relatively limited. Clinton certainly faces a barrage of negativity and a heavy dose of sexism on Twitter. But that mostly appears to come from the right.
And though any and all instances of sexist slurs deserve condemnation, Sanders’ keyboard warriors accounted for just 89 such tweets during the New Hampshire primary. That is a mere 0.17 percent of all the tweets mentioning @HillaryClinton that we examined.
That analysis doesn’t mean Sanders supporters are free of misogyny or sexism. We have not captured all forms of online sexism, only some of the most egregious and overt. And we do not examine abuse of female Clinton supporters, just of Clinton herself.
However, our analysis does provide a better understanding of the extent and character of the attacks lodged against Hillary Clinton online. And it seems relatively little abuse originates from the left. The fact that Sanders, also roundly denounced by the right, receives so little negative attention on Twitter fits with a gendered dynamic. Research shows us that all women —not just politicians, not just Clinton — are much more likely to face harassment and abuse online than are men.
Rebekah Tromble is an assistant professor in the Institute of Political Science at Leiden University. Dirk Hovy is a postdoctoral researcher in computational sociolinguistics at the University of Copenhagen.