Blog / Gender

How do presidents sound?

We used Textio to evaluate presidential speeches

Recently I was watching one of the presidential debates with my daughter. As the candidates fielded questions about immigration, I reflected on how much I’d heard them talk this election cycle. Beyond their views, I had a clear sense of each candidate’s voice: how they spoke, where they focused and got defensive, the language they reverted to under stress.

It started me thinking: what would Textio have to say about these candidates?

Picture of Ted Cruz, Hilary Clinton, Donald Trump, Bernie Sanders, Marco Rubio

Textio is trained on job listings; when you put your document into Textio, we’ll compare it to previously successful listings of its type and tell you how to make yours better. How biased your listing is — which groups your language appeals to — is a key part of how your job performs. It’s not the only factor, but it’s area where Textio is particularly strong.

Because it’s so broadly important, people apply our bias technology to all kinds of docs beyond job listings. People have used Textio to evaluate the bias of email, performance reviews, resumes, marketing websites, even course syllabi and newspaper articles. So we figured, why not presidential speeches?

In all, Textio looked at 126,362 words from campaign speeches by Bernie Sanders, Donald Trump, Hillary Clinton, Marco Rubio, and Ted Cruz.

Here’s what we found.

All the candidates talk about family, community, and contributions.

We quickly saw that we needed to tweak our bias evaluator for this medium: all the candidates talk about contributions, family, and community pretty much constantly (although Clinton talks about family five times as often as any other candidate, and Cruz and Sanders asks for campaign contributions almost twice as often). There’s also some language in political speech that doesn’t show up directly in our usual data set, so Textio had to measure linguistic distance between terms to get the right bias model for this medium.

All the candidates tell more stories about men.

Political speeches are peppered with personal anecdotes: the story of the out-of-work truck driver in Tallahassee or the parents struggling to pay their mortgage in Poughkeepsie. For every thousand words of speaking, the candidates average over four of these stories — and most of them are about men. Even for Clinton, the candidate who is closest to balanced, there are three times more stories about men than women. Cruz includes nine times more anecdotes about men, Rubio 18 times more.

When these stories focus on parents (either the candidates’ own or others), the gender gap is even wider. Rubio has a higher representation of mothers than the rest; he tells about four times more stories about fathers than mothers. (Before you scratch your head reconciling that with the above, it turns out that nearly the only woman he talks about is his own mother.)

The male candidates talk a lot about destroying stuff.

Textio looked at candidates’ specific word choices. While all the candidates include a mix of masculine-biased and feminine-biased language, they also have individual preferences. Clinton is the only candidate whose language bias Textio recognizes as feminine; her speeches are packed with phrases like balance work and family, open our hearts, and in this together. She uses about half as many phrases that Textio recognizes as masculine, phrases like command respect, forced to choose, and set the standard.

All the other candidates include more masculine-biased language — in the case of Bernie Sanders, Marco Rubio, and especially Ted Cruz, a lot more. Their speeches are riddled with language that Textio flags as masculine, phrases like demanding and stringent, obliterate, totally destroy, and towering strength.

The specifics vary from candidate to candidate — Sanders is more likely to wage a decisive campaign against people who are destroying American democracy, while Rubio and Cruz stand unapologetically for Christian values and vow to make the US a powerful force internationally by hunting down ISIS. But regardless of their political stance, the metaphors and verbs these candidates use are highly masculine.

Trump is an oddball.

Trump’s language stands out in a few ways. First of all, there’s simply more of it. His speeches are about 35% longer than everyone else’s.

Much has been written about Trump’s speech. His language is riddled with playground insults like idiot, dummy, and imbecile. Textio recognizes these as masculine-biased — not because most men like them, but because virtually all women don’t. In his case, it goes beyond subconscious bias: this man brought up his penis in a national debate. Trump reminds one a little of the basketball coach in Hoosiers: beyond the overt stuff, language like sheer force of will, hard-driving, nuke, and play hardball is quite common.

But here’s where things get weird. Alongside this language he includes many more feminine-biased terms than the other male candidates — and his use of these terms is increasing over time. He proclaims that he is great for women and talks about his daughter (and her pregnancy) every chance he gets. Trump talks much more than the other male candidates about his beautiful family, his wonderful team, and how well he builds lasting and effective relationships.

Though he is still decidedly male-biased, the whole messy collection leaves his balance of speech closer to neutral than the other men running. The fact that he is such a polarizing candidate is fully reflected in the language he uses.

Trump is really an oddball.

Textio also looked at some syntactic characteristics of candidate speech that go beyond bias: how much candidates repeat themselves, how their sentences are structured, how much they use language that talks about themselves versus language that speaks directly to voters.

It turns out that candidate language has a lot in common across the political spectrum. Political speeches are highly repetitive, with candidates using the same phrases over and over again. Popular speeches also balance between language that empathizes with voters and language that tells the candidate’s own story.

But here again, Trump is not like the others. He repeats himself less. He talks about himself twice as often as Cruz, the candidate with the next most egocentric language. He addresses voters directly less often than anyone else. He uses significantly fewer adjectives.

For fun, we dug up public speech transcripts from other media and business tycoons ranging from Warren Buffett and Rupert Murdoch to Oprah Winfrey and Ted Turner. We wondered: how would the structural patterns in Trump’s speech compare to these others?

All the tycoons we selected show up as male-biased in their speeches other than Oprah Winfrey, Martha Stewart, and Mark Zuckerberg (the latter two show up neutral). But bias aside, when it comes to these structural characteristics, Trump is much more in line with this group than he is with the other political candidates.

Across our data set, Trump’s speech is structurally most similar to Ross Perot’s — the other oddball business tycoon who made a serious run at the presidency.

Want to find out how you sound?

No candidate wants to alienate half the voters. Yet in one of the first US elections where a woman is seriously contending for the presidency, most of the candidates mainly speak to just half the population.

While you may not be running for president, you can’t afford not to know how you sound. If you’re hiring, Textio can help you find out how you sound to the people you are trying to attract.


Topics: Gender, Machine Learning, Politics, Uncategorized, AI