All Attacks are not Created Equal: Identifying Bias in Accusations

MarvelousAI tracks social media narratives surrounding the 2020 US Presidential campaign. We identify differences in how leading male and female candidates are covered, with a focus on Joe Biden, Bernie Sanders, Pete Buttigieg, Elizabeth Warren, Kamala Harris, and Amy Klobuchar.

We’ve found that ‘‘Twitterverse”  is still not accepting of female politicians, but the nature of attacks against specific candidates change over time.

Early on in the presidential primary campaign, there were clear gender-correlated differences in the way male and female candidates are attacked online. Female candidates received notably more character attacks, and more vitriolic attention from right-wing / non-credible sources. The charts below identify bias and credibility of twitter accounts that tweet about the six top Democratic candidates. Left-to-right is political bias, up-down is credibility.

Bias/Credibility Distribution by Candidate

We found a distinct “gender penalty” for candidates. For methodology details and context, see #SHEPERSISTED and “Running While Female: Using AI to Track how Twitter Commentary Disadvantages Women in the 2020 U.S. Primaries.”

As the campaign progressed, attacks got worse for some candidates, especially Kamala Harris. Journalist Aleszu Bajak wrote about it on StorybenchQuantifying the Twitter attacks on Kamala Harris during and following the Democratic debates.

However, there were also more subtle distinctions between candidates. And as the public becomes more familiar with individual candidates, it becomes less plausible to say that all female candidates face the same type of individualized attacks.

At the same time, our data shows front-runner candidates are being attacked on character regardless of gender.
In the wake of the November debate, we identified significant attacks against the top six candidates except Sanders, regardless of gender.

Most of the identified attacks are focused on the character/identity of the candidates, although Kamala Harris also gets disproportionately attacked on electability.

Understanding what the individualized character attacks are actually about, however, is instructive. Biden is accused of corrupt behavior concerning Ukraine and tied to several impeachment inquiry narratives. He’s also, for example, accused of being too old. These attacks originate in the right-wing corners of the ‘Twitterverse.’

Women and non-conforming men (i.e., Buttigieg) are instead treated as inauthentic outsiders, with attacks from all sides of the political spectrum, with the occasional “angry woman” trope. Examples of differences in attacks between male and female candidates:

Mainstream media was guilty of this as well.

We’ve found that any candidate who’s doing well in the presidential primaries will come under attack. While female and non-conforming candidates are attacked for not belonging in their identities, straight, white male candidates face more “familiar” and expected accusations.

There is at least one sign that all is not hopeless. Attacks on Elizabeth Warren are counterbalanced by organic support for her policies and her talents as a politician.

Marvelous AI hopes that tracking, measuring, and exposing the online bias against non-cis, white, male candidates will help make online public discourse fairer.

Related Posts