Misogyny on Twitter

Oliver Keyes bio photo By Oliver Keyes Comment

(Content warning: misogynistic slurs)

Earlier today Tauriq pointed me to a Demos study on online harassment that’s been doing the rounds recently (and Katherine linked me to a copy of the report. Credit to the both of them).

TL;DR actually a really good study but the core assertion in all the media coverage doesn’t hold up, because media.

The press coverage on the study (which is actually ~2 years old) is pretty strong in its claims, because that’s how headlines work; Buzzfeed says that the study shows half of misogynistic abuse on twitter comes from women, as does the Telegraph and Aunty Beeb.

Whenever I see super-strong claims I tend to brace for a poorly designed study, doubly-so when the strong claims turn what we “know” on its head (extraordinary claims require blah blah blah), but I was actually pleasantly surprised.

The study covered the use of “whore” and “slut”, as words, on Twitter, from January-February 2014. Looking specifically at tweets from the UK, the authors applied a classifier to filter out irrelevant tweets (“rape seed oil”). This left about 108k tweets in total, which were both quantiatively and qualitatively analysed (read: a classifier, and going through samples by hand to feed the classifier and check its workings).

The authors took this hand-coded-and-or-classified data and broke it down even further. So direct abuse and harassment (@tweeting slurs at someone) is distinguished from general misogyny (tweeting a slur about someone, but not at them) is distinguished from conversational (“I was such a [slur] 10 years ago”).

The paper is really good. The breakdowns are logical and make sense; the accompanying of automated classification with hand-coding is vital (machines are really really bad at identifying context or emotional tone), and they even published the classifier accuracy data in an appendix so you can check they aren’t just phoning it in. The one open question I have is how gender was identified (and what you do, in identification, with twitter eggs).

The media coverage is total shite, though.

The BBC, the Telegraph, Buzzfeed - they all ran with the premise that misogynistic abuse is equally distributed by gender. That’s what the study found, they say. Except actually that’s not what the study found at all, and the authors know it; their conclusion is:

Women are as almost as likely as men to use the terms ‘slut’ and ‘whore’ on Twitter. Not only are women using these words, they are directing them at each other, both casually and offensively; women are increasingly more inclined to engage in discourses using the same language that has been, and continues to be, used as derogatory against them.

Sounds like the same thing - it’s not. Women are almost as likely to use the term, but the context might be very very different. While the study’s methodology decomposed tweets to look at context - directed abuse and invective versus generally-broadcasted abuse versus casual conversation - the actual gender-breakdown bit doesn’t.

This is vital. There’s a big difference between those slurs as reclaimed terms (for example, the slutwalk), or to refer to oneself in a self-deprecating way, and pointing those slurs at specific individuals. There’s also a big difference between making a claim about individuals into the aether and @tweeting it at the target. Both are toxic and misogynistic, but the actual experience of having it happen about you is very, very different.

The conclusion does say that women offensively direct the slurs at each other, and I believe it (there’s a lot of internalised misogyny lying around, although it’s not my job to critique it directly) - but there’s nothing in the actual body of the paper indicating the rate of that, or performing any fine-grained breakdown of men-vs-women tweeting patterns here.

It’s like one of those bad logic questions. “Half of all slurs come from women and 20% of all slurs are used as abuse, what percentage of abusive slurs come from women?” To which the answer is “I don’t know, you need more data than that to reach a conclusion”.

When you factor in all the author-admitted caveats to the study (and there are quite a few, because all studies have caveats, but it says good things about the scientists that they mentioned them because not everyone does), what the study actually shows is:

  1. Men and women use those slurs in equal-ish amounts;
  2. Well, men and women in the UK;
  3. In early 2014;
  4. Without any drawable conclusions about whether they use them in an abusive or harassing manner at the same rate;
  5. And admitting that this is really a tiny microcosm of the forms that slur-driven online misogyny, let alone online misogyny generally, can take.

It’s a great paper for what it’s designed for, which is to look at language use as it relates to gender and media coverage. It’s a terrible paper for what the media seems to have decided to shoehorn it in for - the claim that half of online misogyny comes from women. But that’s not something where the paper draws deep conclusions, and so you probably shouldn’t cite it as an excuse to draw your own.