Online Harassment: Citation Needed
May 3, 2016
Articles about online harassment have made a resurgence in the last couple of weeks, with social media users sharing data gathered two years ago about online harassment. This research—which analyzed whether men and women experience online harassment differently—was controversial at the time. And with the awareness of online abuse gaining more momentum, readers are seeking solutions. Yet, some of these articles fail to properly understand the very data they report on, and some of that data is not as useful as it seems.
In the last couple of years, we’ve seen articles claiming that men are harassed more than women online, that abuse of men and women can’t be compared because they are contextually distinct, that abuse of men is categorically underreported, or that men are more likely to be verbally abused, but women are more likely to be the victims of serious threats of violence and stalking. Some of these articles are well-written and clearly researched, so I mean no indictment of them as a whole, but with such drastically different conclusions based on a small handful of the same studies, something is up in Journalism Town.
And the cycle only gets worse on social media. The Verge reports that Upworthy and Buzzed have both concluded that a large portion of readers tweet out a link when they’ve only read about a quarter of an article. Add your best friend’s cousin’s Facebook analysis, and the original data is lost in an endless stream of clicks and half-truths.
Click on any of these oft-cited articles, and you will eventually be led to two sources that are most commonly cited: one is a study out of the research outfit Demos, which describes itself as “Britain’s leading cross-party think tank.” The other is a piece of Pew research, which the organization calls its “first in-depth study of online harassment among young adults.” Pew has long earned its reputation as a thorough and effective research organization, so let’s begin with what their research actually says.
In her overview, “5 facts about online harassment,” Pew writer Maeve Duggan writes, “Online men are more likely to experience at least one of the six types of harassment we queried – 44% have had some sort of harassment experience compared with 37% of online women. Men are somewhat more likely to experience certain ‘less severe’ kinds of harassment like name-calling and embarrassment. They are also more likely to receive physical threats online. Women – and particularly young women – are more likely to experience certain types of ‘more severe’ harassment, such as stalking and sexual harassment. Among female internet users 18-24, 26% say they have been stalked online and 25% have been sexually harassed.” Duggan is a research analyst who worked on the report itself.
Putting aside that surveys are, by their nature, self-reported, this report clearly does not indicate that “Men are harassed more than women online.” It would be more accurate to say, “Men are more likely to have been harassed, at least a little, at least once, than women online.” This is, of course, very different from a claim that men bear the brunt of the aggregate abuse online.
The other source, a study of harassment released by Demos, comes closer to that claim. In this study, Demos looked at tweets sent to male and female celebrities on Twitter, and found that “2.54% of the tweets containing the @ username of male public figures contained abuse, compared to only 0.95% of the tweets received by prominent women.” The Daily Beast reported that this study found that “men are harassed more than women online.” But this is a serious overreach.
First, the research took into account a very small subset of the online population (British celebrities). Second, the method of identifying “abusive” messages was incredibly limited. In order to set off the researchers’ abuse-spotting algorithm, the Tweet had to contain a swear word from Google’s search language filter. That list includes (sorry, Mom) words such as “boner,” “crap,” “damn,” “fag,” “bitch,” and four spellings of the word “boobs.” It also includes questionable choices like “LMFAO” and “pawn.”
I wrote to Demos and asked them if this is not an inherently limiting way to look for abuse. For one thing, this system would tag a Tweet saying, “Carrie, you’re the best damn writer in the world” as abusive, but not tag one that said, “Carrie, I want to come to your home and murder you” (“murder” is not on the list; neither is “rape,” “assault,” “maim,” “death,” or “kill”).
Rob Macpherson, a communications officer for Demos responded, acknowledging the issue with false positives:
There are two stages to the filtering process. First we collected tweets that contain one or more of the words in the list. Then the researchers went through and read hundreds of tweets to judge which ones were using the language in an abusive context, and which ones were in a non-abusive context. To use your examples the “best fucking writer on the planet” one would have clearly been classified as not abusive. (emphasis added)
This is good, but still leaves the entire question of what is abusive up to the subjective impressions of the research team. And of the two million tweets the system analyzed, “hundreds” of them would be a small subset indeed. Further, it ignores all the abusive messages that don’t happen to contain Google’s classified swear words.
I asked Macpherson what he thought of headlines such as The Daily Beast’s, using Demos research to conclude that “men are harassed more than women online.”
“I personally think headlines will always provide a simplified synopsis of deeper research,” he said, “and interested readers would always benefit from reading the full story, or even better investigating the analysis themselves.”
Macpherson could not be more right. No study is perfect, and any study will face its limitations. This is why we rely on reporters to analyze and accurately report on research, going beneath the headlines and press releases. Of the aforementioned articles, only one (at Forbes) delved into the methodology behind the Demos study, and asked good questions. As writer Danielle Citron notes, “Typically, cyber harassment is understood as repeated online speech constituting a ‘course of conduct’ that causes substantial or severe emotional distress, not a one-off comment.”
It is also worth noting that none of the articles mentioned that the vast majority of the abusive tweets tagged by the Demos study came from men. Still, many reporters are quick to point out that a previous study showed that women are almost as likely to use the words “slut” and “whore” as men, again ignoring a whole slew of words that add up to harassment. The study also fails to ask how this stacks up to gendered language flung at men. It goes without saying, I suppose, that gendered insults are almost exclusively leveled at women; the only question left is who to blame.
The question of online harassment is one that will be answered only with rigorous research and honest analysis. It appears that much more is needed. Until then, be nice out there.