willowbl00

The process of reporting and responding to online harassment is the least understood and arguably most important part of the problem. Most of what we know about receiving harassment comes from the small number of people who are courageous enough to face the risk of telling their personal stories.

Maeve Duggan's recent Pew report about online harassment helps us understand the experience of harassment across a nationally-representative sample, but efforts by companies to address online harassment are still mostly secret. For all sorts of safety, legal, and business fears, companies are reluctant to reveal the details of how they enforce their policies around harassment and hate speech. So when Women, Action, and the Media approached me to analyze Twitter harassment reports collected over three weeks with the informed consent of participants, I knew it was a unique opportunity to grow our understanding of the problem.

• Read our report: Matias, J. N., Johnson, A., Boesel, W. E., Keegan, B., Friedman, J., & DeTar, C. (2015). Reporting, Reviewing, and Responding to Harassment on Twitter. Women, Action, and the Media. May 13, 2015.
• You can also read WAM!'s recommendations

Together with a truly amazing team, we were able add to the conversation:

• user stories for designers on the people and situations involved in harassment reporting
• summary statistics on who reports, including bystanders and delegates
• summary statistics on what was reported
• statistical tests of what kinds of harassment Twitter handled best and worst at the time
• findings on systematic problems that need to be addressed to better support people facing harassment
• a summary of the work and risks experienced by reviewers of harassment reports

The project, which was independent of any of our institutions, was quasi peer reviewed through a double blind revise and resubmit process. Our team spent many late nights and weekends on this project, and I'm proud of what we have achieved.

Responding to Questions about our Study

Many people online have written with questions and ideas coming out of our study. I've heard from:

• Community & safety teams at several platforms have reported finding it helpful. I heard from staff at one major platform that our section on the work of dealing with harassment was especially helpful, since people usually come to them only to complain.
• NGOs offering support for harassment receivers have found it helpful for thinking through their own initiatives.
• Designers are finding our user stories helpful as they think about designing related systems
• Groups engaged in wider arguments about online harassment, including people embroiled in the GamerGate controversy.

Especially among the GamerGate controversy, people have clamored for us to clarify some points from our report. I'll admit that I'm hesitant, after some GamerGate supporters flooded my Twitter (one later apologized) during my PhD qualifying exams. But I'm the lead author of this report, and I feel a responsibility to all of my readers, so here goes...

Q: Is The WAM! Study Representative of All Online Harassment Everywhere?

Our findings apply to the reports that were sent to WAM! in those three weeks. Our findings should not be taken as representative of some specific sub-group of Twitter, or even all of Twitter. We state this very clearly in our appendix, in great detail, but there has been confusion and misinterpretation, so I want to be very clear.

There are four kinds of evidence that can be used to understand online harassment, and they all have different strengths/limitations:

• stories and collections of the harassment that individuals have received over time, like Anita Sarkeesian's archive of one week of harassment. Stories from notable individuals help a wide audience understand what the experience *can* be like, emboldening many others to say that they share the experience. Those stories and examples are important evidence and should be given our full attention. Since an individual's experience cannot be considered statistically representative, the voices of others who say they experienced similar things offer a powerful and important argument for how widespread that kind of harassment might be.
• nationally representative survey techniques like Maeve Duggan's Pew report offer a statistically representative picture. When Maeve writes that "8% [of Adult Internet users in the U.S] have been physically threatened" online and that "7% have been harassed for a sustained period," she's making a claim that's representative of the US adult population. It's then possible, for example, to extrapolate from the size of the population to conclude that within Pew's confidence intervals, around 6.7 million adult Americans have been physically threatened online. However, Pew's study can't offer any visibility into the harassment reporting, reviewing or responding process.
• reports to platforms or moderators. Most platforms decline to report even how many issues they handle, which makes it incredibly difficult to get a sense of how large the problem is, or how well companies are doing. By treating it as a trade secret or PR risk, platforms often hold back our ability to understand and address this problem. The one concern I *do* find compelling is the privacy risk of publishing this material.
  • Some platforms, especially those that rely on community support and which use pseudonyms, are more open. Wikipedia publishes the list of users who have been blocked, along with reasons. Moderators on Reddit often publish transparency reports to their subreddits, and have even developed tools to automatically publish moderation decisions. For the most part however, this kind of information is kept private by platforms.
  • This data is only representative of what people choose to report to companies. If people don't know they can report it, if they're worried about the personal consequences of reporting (in a domestic violence situation, flagging a Tweet might make things worse), or if they don't think the platform can help, then the platform will never see the report.
• service prividers like WAM!, who offer help to receivers of harassment. There's a long history of research by orgs that offer support for victims of violence against women, and WAM! was adapting that tradition to social media. I wrote about this approach in my Atlantic article on mutual aid accountability in digital labor and environmental conservation. This data cannot support country-level or platform-level claims about people's experience of harassment. However, WAM!'s data does offer an inside view on the experience of reporting, reviewing, and responding to harassment, and our large sample compared to individual stories helps strengthen our understanding of this issue in aggregate. Finally, we *were* able to use this data to do statistical analyses of Twitter's handling of harassment reports.

Q: What Can Be Learned about GamerGate from The WAM! Study?

Very little. Since WAM! only had data about alleged harassment that was reported to it, our findings cannot be used to make any representative claims, positive or negative, about any specific subgroup online, including Gamergate. Did people submit reports to WAM! of alleged harassers that included user accounts that are allegedly associated with GamerGate? Yes. Did WAM! reviewers consider some of those cases to be worth escalating to Twitter? Yes. Does that mean that GamerGate is a movement dedicated to harassment? Our evidence can't answer that question and shouldn't be used to do so.

Many people have misunderstood the reason and findings of our GamerGate analysis. I can understand that; we had a subtle point to make, and since it was a minor point, we put our explanation in the appendix.

Here's why we included a GamerGate-related analysis: We expected that some people who want to ignore the issue of online harassment might look at WAM!s work in November and dismiss it as just part of the GamerGate controversy and therefore not important. We were also curious ourselves: since our sample was dependent on who chose to report to WAM!, what would it mean if most of the reports were associated with that one controversy? Was online harassment on Twitter a passing issue related to a particular conflict, or something deeper and longer?

To test this question, we used two data sources to look for GamerGate-related accounts among alleged harassers: a set of all #GamerGate tweets in December 2015, and the list of accounts on Randi Harper's ggautoblocker. Neither of these lists is statistically representative of supporters of the "GamerGate" movement -- finding representative samples of online social movements is a challenge no one has solved yet. But we thought they might offer a good rule of thumb.

When we saw how small the overlap was between these imperfect lists and our list of alleged harassers (12%), and when we saw how many people reported experiencing long-term harassment, we became confident that our findings were not limited to just this one controversy.

I want to be very clear here. Although our data does identify many reports of alleged harassment by accounts that have tweeted the #GamerGate hashtag or appeared in the ggautoblocker list at some time, our findings cannot condemn or vindicate the GamerGate movement. WAM! and its reviewers are free, if they wish, to make claims about GamerGate from their experience reviewing harassment (I don't think they have). But the research team has not investigated whether reports against alleged GamerGate participants were valid or not. We also haven't looked into whether any reported accounts were genuine participants in the GamerGate movement. Anyone who bases arguments specifically about GamerGate on our report is misinterpreting our findings. 

I have received several detailed questions from people who hope to use math to extract claims about GamerGate, who worry about our samples, or who vigorously question claims they believe we have made about GamerGate. Consider this my answer; it can't be done, and we didn't attempt it in our report.

I hold complicated, semi-appreciative feelings about people who have pointed this out, even when their critiques sometimes vigorously (and I think wrongly) demeaned our abilities and our work. While it's not pleasant to be berated for claims I never made, our detractors were at least trying to dissuade people from applying our findings where they do not apply.

Other Questions? If you have another question, please tweet your question to @natematias.

What's Next For Us?

Although the authors have moved on from this project, we're continuing to do related work. Charlie De Tar is working on ways to support people to learn to be good allies of those experiencing harassment. Whitney, along with Susan Benesch and Bruce Schneier, co-organized an international workshop on misogyny and the Internet. Amy is also working on projects related to our report.

I'm personally starting to focus on the issues of peer support and community moderation:

• Advising the HeartMob team to evaluate the effect of their peer support system on survivors and supporters
• Researching the work of moderators on Reddit (here's my update on the moderator blackout of July 2015)
• Leading a collaboration to develop a literature review about online harassment that goes beyond just tech and law

If you have further questions or are interested in any of my current projects, especially the lit review or the Reddit study, please tweet your question @natematias. I'm tied up with research this week, but I will try to respond as much as I can.

Hover Houses: 12 Cliff-Clinging Homes with a View

Shop in a Swimming Pool: Neglected Space Turned into a Store

Extreme Cliff Living: Modular House Dangles Precariously