|Mathew Ingram , CJR|AIWA! NO!|Any woman who has spent more than a couple of hours on Twitter has likely experienced some form of casual harassment or even outright abuse—and these attacks are even more likely if the woman in question happens to be a journalist or a politician.

On Tuesday, a new report from Amnesty International put some numbers behind those kinds of incidents, and the numbers are not good.

According to the study, which analyzed hundreds of thousands of tweets sent in 2017, female journalists and politicians were subjected to some kind of harassment or abuse on the social network roughly every 30 seconds, and women of color experienced significantly higher levels of abuse: they were 84 percent more likely to be mentioned in abusive or harassing tweets. (The full results, part of Amnesty’s Troll Patrol project, are available here.)

Amnesty says the report, which was conducted with artificial-intelligence software company Element AI, is the largest ever study of the way women are harassed online. The two organizations asked 6,500 volunteers from 150 countries to look through almost 300,000 tweets sent to 778 politicians and journalists in the US and the UK. Politicians were selected from all the major parties, and the journalists came from a wide variety of publications, including The Daily Mail, The New York Times, and The Guardian, as well as news sites such as Pink News and Breitbart. “We have the data to back up what women have long been telling us—that Twitter is a place where racism, misogyny, and homophobia are allowed to flourish basically unchecked,” Milena Marin of Amnesty told The Financial Times.

Tweets were categorized as abusive if they fit Twitter’s definition of abuse, including content that promotes violence or hatred based on gender, race, or ethnicity. (Some of the tweets were later removed, Amnesty said. The study also included tweets that were defined as “problematic,” meaning they contained hurtful or hostile content that reinforced negative or harmful stereotypes about a group of individuals based on race or some other criteria. Amnesty acknowledged that these tweets “may qualify as legitimate speech,” but argued that they can still have the effect of silencing an individual or groups of individuals, and that including them was necessary to highlight “the breadth and depth of toxicity on Twitter in its various forms.”

Amnesty also pointed out that Twitter has often talked about trying to promote “healthy conversation,” and therefore it’s helpful to know the full extent of problematic content that flows through the service. And the study was based on public data provided by Twitter for 2017. By definition, that database would not include any tweets that were deleted for abuse before that date, which suggests that the actual rate of abuse and harassment could be significantly larger. Vijaya Gadde, the head of trust and safety for Twitter, said in a statement seen by Wired that any kind of abuse detracts from the health of the service, and that the company is “committed to holding ourselves publicly accountable toward progress in this regard.”

Twitter CEO Jack Dorsey has said publicly that he wants to improve the civility of discussion on the platform, after years of criticism that the service spent too little time thinking about abuse and harassment and too much time focusing on growth metrics like “engagement.” The Amnesty study shows that there is still a long way to go before female journalists and politicians are likely to see their experience on Twitter as civil or healthy.

Here’s more on Twitter and its fight against abuse:

• Transparency: Last week, Twitter released its regular “transparency report,” and for the first time included detailed statistics on enforcement of its rules around abuse and other violations of its terms of use. It said more than 6 million accounts were reported for possible violations from January to June 2018.
   
• Learning: Twitter has been working with a machine-learning research outfit called Cortico, which grew out of MIT’s Media Lab, in an attempt to find a way of detecting abuse before it is flagged by users. Casey Newton of The Verge spoke with one of the founders of Cortico about that effort.
   
• Engagement: The problem of abuse on Twitter is similar to the problem the company has with trolls and fake accounts, and those who have followed the service from the beginning say it has often downplayed both of these issues because it has been more focused on growth and engagement.
   
• An old story: In 2015, the nonprofit group Women, Action and Media was given access to Twitter’s abuse-reporting system for several weeks and flagged hundreds of tweets. It said 27 percent included hate speech, 12 percent included threats of violence and 67 percent said they had notified Twitter more than once.
   
• Reporting while female: For female journalists, the rampant abuse and harassment on Twitter is just a digital version of the kind of attacks they often suffer in the offline world. Anne Helen Petersen wrote for CJR about the problems journalists can suffer when “reporting while female.”