Telling People to Be Less Racist Online Works, Sometimes

As anyone who has spent much time online could tell you, on the internet, people feel free to do and say terrible things that would be a lot more difficult to do offline. Certain social norms just don’t seem to extend, with their full strength, into the online world, and figuring out why has been a major project for some researchers — especially as the harassment problem on platforms like Twitter has gained more and more visibility.

The most straightforward way to explain the problem is that when people are immersed in online communities, particularly toxic ones, they sense and react to a different set of social norms than they do when they’re out in the world. It is unfortunately the case that in many corners of the online world, you can hurl racial slurs with impunity — with no real damage to your reputation and no sense that you are violating the expectations of others — in a way you can’t offline. (I’d of course be remiss to not point out that since last week, there’s been an uptick in this sort of behavior offline as well.) So changing people’s behavior for the better, which is never an easy thing to do, can be even more difficult online, given all the noxious incentives and norms that abound there.

One possible avenue for nudging people to act better involves addressing those social norms: finding ways to make it clear to the perpetrators of online harassment and their racism that their behavior isn’t acceptable, that the broader community will socially sanction them if they continue to engage in it. In a new paper in the journal Political Behavior, Kevin Munger, a Ph.D. student at New York University’s politics department, offers a test of this theory.

For his study, which is cleverly titled “Tweetment Effects on the Tweeted” — a play on words of treatment effects — Munger used Twitter-scraping tools to come up with a list of Twitter accounts operated by white users who had used the word n*gger to harass black Twitter users, and then culled the list to include only those who had engaged in the most offensive tweeting, as defined by a list which included other slurs as well.

Then, Munger deployed some bots to try to tamp down the offensive behavior. There were four categories of bots: Half appeared to be operated by a white user (based on a cartoony avatar of the type commonly seen on Twitter), and the other half by a black user. Half the bots had a low number of followers (zero to ten) and half had a relatively high number (500 to 550). Munger was curious how these variables would affect the ability of the bots to tweak the abusive users’ behavior; in many real-world settings, humans respond the most to social sanctions from members of their in-group, and particularly to high-status members of it.

The bots, which Munger made look like actual accounts run by humans, responded to one of the offensive tweets with the line, “Hey man, just remember that there are real people who are hurt when you harass them with that kind of language.” As it turned out, the intervention worked — the targeted accounts became less abusive. “The effect persisted for a full month after the application of the treatment,” Munger writes. Overall, he believes the intervention “caused the 50 subjects in the most effective condition to tweet the word ‘n****r’ an estimated 186 fewer times after treatment.”

Now, there’s a pretty big but. The intervention only worked when the bot reprimanding the abusive Twitter personality appeared to be operated by a white user, and appeared to have a relatively high number of followers. This tracks with the idea that sanctions from prominent (or somewhat prominent) members of in-groups are the most effective. It’s the same logic the authors of a study on a promising anti-bullying intervention used in New Jersey middle schools: By mapping out schools’ social networks and targeting the most connected individuals, it appears they were able to “seed” the networks with anti-bullying norms that then spread, helping to tamp down on harmful behavior.

This isn’t the exact same, because a Twitter user with 500 followers you’ve never heard of before isn’t “prominent” in quite the same way. But still, in many cases the abusive twitter users targeted by Munger’s study may just have never received much pushback for their behavior. Many of them probably hang out with (online), and retweet, similarly hateful personalities. It’s useful information to know that a simple tweet appeared to put the brakes on their behavior a bit.

One interesting question is how this paper fits into the broader debate over online shaming and callout culture. There’s some evidence from sociology and political science that if you aggressively hector people for acting like horrible human beings, they will either double down on their behavior or ignore you. But Munger explained in an email that the phrasing of his intervention was “designed to be humanizing and to push back on the group norm (of the groups my subjects tended to belong to) of using this kind of language.” The bots didn’t yell at people for being racist — it just reminded them that there’s a broader community out there that finds such behavior unacceptable.

As Munger explained:

I think group norms/identity are real mechanism behind online incivility and harassment: the physical distance and anonymity enable people to be nasty in ways that are far more difficult (in terms of emotional cost) offline, and group speech norms are extremely important because they’re the main way of signalling group identity. More extreme groups are willing to use more extreme language, which is an effective identity signal precisely because it’s offputting to most people. || So, my work is trying to test ways of priming people’s offline identities and thus reminding them that what they’re doing (using extreme language to make people feel bad) violates norms of behavior except in the online communities to which they belong.

It’s that distance between online and offline personas that drives so much nastiness: All else being equal, the more that gap can be bridged, the less awful people will act online.

Telling People to Be Less Racist Online Works, Sometimes