Twitter Bots Can Reduce Racist Slurs—if People Think the Bots Are White
NYU student Kevin Munger began his experiment by identifying 231 Twitter accounts with a propensity for using the n-word in a targeted manner (meaning, the message included the “@” symbol and used second-person language). All of these accounts were at least six months old and had used the n-word in at least three percent of their posts during the period Munger monitored them (late summer last year). Munger explains that he chose white men as the study’s subjects “because they are the largest and most politically salient demographic engaging in racist online harassment of blacks,” and also to control “the in-groups of interest (gender and race).”
He created fake Twitter accounts to target each of these users with a simple phrase, always in response to an apparently harassing use of the n-word: “@[subject] Hey man, just remember that there are real people who are hurt when you harass them with that kind of language.”
Enlarge / Munger’s flowchart of criteria for whether a Twitter account was racist enough for his study.
Munger’s fake accounts were puffed up with varying numbers of followers, which he purchased through a fake follower-bot service, and they were identified by names traditionally associated with white or black people, along with a cartoon avatar of a white or black man to match. Munger wanted to test a few things: whether the admonishing account’s race or follower count would draw a different response, and whether a user’s anonymity would influence their behavior. (Anonymity was scored based on whether a Twitter account had no real name, photo, or identifying information in either its profile or its posts.)
Munger’s data shows that a rebuke from an apparent white user with a high follower count had the most impact, and this impact carried more weight with the most anonymous Twitter users. In these cases, future posts containing the n-word dropped by roughly 27 percent compared to a control group in the following week. That drop-off leveled out somewhat in two-week and one-month follow-ups, but it remained. (As Munger puts it, “the 50 subjects in the most effective treatment condition tweeted the word ‘nigger’ an estimated 186 fewer times in the month after treatment.”)
More: Twitter bots can reduce racist slurs—if people think the bots are white