Munger (2017) seeks to understand how different social statuses and in-group and out-group effects might impact reducing racist harassment in online spaces. He sought to materialise his hypothesis as Twitter profiles. In-group and out-group characteristics were defined by the race of profile pictures. The social status was visible in the follower counts of Twitter profiles. In total, there were four profile pages in Twitter.
Following this, he developed an experimental procedure where the Twitter bot would react to racist slurs by tweeting the sender the following message: ` @[subject] Hey man, just remember that there are real people who are hurt when you harass them with that kind of languageâ. For experimental conditions, he rotated which of the four profiles responded to the tweet, thus creating four buckets.
To examine the impacts these bots had on user behaviour, he examined later tweets sent by the subjects in all of the four buckets and explored if their behaviour changed - if they used fewer racist slurs afterwards. He concluded that the results were tentatively true:
The primary prediction expressed in H1, that the In-group/High Follower treatment would cause the largest reduction in racist language use, was borne out. This effect was larger than either the In-group/Low Follower or Out-group/High Follower treatments, although these latter two reductions were not significant as expected. Overall, this is evidence of a multiplicative effect of the two treatments, as neither had an effect in isolation. I found evidence for both social identity theory in terms of in-group norm promotion and the theory that influential community members drive changes in normative group behavior.