Abstract

Is platform self-regulation of social media effective in moderating online content? Or does it shift abusive posts to darker corners of the internet? This paper addresses these questions by analyzing the effect of social media regulation from a cross-platform perspective. I exploit a policy of Twitter against racist hate speech and investigate whether regulation is effective in curbing this harmful content or whether it shifts abusive content to unregulated platforms. I exploit synthetic difference-in-differences and regression discontinuity in time to assess the effects of Twitter's policy. I find that hate speech and racist hate speech content decrease on Twitter after the policy, but there is no apparent increase in hate speech and racist hate speech on Parler for users that have accounts on both platforms. I finally exploit word embedding to analyze whether the language used for abusive content changes after the policy.