Do Users Ever Draw a Line? Offensiveness and Content Moderation Preferences on Social Media

Abstract

When is content on social media offensive enough to warrant content moderation? While social media platforms impose limits to what can be posted, we know little about where users draw the line when it comes to offensive language, and what measures they wish to see implemented when content crosses the boundary of what is deemed acceptable. Conducting randomized experiments with over 5,000 participants we study how different types of offensive language causally affect users' content moderation preferences. We quantify causal effects of uncivil, intolerant, and threatening language by randomly introducing these aspects into fictitious social media posts targeting various social groups. While overall there is limited demand for action against offensive behavior, the severity of the attack matters to the average participant. Amongst our treatments, violent threats cause the greatest support for content moderation of various types, including punishments that would be viewed as censorship in some contexts, such as taking down content or suspending accounts.

Partisan conflict over content moderation is more than disagreement about facts

“…In this study, we focus on political headlines, and more specifically those denigrating out-party politicians instead of flattering in-party politicians. Future research should investigate further how other types of content-political misinformation flattering in-party politicians, nonpolitical misinformation, hate speech, voter suppression content-and different contexts influence the preference gap and party promotion (67). Another potentially interesting research question is to what extent individual-level drivers of content moderation are decisive at the level of content moderation systems with thousands of often professional content moderators, and which other factors might be at play in those systems.…”