Misinformation and harmful language are interconnected, rather than distinct, challenges.

Abstract

There is considerable concern about users posting misinformation and harmful language on social media. Substantial-yet largely distinct-bodies of research have studied these two kinds of problematic content. Here, we shed light on both research streams by examining the relationship between the sharing of misinformation and the use of harmful language. We do so by creating and analyzing a dataset of 8,687,758 posts from N = 6,832 Twitter (now called X) users, and a dataset of N = 14,617 true and false headlines from professional fact-checking websites. Our analyses reveal substantial positive associations between misinformation and harmful language. On average, Twitter posts containing links to lower-quality news outlets also contain more harmful language (β = 0.10); and false headlines contain more harmful language than true headlines (β = 0.19). Additionally, Twitter users who share links to lower-quality news sources also use more harmful language-even in non-news posts that are unrelated to (mis)information (β = 0.13). These consistent findings across different datasets and levels of analysis suggest that misinformation and harmful language are related in important ways, rather than being distinct phenomena. At the same, however, the strength of associations is not sufficiently high to make the presence of harmful language a useful diagnostic for information quality: most low-quality information does not contain harmful language, and a considerable fraction of high-quality information does contain harmful language. Overall, our results underscore important opportunities to integrate these largely disconnected strands of research and understand their psychological connections.