Who gets targeted? Intersectional patterns of hate and toxicity in social media

This study examines the prevalence, intensity, and diffusion of toxic messages across four social media platforms (X, Facebook, Instagram, and TikTok), focusing on five discursively stigmatised target groups: migrants, Muslims, Roma people, Jews, and LGBTI people. Attention is given to posts that co-mention two or more of these groups within the same message. Throughout the manuscript, we refer to these cases as “intersectional patterns” in the specific sense of multi-target co-targeting in discourse—this message-level operationalisation should not be interpreted as identifying the intersectional social position of a specific individual. Based on a corpus of 798,619 Spanish-language messages published between April 1 and August 13, 2024, automatic toxicity detection tools (Perspective API) were applied to assess six dimensions of hostile language, and engagement was measured using likes and comments. Results show that multi-target posts are associated with higher toxicity intensity in continuous outcomes on Facebook, Instagram, and TikTok, while corresponding estimates for X are small and not statistically distinguishable from zero in covariate-adjusted intensity models. For threshold-based outcomes, multi-target posts have substantially higher odds of exceeding toxicity cutoffs on Facebook and Instagram, whereas effects are mixed on X and imprecise on TikTok depending on the operationalisation. Descriptively, X exhibits the highest average toxicity scores and the largest share of posts exceeding TOXICITY ≥0.5. Engagement differences among toxic posts are platform-contingent: Facebook shows consistently lower engagement for multi-target toxic posts, and X shows a selective increase in comments (but not likes or total engagement) under an inclusive toxicity definition. These findings underscore the need for intersectionality-sensitive approaches to understanding and measuring hostile discourse across platforms.