Negativity vs. Incivility: How We Engage with Partisan Content on Facebook Over Time

Social media algorithms facilitate the visibility of content aligned with users’ beliefs, a phenomenon further amplified by individuals who consume and share similar content. This study investigates the dynamics of affective language on 1,493 U.S. Facebook partisan Pages, focusing on social engagement (e.g. comments, shares, likes, and other reactions) over a 13-year period. Linguistic attributes of negativity and incivility were identified using LLM-based classifiers and validated through human review. The findings showed that negativity and incivility were associated with distinct engagement patterns. In static models, negativity was more strongly associated with expressive reactions such as “Angry,” but corresponded to lower expected levels of approval-oriented and amplification-oriented engagement, including liking, “Love,” and sharing. Over time, the standardized association between negativity and most engagement types attenuated, with trajectories varying across engagement metrics. In contrast, incivility was more strongly and consistently associated with higher expected engagement across multiple behavioral signals and maintained relatively stable associations over time. Comparative analyses highlighted how different forms of affective language were differentially rewarded through engagement across Facebook’s platform evolution, underscoring how engagement-based ranking systems may reinforce distinct communicative styles. The discussion considers implications for user behavior, platform incentives, and democratic discourse.