More Polarized, Less Toxic: How Electoral Discourse on Instagram Shifted Over Four Years in Brazil

Most studies of electoral discourse on social media examine a single campaign, leaving changes across election cycles poorly understood. We compare political discussions on Instagram during the 2018 and 2022 Brazilian presidential elections using more than 5 million comments and complete post metadata from prominent political profiles. We examine engagement, sentiment, toxicity, and topic dynamics over matched campaign periods. Although the three profiles observed in both elections gained substantially more followers, likes and comments per follower declined, while overall comment activity became more concentrated around the two election rounds. In 2022, negative comments increased from 36.3% to 49.5%, while neutral comments fell from 26.4% to 10.4%, a pattern consistent with stronger affective polarization. At the same time, median scores decreased across six harmful-language attributes, particularly insult and severe toxicity. Topic analysis shows continuity around religion, democracy, regional ideology, and election fraud, alongside greater attention to civil rights and environmental issues in 2022. These findings show that indicators of affective polarization and explicit toxicity need not move together and demonstrate the value of cross-election analysis for understanding platform-mediated democratic debate.