Research on gender and political communication consistently shows that men and women politicians differ in how they present themselves to the public. Yet most studies focus on election campaigns or traditional media, where communication is mediated by party strategies, journalistic norms, and gender-equality pressures. These studies usually rely on limited samples of politicians, and small-N manual coding, offering only a snapshot of gendered communication under the heightened visibility of campaign periods. As a result, we know little about whether such patterns persist in the routine communication of politicians - when external constraints are weaker but when most public exposure to political discourse actually occurs (Aaldering & van der Pas, 2020). This paper addresses this gap by examining gender differences in how elected representatives use social media during non-campaign periods.
Building on theories of gendered political socialization, reinforced by role congruity expectations, we argue that women politicians face stronger normative constraints against overt aggression and incivility (Schneider & Bos, 2019). We therefore expect that women are less likely than men to engage in negative discourse overall, but that gender differences also emerge in how negativity is expressed. When women do go negative, they are expected to favor issue-based, policy-oriented criticism over personal or trait-based attacks. Extending insights from issue ownership and gendered issue framing, we also expect women to emphasize soft-policy over hard-policy areas and to express negativity primarily within soft-policy areas where they hold greater perceived credibility (Eagly & Karau, 2002).
Empirically, the study analyzes the complete population of tweets -nearly 250,000 posts- published between May 2019 and April 2023 by all 425 elected members of the Belgian federal, Flemish, and Walloon parliaments. This comprehensive dataset enables an unprecedented longitudinal analysis of political communication in a multi-party, gender-balanced system beyond the US–UK focus of existing research. Using large language models (LLMs) for computational text coding, we classify tweets by attack behavior, incivility, attack type, and issue domain, producing fine-grained measures of political tone and issue focus. This enables a systematic and generalizable account of how gendered expectations shape both the likelihood and form of negativity in everyday digital politics.
