When, Where, and With Whom We Polarize
Contextual Drivers of Ideological Signaling on Social Media

Political polarization in online spaces is often analyzed in terms of the ideological distance between interacting users. Yet, less is known about whether users reveal their ideological identities in their interactions, which limits our understanding of how polarization in online spaces is perceived and/or experienced by users. This study offers analyzes how users express---either intentionally or unintentionally---their ideological leanings on social media and how these cues are perceived by others. To measure the prevalence and strength of ideological signals, we analyze hundreds of millions of comments and likes from millions of users across hundreds of public Facebook pages during the 2016 U.S. election cycle (January 2015–February 2017). We fine-tune large language models to infer users' symbolic ideology from short-form comments, and calibrate the resulting predictions with human-annotated survey data, enabling us to construct a measure of human-perceived ideological signals embedded in text. We find that ideological signaling is pervasive in public discourse, even in ostensibly nonpolitical discussions. These signals intensify in the lead-up to elections, decline afterward, and resurface around major political events. Unlike offline settings where overt displays of partisan cues are discouraged, online interactions reveal a reversed pattern: strongly ideological messages--whether from co-partisans or opponents--elicit heightened ideological signaling in replies. Together, our study highlights the central role of ideological signaling in understanding online polarization and demonstrates a scalable, language model-based approach for measuring it from social media content.