The Polarization Paradox: Why More Connections Can Divide Us

Abstract

We develop a simple model of content filtering — the tendency of individuals to selectively forward information that aligns with their ideological preference — to study how network structure shapes the distribution of political content. In our framework, individuals and content are horizontally differentiated into three types (left, middle, right). We show that content filtering can amplify the middle or the extremes and may result in only centrist content (full moderation) or only extreme content (full polarization). The outcome depends on the interaction between two forces: a preference advantage from the relative prevalence of types in the population, and a pairwise comparison advantage that systematically favors centrist content. Network density plays a critical role. Sparse networks robustly yield moderation, even when extreme types dominate the population, while dense networks replicate the population's type distribution. Intermediate densities generate non-monotonic comparative statics, including sharp transitions between moderation and polarization. These findings complement existing empirical results that emphasize the types of connections individuals have on social media by highlighting how the number of connections, holding their composition fixed, may fundamentally shape the information environment in ways that foster/mitigate populism and polarization. Joint with Arthur Campbell, Matt Leister, and Yves Zenou.