A Model of Polarization on Social Media

Lower linking costs don't automatically reduce polarization: The model shows that making it easier for users to follow others can paradoxically increase exposure to extreme content and heighten polarization, contradicting the intuition that connectivity reduces division.

Moderate voices are crucial but fragile: When linking costs become very low, moderate content creators are discouraged from participating altogether, which actually worsens polarization. This suggests platforms need to actively support balanced viewpoints, not just remove barriers to connection.

Exposure-based interventions have limited effectiveness: The paper explains why common platform policies focused on controlling what users see (like algorithmic feed changes or attention reallocation) produce "ambiguous effects on polarization"—they don't address the underlying problem of who chooses to create content.

Incentivizing content creators matters more than managing links: Policies that encourage certain types of content production generate "large welfare gains," suggesting platforms should focus on who participates rather than just how users connect or what they're exposed to.

Algorithmic bias isn't the only polarization driver: By developing a model without algorithmic bias, the authors demonstrate that polarization emerges from basic structural features like linking costs, meaning some polarization problems persist even on "neutral" platforms.