Yizhou Jin, University of Toronto
Preliminary draft, November 2026.
[Paper]
This paper studies how information asymmetry between producers and readers drives selection in the production and amplification of social-media posts, and how network structure shapes that selection. We first develop an empirical framework with selection on two margins: producers with heterogeneous accuracy and amplifiers with different demand for credibility. We then micro-found this demand in a network sharing game, linking it to two network features: reciprocity and overlap. Empirically, we collect viral posts about the 2022 mpox outbreak on Twitter and develop a large-language-model protocol that scales fact-checking of each post while aligning closely with human judgment. Our empirical tests demonstrate adverse selection on both production and amplification margins within reciprocal or high-overlap networks, and advantageous selection in one-way and diverse (low-overlap) networks.