I study collective outcomes in social networks, examining how network structure and strategic incentives shape belief formation, cooperation, and conflict. My research combines microeconomic theory, experiments, and data-driven approaches to understand when networks succeed in aggregating information and sustaining coordination, and when they fail.
Published
“Enhancing Resiliency and Promoting Prosocial Behavior among Tanzanian Primary-School Students: A School-Based Intervention” (with R. Berger, J. Benatov, R. Cuadros, and M. Gelkopf). Transcultural Psychiatry, 2018, 55(6): 821-845. [PDF] [DOI]
Children in Sub-Saharan Africa face chronic adversity but often lack access to mental health care. We evaluate a culturally adapted school-based intervention designed to enhance resiliency and prosocial behavior among Tanzanian primary school students. Students randomly assigned to the intervention showed significant improvement across all outcomes of social difficulties, anxiety, school functioning, and prosocial behaviors both post-intervention and at 8-month follow-up.
Working Papers
“Cooperation in a Dynamic Network” (with Nicole Immorlica, Brendan Lucier, and Brian W. Rogers). [PDF]
We study cooperation in evolving networks where agents play repeated prisoner’s dilemma games and can sever relationships. We characterize simple stationary equilibria in which cooperation and defection coexist even as players become arbitrarily patient, capturing the observation that many social networks exhibit high but not universal cooperation. Low levels of cooperation can be sustained only through exclusivity among cooperating agents.
“Belief Formation on Networks” (with Junya Murayama, Brian W. Rogers, and Xiannong Zhang). [PDF]
We design laboratory experiments to study how people combine interdependent information sources. About half of subjects solve problems where source values can be exactly deduced, but essentially none solve structurally similar problems with non-degenerate posteriors. Success in the former does not predict performance in the latter. Most subjects fail to identify even directional effects of signals, suggesting their success relies on fragile heuristics rather than robust understanding.
Work in Progress
“Learning Who to Trust: A Joint Inference Theory of Persistent Disagreement”
This paper offers an explanation for persistent disagreement based on joint inference rather than cognitive biases or non-standard preferences. When agents must simultaneously learn what is true and whom to trust, observed disagreement becomes rational evidence of low source reliability. This trust erosion fragments networks and prevents consensus, providing Bayesian microfoundations for belief fragmentation and information echo chambers in social learning.
“Rational Joint Inference: Experimental Tests of Prior-Based Belief Updating”
Two complementary experiments test whether prior-based inference reflects rational Bayesian updating when signal precision is unknown. Subjects must jointly learn about the state of the world and how informative their signals are. In this information environment, signal realizations that are more closely aligned with one’s prior are rationally perceived as more precise.
“Spectral Polarization: A Network-Aware Measure with an Application to Civil Conflict”
Existing measures of ethnic polarization reduce complex network structure to scalar functions of pairwise distances, fundamentally limiting what information about network topology can be encoded. This paper formalizes this insufficiency and introduces a new spectral measure, derived from the Fiedler vector, that captures structural features pairwise indices cannot. As an empirical application, the Spectral Polarization index significantly outperforms traditional ethnic polarization measures in predicting civil conflict.
“Seeding Adoption Cascades in Homophilous Networks”
Optimal network seeding typically requires detailed individual-level data rarely available to policymakers. This paper develops a targeting framework for homophilous networks that exploits group-level structure and the timing of local cascades, achieving substantial cost reductions over standard one-shot seeding using only aggregated data. An empirical implementation uses Facebook’s Social Connectedness Index, which provides the group-level network structure the framework requires.
“The Effects of Media Sentiment on Product Sales: Evidence from the Soft Drinks Industry” (with Sumedha Rajbanshi).
We estimate the effect of media sentiment on soft drink sales in New York State. Using natural language processing to analyze media coverage of Coca-Cola and PepsiCo (2006-2012), we find lagged sentiment significantly predicts sales, with stronger effects for core products bearing the firm name. These findings suggest that brand salience moderates consumer response to reputational information.
