“From Search to Marriage: Do Observable Scalar Rankings Suffice for Matching?” (with Pierre-Andre Chiappori, Yu Yang, Junsen Zhang)
Scalar indices based on observable traits are widely used in empirical matching models and often treated as sufficient statistics for search, equilibrium sorting, and downstream outcomes, but this assumption is rarely tested. Using height in China’s marriage market, we combine a large-scale randomized online marriage-platform experiment with nationally representative marriage data and show that women exhibit strong heterogeneity in search-stage height–income trade-offs, and that the experiment-implied index attenuates but does not eliminate assortative matching on height, with trade-offs sharply compressed in equilibrium marriages. Finally, household income displays strong non-separable interactions across spouses’ heights that are inconsistent with additively separable single-index outcome models, implying that the search-stage index is insufficient to rationalize equilibrium matching or income outcomes.
"Competitive Attitude and Partner Income: Gender Differences in Mating and Cross-Productivity Effects " (with Gahye (Rosalyn) Jeon) (under review)
The study examines how competitiveness impacts future incomes of individuals and their partners in Dutch households. Individual competitiveness is linked to higher incomes for single and partnered women, as well as partnered men, but not for single men. Women's competitiveness positively affects their partner's income, indicating gender differences in competitiveness effects on incomes.
"Heterogeneous Risk Aversion in Multi-Prize Promotion Contests" (with Zhouqiong Chen, Ella Segev, and Zekai Zhang)
We study a multi-prize all-pay auction in which contestants differ in both valuations and attitudes toward risk and loss, capturing promotion and high-potential selection environments where effort is sunk regardless of outcome. We characterize the unique mixed-strategy equilibrium and show that greater risk or loss aversion induces more aggressive bidding and higher winning rates even at lower valuations, implying that contests may reward psychological traits rather than long-run organizational value, with implications for promotion-system design.
“Mutual Certification as Endogenous Information Design”
In credence-good markets, experts may form associations that coarsen outcome information through voluntary “mutual certification.” I characterize equilibria in which such associations implement the same information partition a planner would choose under risk-averse demand, emerging when variance reduction outweighs pooling losses and dissolving as reputations become precise, yielding a unique, welfare-improving, and dynamically transient structure that links certification, licensing, and reputation networks.
“Virtue as Signal: Pro Bono Work and Trust in Credence Goods Markets”
Professionals often provide pro bono services, which may signal trustworthiness in credence-goods markets where clients cannot verify quality. I develop a model in which only belief-sensitive experts—who value esteem and feel guilt from disappointing others—use costly, observable pro bono work to separate from opportunists, sustaining trust and higher wages, while mandatory pro bono destroys this signaling value and can reduce average quality even as redistribution rises.
“Belief-Dependent Screening: How Disclosure Raises Bribery Efficiency”
Gifts are widely used to influence experts when reciprocation is unobservable and contracts are unenforceable, and evidence shows that increasing the salience of bribing intent reduces gift acceptance but increases reciprocity among those who accept. I develop a theory of psychological screening in which bribers manipulate beliefs about intent through gift size or explicitness, exploiting shame of acceptance and guilt of non-reciprocation to screen types and concentrate influence on the intensive margin, implying that disclosure and transparency policies may reduce visible corruption while increasing its efficiency by eliminating moral ambiguity.
“Pairwise Comparisons versus Likert Ratings: Structural Efficiency, Finite-Sample Behavior, and Evidence from Face Judgments” (with Zirui (Glenn) Fung)
Researchers collecting human judgments must choose between Likert ratings and pairwise comparisons, but their relative efficiency under realistic noise and finite response budgets remains unclear. We argue the difference is structural: Likert ratings estimate levels independently so uncertainty compounds when taking differences, whereas pairwise comparisons measure contrasts directly and reduce uncertainty through redundancy and path averaging, a mechanism we formalize in toy models and propose to test empirically using a large face-judgment dataset.