Publications:
School Choice Under Complete Information: An Experimental Study.
Yan Chen, Yingzhi Liang, Tayfun Sönmez. Journal of Mechanism and Institution Design (2016)
We present an experimental study of three school choice mechanisms under complete information, using the designed environment in Chen & Sönmez (2006). We find that the top trading cycles (TTC) mechanism outperforms both the Gale-Shapley deferred acceptance (DA) and the Boston immediate acceptance (BOS) mechanism in terms of truth-telling and efficiency, whereas DA is more stable than either TTC or BOS. Compared to the incomplete information setting in Chen & Sönmez (2006), the performance of both TTC and BOS improves with more information, whereas that of DA does not.
A Dynamic Matching Mechanism for College Admissions: Theory and Experiment.
Binglin Gong, Yingzhi Liang. Management Science (2025)
Market design has provided many managerial insights into why certain market institutions fail while others succeed in allocating scarce resources in both the for- and non-profit sectors. In this paper, we analyze a new form of dynamic matching mechanism enabled by innovations in information technology. We provide a theoretical and experimental examination of this mechanism in the context of college admissions in Inner Mongolia, China, where students are given real-time allocation feedback and are allowed to revise their choices. Theoretically, we show that efficient and stable outcomes arise in every rationalizable strategy profile if there is a sufficient number of revision opportunities. Experimentally, we find that in an environment with high strategic complexity, the Inner Mongolia Dynamic mechanism performs better than theoretical predictions: It is as stable as the Deferred Acceptance mechanism and as efficient as the Boston mechanism, with higher truth-telling rate than both of them. These results suggest that the Inner Mongolia Dynamic mechanism can be a good substitute for static mechanisms in complex environments. The Inner Mongolia Dynamic mechanism may also be useful in matching potential employers and employees in the labor market.
Working Papers:
Revealing AI Reasoning Increases Trust but Crowds Out Unique Human Knowledge.
Zenan Chen, Ruijiang Gao, Yingzhi Liang. Working Paper (2025)
Effective human-AI collaboration requires humans to accurately gauge AI capabilities and calibrate their trust accordingly. Humans often have context-dependent private information, referred to as Unique Human Knowledge (UHK), that is crucial for deciding whether to accept or override AI's recommendations. We examine how displaying AI reasoning affects trust and UHK utilization through a pre-registered, incentive-compatible experiment (N = 752). We find that revealing AI reasoning, whether brief or extensive, acts as a powerful persuasive heuristic that significantly increases trust and agreement with AI recommendations. Rather than helping participants appropriately calibrate their trust, this transparency induces over-trust that crowds out UHK utilization. Our results highlight the need for careful consideration when revealing AI reasoning and call for better information design in human-AI collaboration systems.
Acting to Believe.
Wei Huang, Yingzhi Liang. Under Review (2026)
Can people build confidence simply by misremembering the past, or must they act to earn it? We examine these two routes in a field experiment with university students completing two rounds of self-control tasks. Before the second round, students were randomly assigned to receive no reminder, an expected reminder, or an unexpected reminder of first-round performance. Reminders correct potential memory bias and separate confidence generated by biased recall from confidence generated by effort. Students without reminders formed a positive memory bias, recalling first-round performance as better than it was, but this bias did not robustly increase confidence or second-task participation relative to the unexpected-reminder group. In contrast, students expecting a reminder exerted significantly more first-round effort, then reported higher confidence in self-control and participated more in the second task. We develop an intrapersonal self-signaling model driven by either a self-regulation or a self-image motive. Our findings support the self-regulation motive.
Team Size under Complementary Efforts: Theory and Experiments.
Yan Chen, Yingzhi Liang. R&R at Journal of Public Economics (2026)
Our study examines size effects for a team providing public goods. We show that the optimal team size depends on the degree of complementarity of the production function. Within the class of Constant Elasticity of Substitution (CES) production functions, we first prove that the low-complementarity technology admits a lower bound for the optimal team size, whereas the high-complementarity technology admits an upper bound. We then validate this prediction with a laboratory experiment. Specifically, we find that the average contribution in the first period is higher in ten-person teams than in four-person teams under the low-complementarity technology, whereas the reverse is true under the high-complementarity technology. Furthermore, we observe that the contribution gap increases over time, approaching the theoretical maximum but not the minimum contribution. A learning model incorporating social preferences explains both the convergence and non-convergence observed in our experiments.
Incorporating Private Information Into Centralized Algorithms: A Field Experiment at a Ride-Sharing Platform.
Yingzhi Liang. Working Paper (2025)
Many have seen the gig economy as the "future of work". Despite having more flexible working hours than workers at traditional workplaces, ride-sharing drivers have little power over the algorithm that assigns them tasks. As a result, there can exist a misalignment between driver location preference and assigned trips. We examine the effect of allowing drivers to self define working regions through a natural experiment on the largest ride-sharing platform in China, DiDi. We find that treatment drivers increase working hours and income by 5% while maintaining productivity, measured by hourly earnings. We also find no evidence that the treatment lowers matching efficiency, measured by passengers and drivers' waiting time.
Work in Progress:
Revitalizing Dormant Teams in Online Communities: A Field Experiment on Kiva.org.
Wei Ai, Roy Chen, Yan Chen, Yingzhi Liang, Qiaozhu Mei.
We investigate the effect of repeated interventions in charitable giving on microfinance platform Kiva.org. After sending forum messages to inactive teams every month for six months, we find that the treatment lenders lend significantly more in the first month, but the effect gradually decreases over the course of the experiment.
Team Composition: Friends or Strangers?
Yingzhi Liang, Tanya Rosenblat.
We study peer effect by randomly assigning group members in an undergraduate introductory programming course. Students are paired with a friend or a stranger in the same class to complete a group assignment. We find that students paired with friends have a 6% higher completion rate than students paired with strangers. A follow-up survey reveals that the effect comes from students being more accountable towards their friends than unfamiliar classmates. Our findings confirm the effectiveness of using strong ties as a commitment device in the education setting.
Predicting Students Academic Success Using Simple Economics Games.
Yan Chen, Yingzhi Liang, Qiaozhu Mei, Dongwu Wang, Stephanie Wang.
We predict undergraduate student GPA using their choices and behavioral patterns from a set of simple economics games, including the trust game, beauty contest, competitiveness game, risk lottery, and knapsack problem. Among machine learning models used, LassoLars performs the best, but the overall predicting power of these economic games is small. A self-reported procrastination measure is more predictive than any game choices.