Xiaojie Zhang

PhD Candidate in Economics

Hello! I'm a behavioral scientist and PhD candidate in Economics at Nanyang Technological University, Singapore, expecting to graduate in 2027. I am on the 2026–2027 Academic Job Market.

I study how people interpret information, allocate attention, and form judgments and decisions, particularly about discrimination, morality, and inequality. I also investigate AI through a behavioral economics lens.

My research uses quantitative approaches, including controlled behavioral experiments and studies of large language models (LLMs). I integrate process-tracing tools such as eye-tracking and emotion detection to capture patterns of attention and emotional responses, helping explain the processes underlying judgment and decision-making.

Research

Job Market Paper

Abstract

Organizations often treat workforce diversity as evidence of fairness, yet diversity itself may shape how discrimination is recognized. We investigate this possibility in a preregistered experiment with 1,159 U.S. adults using webcam-based eye tracking. Participants evaluated hiring outcomes in which discrimination against a focal racial minority group (Asian American or Black American, assigned between subjects) was held constant while the demographic composition of other selected candidates varied. We find that greater diversity reduced perceived discrimination against both focal groups, but its effects on visual attention were asymmetric. When the focal applicant was the only racial minority selected, Asian American and Black American applicants received comparable attention. As the hiring outcome became more diverse, attention to Asian American applicants declined, whereas attention to Black American applicants was maintained or increased. Exploratory analyses suggest two potential mechanisms: diversity may serve as a signal of fairness, reducing perceived discrimination, while competition for limited attention may disproportionately reduce scrutiny of Asian American applicants. These findings reveal that workforce diversity may not only shape how fair an organization appears, but also whose disadvantage receives attention and whose is overlooked.

Working Papers

Demand for Uninformative Predictions under Risk: Performance Histories, Transparency, and Prosocial Value

with Nattavudh Powdthavee and Yohanes E. Riyanto

Revise and resubmit at Journal of Behavioral and Experimental Economics

Abstract

Predictions that are statistically independent of the outcomes they forecast have no instrumental value, yet people may still acquire them after observing a record of success. We examine whether disclosing their random generation reduces demand and whether demand persists when acquisition benefits charity. In a preregistered laboratory experiment, 750 participants in 39 sessions bet on five fair coin tosses and could surrender part of their endowment to see a prediction before each toss. A 2 × 2 design varies disclosure of the prediction process and whether the surrendered amount is retained by the experimenter or donated to charity. Before any performance history is observed, disclosure reduces acquisition from 30.5% to 13.5% under purchase but only from 30.3% to 28.2% under donation. Successful histories subsequently increase acquisition in all four conditions, including under full disclosure, and the data do not establish that disclosure weakens this response. In the full-information donation condition, 87.8% of bets by acquirers following successful histories match the prediction, against 48.4% among non-acquirers. Disclosure can therefore reduce conventional demand for an uninformative prediction but does not eliminate responses to random success. When acquisition is prosocial, initial demand may persist, and acquirers still use the prediction.

Perceived Similarity, Life Satisfaction Judgments, and Redistributive Allocation

with Anthony Lepinteur and Nattavudh Powdthavee

Revise and resubmit at Journal of Wellbeing Economics

Abstract

It is well-established that people tend to allocate more resources to ingroup or socially similar recipients. Yet it remains unclear whether these transfers reflect a belief that similar recipients suffered larger life satisfaction losses, or whether similarity influences allocation through some other channel. We test this in a preregistered experiment with U.S. participants. Evaluators reported perceived similarity to a target profile, judged changes in a target’s life satisfaction following mirror-opposite income, career, and health shocks, then allocated funds from a temporary public-assistance program to the same target. While randomly assigned demographic similarity (shared gender, age proximity) had no effect on judged welfare loss or allocation, with Bayesian evidence favoring the null, we find that perceived similarity predicted both. A one-standard-deviation increase in perceived similarity predicted percentage points more allocation, though greater similarity was also associated with smaller judged effects of negative shocks. Similar recipients received more support despite being judged to have suffered less, not more. Judgments showed pronounced loss aversion: negative shocks were judged to reduce life satisfaction about times as much as equivalent positive shocks increased it. These findings separate judgments of subjective well-being loss from distributive priority, indicating that similarity-based allocation reflects in-group favoritism rather than a belief that similar recipients suffered larger life-satisfaction losses, with implications for how support is targeted in public-assistance and redistribution settings.

Credit and Responsibility in Human-AI Collaboration: Attribution in AI-Assisted Knowledge Production

with Yao Qu

Abstract

This study examines how individuals make ethical judgments about credit and responsibility in human-AI collaboration and identifies the psychological mechanisms underlying attribution in this context. Drawing on attribution theory, we contrast motivational and informational accounts using a randomized experimental design that isolates attribution from performance-based inference. The results show that attribution follows an informational logic: negative outcomes are treated as diagnostic of responsibility and lead to a systematic concentration of attribution on the human actor, while evaluative perspective plays a limited role. This pattern is further conditioned by how AI is perceived, with responsibility becoming more distributed when AI is viewed as a collaborative contributor rather than a tool. We argue that this responsibility-centered pattern reflects the liminal status of AI in knowledge production, where contributions are distributed but accountability remains anchored to humans. These findings have implications for moral accountability, authorship, and the ethical governance of knowledge work in AI-assisted contexts.

Selected Work in Progress

AI Prediction Through Human Eyes

with Josie Chen, Tai-sen He, and Nattavudh Powdthavee

AI versus Police Officers: Who Is More Susceptible to Information-Order Effects?

with Tom Burrows, Tom Kirchmaier, and Nattavudh Powdthavee

Gender, Specialization and Credit

with Boon Han Koh and Jonathan Yeo

Identity, Endowment Source, and Redistribution

with Boon Han Koh, Jonathan Yeo, and Ruge Zhang

Inter-generational Fairness

with Nathamuni Aadhityaa and Jonathan Yeo

Contact


I'm interested in conversations and collaborations across behavioral science, experimental economics, human–AI interaction, and public policy.

xiaojie001@e.ntu.edu.sg

Based in Singapore · Nanyang Technological University