Research

Working Papers

[Job Market Paper] Time Is of The Essence: Experimental Evidence on The Effects of Job Application Timing
Abstract: This paper studies how job application timing affects employer consideration through firms’ sequential screening. Repurposing the correspondence experiment of Kline, Rose and Walters (2022), I show that, holding vacancy and resume content fixed, first-half experimental applications have a contact rate 1.16 percentage points higher, equivalent to 57 percent of the Black-White contact gap in the same setting. Each day earlier raises contact probability by 0.40 percentage points. The key mechanism is that firms often begin contacting while applications are still being received. Using randomized resume attributes, I instrument whether contact has already begun with the highest predicted contact probability among previously submitted applications. Stronger prior applications trigger earlier initiation, and IV estimates confirm a post-initiation contact penalty. The timing gradient is steeper in tighter labor markets. A sequential model separates applicant arrival rates from the shadow cost of delay and quantifies how much screening value firms forgo by moving sooner. Conditional on the firm contacting an experimental application, the highest-ranked application has not yet arrived in 51.6 percent of cases, yielding a 6.6 percent screening-value shortfall. Together, the results provide the first large-scale causal evidence that firms’ sequential contact decisions make queue position consequential within vacancies, revealing an overlooked employer-side margin through which labor market conditions shape who is considered and how applicant pools turn into matches. draft
presented at: All-California Labor Economics Conference 2024 (poster), PolMeth Europe 2025, WEAI 100th Annual Conference, 3rd Annual AEA CSQIEP Mentoring Conference, 6th World Labor Conference, 20th EGSC at WashU St. Louis, 90th Annual MEA Meeting (2026), 5th AMIE Workshop in Applied Microeconomics, 8th QMUL Economics and Finance Workshop for PhD & Post-doctoral Students

When Hierarchies Teach: Selective Review and Decision Alignment in U.S. Asylum Courts with Daniel L. Chen

Abstract: Hierarchical review evaluates experts' first-instance decisions and informs their subsequent choices, but this feedback is selectively generated. We study this process in U.S. asylum courts, where feedback follows a denial, applicant appeal, and Board of Immigration Appeals (BIA) disposition. Using linked immigration-court and BIA records from 1986 to 2019, we develop a rolling prediction model that provides an observable ex ante reversal-risk benchmark for classifying realized reversals by signal diagnosticity. Weekly event studies show a 1.57-percentage-point larger first-week increase in lower-court grant rates after surprise than after expected reversals, with no broad decline in observable hearing-diligence proxies. We develop a threshold model of response magnitude and distributional alignment with the BIA standard. High-response judges have flatter marginal treatment effects (MTEs) than low-response judges, especially at below-average predicted grant probabilities. The pattern appears on both appeal filing among denials and BIA success conditional on appeal, with the clearest contrast on appeal filing. This alignment gap is smaller in defensive than affirmative proceedings, consistent with structured hearings making feedback more usable. Review's teaching function therefore depends on feedback selection, signal diagnosticity, and recipient responsiveness.. draft
presented at: ALEA 2026 Meeting, CELS 2026

From AI Exposure to Training Access: Evidence from U.S. Workforce Boards

Abstract: I link 2017-2024 administrative records from local boards under the Workforce Innovation and Opportunity Act (WIOA) to training provider listings and state funding to provide new empirical evidence on how public training systems respond to post-GenAI changes in expected skill demand. After ChatGPT's launch in 2022, boards serving more AI-exposed workers and with greater prior experience show a 3-percentage-point larger rise in training participation, about 30 additional trainees per 1,000 participants. Training content shifts much less toward AI-related fields. Eligible-provider listings expand, but listed programs do not automatically translate into funded AI training. The access response is larger in state-years with greater formula allotments, and training activity becomes more concentrated in boards with established routines. The findings show that public workforce systems can expand access when resources and delivery experience are in place. Targeted AI retraining, however, depends on whether those systems convert provider availability and fiscal support into funded participation. draft
presented at: ALEA 2026 Meeting, CELS 2026

Work in Progress

Employer Networks as Organizational Capital in Professional Education: Evidence from U.S. Law Schools draft available upon request
Reallocating Expertise in the AI Era: Evidence from the U.S. Patent Office draft available upon request
Delegation under Deadlines: Evidence from U.S. Patent Prosecution draft coming soon
Strategic Predictability: Theory and Evidence from Professional Tennis draft available upon request
Beyond the Altar: Catholic Convents and Gender Norms in the United States, 1850-1940 with Monia Tomasella and Juan David Torres

Research Statement

ResearchStatement_SaiZhang.pdf

Page updated

Google Sites

Report abuse