Research

Job Market Paper

"The Professional Costs of Speech Sanctioning Events: Evidence from Academia" with Lorenzo Palladini (American University of Beirut, Mediterraneo)

Presenting at the NBER Productivity Seminar in November 2026

Conference Awards:

Steven Klepper Award for Best Young Scholar Paper DRUID 2024
Nominated for the 2026 SMS Conference Best PhD Paper Prize (October 2026)
Competitive research fellowships for this research:

Institute for Humane Studies Research Award Fellowship (2024)
Institute for Humane Studies Junior Fellowship (2026)
SAR Fellowship (2026)

Summary: How do organizations respond when an employee is publicly sanctioned for speech? Using a novel dataset of U.S.-based scholars and a staggered difference-in-differences design, we find that sanctioned scholars publish approximately 20% fewer papers and receive 4.8% fewer citations to their prior work. Their close prior coauthors also publish 12.5% fewer papers. Citation declines are concentrated among prior coauthors and same-institution colleagues and persist when the contested speech falls outside the scholar’s area of expertise. These patterns are consistent with professional distancing and show how reputational shocks propagate through scientific teams and organizations.

Case Studies

"Harvard Business School Case 625-048, August 2024 "Managing Science: Perspectives from Postdocs" with Kyle Myers (Harvard Business School), Rem Koning (Harvard Business School), Solène Delecourt (Berkeley Haas), and Katelyn Cranney (Stanford).

Peer-Reviewed Publications
"When Government Shuts Down Science: Evidence from U.S. Antarctic Research" with Christian Helmers (Santa Clara)
Forthcoming at Research Policy.
Media Coverage: Nature

Summary: Can brief political disruptions have lasting effects on scientific production? Exploiting the 2013 U.S. federal government shutdown in a difference-in-differences design, we find that affected scientists’ subsequent publication output fell by approximately 11%. Although researchers partially adapted by changing their collaboration patterns and reallocating effort, these adjustments were insufficient to offset the loss. The findings show that even brief disruptions can generate persistent declines in knowledge production when research depends on time-sensitive, geographically fixed, and difficult-to-substitute inputs.

Working Papers

"Do Women Ask For Less? The Gender Ask Gap in Science" with Valentina Tartari (Stockholm School of Economics) , H.C. Kongsted (Copenhagen Business School), Astrid Ulv Thomsen (Copenhagen Business School), and Lorenzo Palladini (American University of Beirut, Mediterraneo).
Reject and Resubmit at Nature
Media Coverage: Times Higher Education

Summary: Do gender disparities in scientific funding emerge before proposals are evaluated? Using the universe of approximately 8,600 funded and unfunded applications submitted to two large European science funders, we find that female principal investigators request 5.0–5.4% less funding than comparable male applicants, accounting for approximately 70–80% of the corresponding gap in awarded funding. Smaller requests do not increase the probability of success, suggesting that the gap does not reflect strategic under-requesting. The gap persists in team-based applications, even when female PIs apply with male co-applicants, although mixed-gender teams partially attenuate it. It is concentrated among early-career and first-time applicants and disappears with repeated applications, consistent with learning and information frictions in how researchers calibrate the resources needed to pursue their ideas.

"Benchmarking the Future of Work: From AI Progress to Economic Capabilities" with Jake Prokopets.
Under Review at Management Science
Summary: AI benchmarks measure what models can do, but not where those capabilities matter economically. We introduce the Benchmark-to-Work Index (BWI), which translates AI and robotics benchmark performance into a dynamic measure of occupational exposure by assessing capabilities relative to humans and their relevance to real-world work. BWI predicts subsequent patterns of AI use, providing an early signal of where technological capabilities are advancing and where firms may find opportunities to redesign workflows and invest in complementary capabilities.

"CentaurBench: Benchmarking LLM Assistance on Real-World Work" NBER Working Paper 35663. with Kenny Wongchamcharoen, Min Min Fong, and Abhishek Nagaraj (Berkeley Haas).
Accepted at NeurIPS 2026 Workshop on Agentic AI Benchmarks and Applications for Enterprise Tasks (AABA4ET)

Summary: Are the large language models that perform best autonomously also the most effective at assisting workers? Across seven economically grounded tasks, we compare models completing work independently with models providing guidance to a fixed lower-capacity worker model. Automation and augmentation rankings differ substantially, and assistance does not consistently improve performance, showing that organizations should evaluate AI models according to the specific roles and tasks in which they will be deployed.

"Does Politics Permeate Science? Evidence from a Field Experiment on Political Bias in Academic Opportunity" with Jessica Khan (Northwest Florida State College). Draft available on request.

Presented at the NBER Investments in Early Career Scientists 2026

Summary: Critics frequently allege political discrimination in academia, but does it actually shape access to academic opportunities? Across three audit experiments involving approximately 10,000 U.S. professors, we randomly vary the political orientation signaled by fictitious prospective students. We find no evidence of differential treatment among STEM professors, but document bias against conservative applicants in the social sciences and humanities. The findings show that political gatekeeping is field-specific and informs current debates over federal funding interventions in higher education: broad university-wide funding cuts may not be warranted.

"The Missing Colleague: AI and Informal Intellectual Collaboration" with Raymond Kim and Jake Prokopets. Very early draft - more empirical work to come soon.

Summary: Does generative AI replace the informal intellectual contributions of colleagues? Using over 14,000 NBER and arXiv economics working-paper records, we find that AI-flagged manuscripts acknowledge approximately 21% fewer informal collaborators in specifications with researcher fixed effects and extensive controls. We find suggestive evidence that AI may substitute for domain-specific expertise and reduce reliance on harder-to-reach colleagues, highlighting a potential trade-off between AI assistance and the human relationships that sustain scientific knowledge production.

"Historical and Modern Legacies of the Centrality Gradient" Justin Cook (Tulane) and Raymond Kim (Westmont College). Submitted.

Summary: Using 873 polities spanning 3400 BCE to 1500 CE on a fixed quarter-degree grid, we estimate a centrality gradient with great-circle distance and human mobility cost, showing cells farther from the center of a historical polity were less densely populated. Cross-sections show this gradient steepens with a cell’s cumulative years under rule, and a within-cell panel identifies it from plausibly exogenous border movements. As evidence of public good provision and accumulation, Roman road density declines with distance holding population fixed. Distance to the historical center still explains where modern capitals sit, but not modern population density or nighttime lights.

Early Work in Progress

"AI Science RCT" with Rem Koning (Harvard Business School)

"Tasks in the Age of Automation" with Abhishek Nagaraj (Berkeley Haas) and Isaac Robinson (Mercor)

"When Researchers Self-Censor: Political Pressure and the Direction of Science" sole-authored.

"Identifying AI R&D Uplift: A Research Agenda for Frontier-Lab Data" Coming soon - a conceptual piece!

Page updated

Google Sites

Report abuse