U.S. export-control sanctions cut publication volumes from listed Chinese universities but raise average research impact, as scholars pivot topics and route collaborations through domestic intermediary institutions rather than severing U.S. links.

Main Finding

Entity List sanctions on Chinese academic institutions causally reduce the quantity of research output but increase its quality. Sanctioned researchers adapt by diversifying research topics and reconfiguring collaboration networks—particularly by partnering with non-listed Chinese institutions that act as intermediaries—rather than by cutting ties with U.S.-based collaborators. These behavioral responses explain the opposing effects on quantity and quality and show that sanctions reconfigure, rather than sever, global academic networks.

Key Points

• Treatment and outcome
  • The U.S. Entity List inclusion of Chinese academic institutions is treated as a staggered, exogenous shock.
  • Primary outcomes: lower publication counts (quantity) and higher bibliometric quality (e.g., impact/citations or venue prestige).
• Causal evidence
  • A stacked event study design is used to estimate causal effects of listing on research productivity.
  • Effects are robust to dynamic event-study checks (pre-trends tested) and supported by complementary analyses.
• Mechanisms
  • Research diversification: sanctioned researchers shift into new topics and lines of inquiry.
  • Network reconfiguration: researchers form more collaborations with non-listed Chinese institutions that function as intermediaries, restoring access to restricted resources and introducing new knowledge.
  • U.S. ties remain largely intact: no significant decline in collaboration with U.S.-based researchers, indicating reconfiguration rather than isolation.
• Qualitative support
  • Semi-structured interviews corroborate quantitative mechanism claims, documenting how intermediaries and topic shifts operate in practice.
• Theoretical contribution
  • Extends the “disruption-as-catalyst” logic to geopolitically imposed resource constraints: sanctions can trigger productive adaptation and diversification.
  • Reconciles prior inconsistent findings by showing sanctions can simultaneously depress output volume while improving output quality through identifiable behavioral channels.

Data & Methods

• Data
  • Publication-level data drawn from the Open Academic Graph, providing comprehensive coverage of authorship, affiliations, publication venues, and citation information.
  • Identification of sanctioned (Entity Listed) Chinese institutions and affiliated researchers.
• Empirical design
  • Stacked event study (difference-in-differences with staggered treatment timing) to estimate the dynamic causal impact of Entity List inclusion.
  • Comparison between treated researchers and appropriate control groups, with event-time fixed effects to control for common shocks.
• Mechanism analysis
  • Network and topic measures constructed from coauthorship and text/keyword data to capture collaboration patterns and topic diversification.
  • Mediation-style tests linking changes in networks/topics to observed changes in quantity and quality.
• Qualitative validation
  • Semi-structured interviews with affected researchers and institutional actors to validate the quantitative mechanisms and provide contextual detail.

Implications for AI Economics

• Supply-side consequences for AI/tech research
  • Sanctions can reduce the short-run supply of publications from targeted institutions but may increase the average impact of surviving work, altering the shape of global research production.
  • For AI subfields that rely heavily on specialized hardware, data, or cross-border collaboration, intermediaries and network reconfiguration can enable continued progress despite restrictions.
• Knowledge diffusion and competition
  • Intermediary institutions and topic diversification can create alternative pathways for knowledge transfer, potentially preserving or even enhancing competitive capabilities of sanctioned actors.
  • Policymakers aiming to slow technological capacity via sanctions should account for adaptive responses that may partially offset intended effects.
• Policy design and enforcement
  • Targeted sanctions produce complex, heterogeneous effects—reducing quantity but potentially raising quality—so evaluations of their effectiveness should consider both dimensions.
  • Enforcement and export-control policy may need to anticipate and address intermediary channels that restore access to restricted inputs.
• Modeling and empirical practice in AI economics
  • Models of innovation and technology diffusion should incorporate endogenous network reconfiguration and topic-switching as adaptive responses to constraints.
  • Empirical studies should measure both quantity and quality outcomes and explicitly examine network mechanisms when assessing geopolitically driven interventions.
• Future research directions
  • Long-run effects on cumulative innovation, commercialization, and human capital mobility in AI.
  • Field-level heterogeneity: how effects differ across AI subfields (theory, systems, applied ML) with varying dependence on hardware, data, and international collaboration.
  • Welfare implications for global AI development and strategic competition.

Limitations to bear in mind: identification relies on standard parallel-trends assumptions for event studies and on the ability to observe intermediaries and topic changes; measurement of “quality” is bibliometric and may not capture downstream applied impact.