Seokbeom Kwon

Research Highlights

Featured work

Overview of five featured papers — each with the main figure from the paper and a one-line read.

Scientific importance of dual-use research

Field-year normalized citations: DUR vs non-DUR papers, 1981-2005

DUR are scientifically more influential than non-DUR, sustained from 1981 to 2005. Shading indicates 99% confidence intervals.

Science 2026 · 392(6802): 1032–1035 · Solo-authored

Dual-Use Research Under Scrutiny

As the U.S. tightens ex-ante oversight of dual-use research (DUR) for security reasons, two questions arise. How scientifically important is DUR — and is U.S. oversight even reaching it, given that science is global?

  • Data ~600,000 research articles published 1981–2005, classified by dual-use potential
  • Method Three-tiered security screening at USPTO; patent-paper citation analysis
  • Finding 1 DUR is scientifically more influential, sustained over decades
  • Finding 2 An increasing share of DUR emerges without U.S. federal government involvement
  • Implication Heavy ex-ante oversight risks delaying diffusion of important science while losing institutional reach as DUR globalizes — pushing reliance toward costlier ex-post measures
  • Open data Replication dataset on Dryad
Read paper → Dataset

Event study + recipient–provider similarity

Event-study estimates showing parallel trends and null post-treatment effect

Event-study estimates. Productivity (Poisson) of data providers vs. matched comparison group, 5 years before to 4 years after data sharing. Pre-treatment trends are parallel; the post-treatment trajectory shows no significant divergence.

Bar chart comparing similarity within and across provider-recipient pairs

Why no penalty? Provider–recipient research pairs are less similar (lower ZSimilarity, fewer shared subject codes and journals) than provider–provider or recipient–recipient pairs. Data recipients pursue distant questions, so diversion dominates competition.

Research Policy 2025 · 54(9): 105308 · Solo-authored

Competition or Diversion? Effect of Public Sharing of Data on Research Productivity of Data Provider

Scientists worry that publicly disclosing research data hurts their own publication opportunities — rivals can race ahead with the same data. But the empirical evidence has been scant. Does mandatory data sharing harm the original data creators' research productivity?

  • Setting NIH mandate to share data via dbGaP archive
  • Method Panel DID + synthetic control on data providers vs matched controls
  • Finding No negative impact on data providers’ research productivity
  • Why Data recipients pursue research questions distant from providers’ — diversion dominates competition
  • Implication Sustainable data-sharing policies can be designed without penalizing original teams
Read paper →

US patent-paper pairs: govt acknowledgment

28% UNDISCLOSED 28% — federal support hidden in patent records 72% — properly acknowledged ~84,000 US PATENT-PAPER PAIRS ANALYZED
Science 2024 · 385(6712): 936–938 · Solo-authored

Underappreciated Government Research Support in Patents

Federally funded research drives a huge share of US innovation. But when patents fail to acknowledge that government support, the public loses twice: government’s contribution is underestimated, and its right to exercise patents for public health or safety is undermined.

  • Data ~84,000 US patent-paper pairs (PPPs) linking research outcomes to patents
  • Finding 28% of patents on federally-funded research outcomes did not acknowledge the US government’s support
  • Implication Government’s role in innovation is systematically underestimated; potential public-interest exercise of patent rights is weakened
  • Open data Replication dataset available on Dryad
Read paper → Dataset

Synthetic control evidence on follow-on innovation

Four-panel synthetic control results: regulated vs control, treatment effect, placebo, p-value

Synthetic control results. Top-left: follow-on citations (PostCiteOIN) for the regulated group rise above the synthetic counterfactual after the 2011 intervention. Top-right: the average treatment effect grows steadily over four post-intervention years. Bottom: placebo tests against ~500 alternative units — the observed effect lies in the tail of the placebo distribution (p = 0.02).

Research Policy 2021 · 50(9): 104295 · with A. Marco

Can antitrust law enforcement spur innovation? Antitrust regulation of patent consolidation and its impact on follow-on innovations

When a firm consolidates patents covering substitute technologies, it can choke off competitors’ follow-on innovation. Can antitrust law — usually framed in terms of consumer prices — serve as a tool for innovation policy?

  • Case 2011 US DOJ regulation blocking transfer of Novell’s software patents to Microsoft, Oracle, EMC, Apple
  • Data US patent, trademark, and copyright records
  • Finding Antitrust intervention increased competitors’ follow-on innovation
  • Implication Antitrust law can be a complement to patent law in promoting innovation, not just a constraint on monopoly
Read paper →

Bibliometric evidence across three emerging domains

Pairwise correlations between IES and IFWD across NEDD, SynBio, AutoV

IES–IFWD correlations

IFWD distribution for NEDD

NEDD — IFWD distribution

IFWD distribution for synthetic biology

SynBio — IFWD distribution

IFWD distribution for autonomous vehicles

AutoV — IFWD distribution

Top-left: positive IES–IFWD correlations across all three domains. Other panels: in each field, papers with higher emerging-idea content (red) sit visibly to the right of the comparison distribution (blue) — the relationship is not driven by outliers.

Research Policy 2019 · 48(9): 103834 · with X. Liu, A. Porter, J. Youtie

Research Addressing Emerging Technological Ideas Has Greater Scientific Impact

Funders, journals, and labs all face the same gamble: invest in the new and unproven, or stick with established lines. Does engaging with emerging ideas actually pay off in scientific impact?

  • Domains Nano-enabled drug delivery, synthetic biology, autonomous vehicles
  • Method Bibliometric indicators detecting emerging terms in titles + abstracts; regression on future citations
  • Finding Papers containing more emerging ideas received more citations — both within and across fields
  • Implication Engaging with technological emergence is associated with broader scientific influence, not just niche relevance
Read paper →

See further publications.

View all publications