Beamtimes and knowledge production times: how big-science research infrastructures shape nations’ domestic and international science production

The paper investigates how a major big-science research infrastructure—China’s Shanghai Synchrotron Radiation Facility (SSRF)—affects the production of scientific knowledge within national borders and through international collaboration. The central research question asks whether, and how, the opening of SSRF altered the quantity and quality of high-impact national versus international publications across scientific disciplines. Situated in the broader context of nations investing in large-scale facilities to reach the scientific frontier, the study addresses the tension between serving national interests (domestic capability and competitiveness) and advancing the common global good via collaborative science. Drawing on catch-up theory, the authors posit that latecomer countries evolve from reliance on international collaboration toward more indigenous innovation and evaluate whether SSRF facilitated this transition by improving discipline-level outcomes among synchrotron-reliant fields.

The study builds on literature on big science’s role in frontier discoveries (Weinberg; Price; Hallonsten) and spillovers to innovation and national prestige. It engages center–periphery and international collaboration perspectives (Hwang; Traweek), and policy debates on returns to public R&D, assessment of large facilities, and the balance of big versus small science (Alberts; Fortin & Currie; OECD; ESFRI). Prior work often uses descriptive bibliometrics to assess facility outputs (e.g., synchrotron and astronomy cases). A related DID study on China’s National Supercomputing Center examined regional innovation at the city level; this paper advances the literature by employing a discipline-level DID and distinguishing national versus international publications to address endogeneity concerns more rigorously.

Data: The authors compile 1,908,610 English-language WoS-indexed journal publications with at least one China-based author (including Hong Kong, Macau, Taiwan) from 1970–2015. They exclude 55,584 papers from 76 multidisciplinary journals (e.g., Nature, Science) to avoid confounding and merge journal impact factors (IF) from Journal Citation Reports. Because IF coverage is systematic from 1998 onward, the final sample is 1,750,298 publications from 1998–2015.

Design: The opening of SSRF in 2009 is used as a quasi-natural experiment. Only some disciplines can directly use synchrotron radiation; these form the treatment group, while non-reliant disciplines serve as controls. Publications are categorized via WoS’s 252 disciplines (social sciences and humanities excluded). Treatment assignment combined bibliometric tracing of SSRF-originated publications and expert validation by SSRF user managers and beamline scientists, yielding 117 treatment disciplines and 96 control disciplines (213 total analyzed).

Outcome measures at the discipline-year level: (1) High-IF publication percent: share of publications in journals with IF > 10 (robust to alternative cutoffs such as IF > 5); (2) Mean IF of journals in which the discipline’s publications appeared. Outcomes are separately computed for national (all authors based in China) and international (cross-border coauthorship) publications.

Empirical model: A difference-in-differences regression with discipline and year fixed effects estimates the effect of SSRF: Y_it = β1 (Treatment_i × After_t) + β2 Controls_it + δ_i + δ_t + ε_it, where After_t = 1 for years after 2009 (post-opening) and 0 otherwise. Controls include: (a) NSFC funding (log) by broad field-year, mapped from WoS disciplines; (b) Graduate students (log) by broad field-year from the China Statistical Yearbook; (c) Overseas returnees (log) annually from the Ministry of Education (aggregate).

Identification and validation: Parallel pre-trends are visually checked and formally tested using event-study specifications (year-relative-to-2009 dummies). Robustness includes (i) falsification with 1,000 randomized placebo treatment assignments; (ii) placebo timing tests setting pseudo-shocks 1–5 years prior to 2009; and (iii) lagging all regressors by one year to account for publication delays. Heterogeneity analyses re-estimate the DID by grouping treatment disciplines into broader fields (physics, chemistry, materials science, engineering, biomedical science, environmental science). Standard errors are clustered at the discipline level.

• Baseline DID (1998–2015):
  • High-IF share: Treatment × After increases the percentage of high-IF national publications by 12.4% (statistically significant). The effect on the high-IF share of international publications is not statistically significant.
  • Mean IF: Treatment × After raises Mean IF by 0.143 for national publications (≈9.47% over the sample mean) and by 0.334 for international publications (≈17.67% over the sample mean), both statistically significant, indicating quality improvements for both national and international outputs.
• Event-study: Pre-2009 coefficients are near zero and insignificant, supporting parallel trends. Post-2009 coefficients become positive and significant (especially after two years for national high-IF share); for Mean IF, post-2009 coefficients are significantly positive for both national and international outputs.
• Robustness:
  • Randomized treatment falsification: Coefficients from randomized assignments center near zero and differ significantly from actual estimates.
  • Placebo timing: Pseudo-shocks before 2009 yield coefficients that are mostly insignificant or negative, supporting that effects arise post-SSRF opening.
  • One-year lag model: The increase in the national high-IF share is 14.4%; Mean IF rises by 0.175 (national) and 0.334 (international), consistent with baseline magnitudes.
• Heterogeneity across fields (Mean IF outcome): International publications show significantly positive SSRF effects across all broad fields. For national publications, effects are insignificant in physics, biomedical science, and environmental science, suggesting greater reliance on international collaboration in these areas.
• Additional bibliometric corroboration: Among top SSRF-linked disciplines, optics, environmental science/engineering, condensed matter physics, and cell biology show high international collaboration shares, aligning with the field heterogeneity results.

The findings directly address the study’s question by demonstrating that SSRF substantially benefits disciplines reliant on synchrotron radiation, increasing the quality and share of high-impact national publications while also elevating the overall quality of both national and international outputs. This supports the thesis that big-science infrastructures help latecomer nations both strengthen indigenous research capacity and engage the global scientific frontier—effectively “walking on two legs.” Policy-wise, the results argue for balanced portfolios that include big-science infrastructures alongside curiosity-driven and mission-oriented programs. They also inform assessment practices, indicating that facility impacts extend beyond direct outputs to discipline-wide quality improvements and that international collaboration remains crucial in specific fields even as domestic capability rises.

This study contributes a discipline-level DID analysis distinguishing national and international science to estimate the causal impact of a major big-science infrastructure. SSRF’s opening in 2009 increased the share of high-impact national publications and improved Mean IFs for both national and international outputs in synchrotron-reliant disciplines. The results highlight big-science infrastructures’ dual role in bolstering domestic scientific strength and enabling international collaboration. Future research should broaden scope to additional infrastructures and countries, integrate qualitative and quantitative approaches, measure wider spillovers (e.g., industrial R&D, networks, human capital), explore finer-grained heterogeneity across subdisciplines, and refine causal strategies where multiple shocks or staggered adoptions exist. Generalization to other types of big-science infrastructures should proceed cautiously.

• Potential confounding from other post-2009 shocks that might differentially affect treatment versus control disciplines.
• The one-time opening of SSRF allows identification of average treatment effects on treated disciplines but not nuanced dynamic heterogeneity across all narrower subfields.
• Treatment classification, while expert-validated, may still introduce measurement error across WoS categories.
• Outcomes focus on publications and journal IFs; other impacts (industrial R&D, networks, human capital, regional innovation) are not measured.
• External validity is limited: SSRF is a synchrotron facility; effects may differ for other big-science modalities. Results pertain to 1998–2015 and to China’s research system during that period.