The Securitization of Science: A Systems Perspective on Policy and Measurement
Pith reviewed 2026-05-21 01:27 UTC · model grok-4.3
The pith
Research security policies did not reduce international collaborations most in sensitive fields.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The share of China's internationally co-authored publications fell across all fields after 2018, consistent with domestic capacity building efforts. Contrary to deterrence expectations, high dual-use fields including computer science, materials science, physics, chemical engineering, and engineering experienced the smallest reductions of 3 to 6 percentage points. Fields with low dual-use relevance such as psychology, nursing, dentistry, and health professions saw the largest drops of 8 to 24 percentage points. Researchers appear to have maintained their frontier collaborations while modifying the form and language used in their work.
What carries the argument
Differential contraction in international co-authorship shares between high and low dual-use proximity fields as measured by Scopus categories.
If this is right
- Friction from security policies does not redirect scientific networks away from sensitive topics.
- Adaptive participation allows researchers to sustain collaborations by altering presentation rather than content.
- Deterrence predictions fail to match the observed field-specific patterns of decline.
- Policy outcomes are better explained by general network logic than by targeted national security measures.
Where Pith is reading between the lines
- If China's domestic initiatives are the main driver, similar declines would appear even without external policies.
- Scientists might increase use of informal channels or domestic partners for sensitive topics over time.
- Other countries adopting research security measures could see parallel adaptive responses in their data.
- Tracking changes in collaboration language or publication venues could provide further tests of adaptation.
Load-bearing premise
That the observed post-2018 shifts in co-authorship shares stem primarily from research security policies rather than simultaneous Chinese domestic science capacity programs, with Scopus categories serving as reliable indicators of dual-use sensitivity.
What would settle it
A study finding that researchers in high dual-use fields halted all international collaborations in detailed project-level data after 2018 would falsify the claim of adaptive participation.
read the original abstract
International scientific collaboration is organized primarily by researcher-level logic and network dynamics in which scientists seek partners at the frontier of their field with little consideration of national affiliation. Research security policies that raise the friction costs of international collaboration are often assumed to operate against this logic, aiming to produce selective withdrawal from sensitive fields to deny knowledge transfer to specific countries. This paper tests that assumption against eight years of bibliometric evidence from China's scientific output from 2018 through 2025 across 27 Scopus subject categories. China's internationally co-authored publication share declined universally across all fields after 2018, consistent with China's push for domestic capacity maturation, but the pattern of decline is inconsistent with the deterrence prediction. Fields with high dual-use proximity, including computer science, materials science, physics and astronomy, chemical engineering, and engineering, showed the least contraction (3-6 percentage points), while fields with minimal dual-use salience (psychology, nursing, dentistry, and health professions) showed the steepest declines (8-24 percentage points). The findings suggest that governments raised the friction level but did not redirect the network nor suppress collaboration in sensitive areas. Researchers practiced adaptive participation, maintaining frontier collaborations while adjusting the form and language of their engagement. This pattern matches what the invisible college theory and the Ostrom knowledge commons framework predict.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper analyzes eight years of bibliometric data on China's scientific output (2018–2025) across 27 Scopus subject categories. It reports a universal post-2018 decline in the share of internationally co-authored publications, which the authors attribute to domestic capacity-building efforts, but finds that fields with high dual-use proximity (computer science, materials science, physics and astronomy, chemical engineering, engineering) experienced smaller contractions (3–6 percentage points) while low-salience fields (psychology, nursing, dentistry, health professions) showed larger declines (8–24 percentage points). This differential pattern is interpreted as inconsistent with the deterrence prediction of research-security policies and as evidence of adaptive researcher participation consistent with invisible-college and knowledge-commons frameworks.
Significance. If the reported differential contraction pattern proves robust to alternative classifications and statistical scrutiny, and if it can be credibly linked to policy-induced friction rather than concurrent domestic initiatives, the result would supply a concrete empirical counter-example to the assumption that security policies produce selective withdrawal from sensitive domains. The work supplies a multi-year, field-disaggregated test of how network-level collaboration dynamics respond to raised transaction costs, thereby furnishing falsifiable support for the invisible-college and Ostrom commons predictions.
major comments (2)
- [Results (description of percentage-point changes)] The central claim that the observed pattern is 'inconsistent with the deterrence prediction' rests on the differential decline (3–6 pp vs. 8–24 pp) between high- and low-dual-use-proximity bins. No error bars, confidence intervals, or formal statistical tests comparing the two groups are reported, nor are robustness checks shown for alternative groupings of the 27 Scopus categories. Without these, it is impossible to determine whether the reported difference is distinguishable from sampling variation or from misclassification of sub-fields.
- [Methods (dual-use proximity classification)] The mapping of the 27 broad Scopus subject categories onto 'high dual-use proximity' (computer science, materials science, physics, chemical engineering, engineering) versus minimal salience is presented without a validated, policy-grounded coding scheme. Because US and allied measures target narrow domains (advanced semiconductors, specific quantum or AI applications, hypersonics) rather than entire fields, the lesser contraction in the high-proximity aggregate could be driven by non-sensitive sub-areas within those categories; if so, the pattern would be consistent with selective deterrence rather than inconsistent with it.
minor comments (2)
- [Data and methods] The exact operational definition of 'internationally co-authored publication share' (e.g., whether it counts fractional authorship, full counting, or only first/last-author international links) should be stated explicitly to permit replication.
- [Results] The manuscript would benefit from a short table listing each of the 27 categories together with its assigned dual-use label and the observed percentage-point change, so readers can inspect the raw pattern directly.
Simulated Author's Rebuttal
We appreciate the referee's detailed and constructive feedback on our manuscript. The comments highlight important areas for strengthening the statistical analysis and clarifying the classification methodology. We have revised the paper accordingly and provide point-by-point responses below.
read point-by-point responses
-
Referee: The central claim that the observed pattern is 'inconsistent with the deterrence prediction' rests on the differential decline (3–6 pp vs. 8–24 pp) between high- and low-dual-use-proximity bins. No error bars, confidence intervals, or formal statistical tests comparing the two groups are reported, nor are robustness checks shown for alternative groupings of the 27 Scopus categories. Without these, it is impossible to determine whether the reported difference is distinguishable from sampling variation or from misclassification of sub-fields.
Authors: We agree that the absence of formal statistical tests and robustness checks limits the strength of our claims. In the revised version, we now report bootstrap-derived 95% confidence intervals for each field's percentage-point change and conduct a two-sample t-test (or equivalent non-parametric test) comparing the average decline in the high dual-use proximity group versus the low-salience group. We also include robustness analyses using two alternative classification schemes: one based on a finer mapping to specific dual-use technologies and another excluding engineering subfields with lower security salience. These checks confirm that the differential contraction remains significant and directionally consistent. revision: yes
-
Referee: The mapping of the 27 broad Scopus subject categories onto 'high dual-use proximity' (computer science, materials science, physics, chemical engineering, engineering) versus minimal salience is presented without a validated, policy-grounded coding scheme. Because US and allied measures target narrow domains (advanced semiconductors, specific quantum or AI applications, hypersonics) rather than entire fields, the lesser contraction in the high-proximity aggregate could be driven by non-sensitive sub-areas within those categories; if so, the pattern would be consistent with selective deterrence rather than inconsistent with it.
Authors: We recognize that broad subject categories aggregate sensitive and non-sensitive research. Our initial classification was informed by policy literature on research security priorities, including US export control lists and reports from the National Science Foundation on critical technologies. To address the concern, we have expanded the methods section with a detailed justification table linking each category to specific policy instruments. Furthermore, we performed an additional analysis using only the most security-sensitive subfields within the high-proximity categories (where data permits) and find that the smaller decline pattern holds or is even more pronounced. This bolsters our interpretation of adaptive behavior over selective deterrence, though we acknowledge that finer-grained data would provide even stronger evidence. revision: partial
Circularity Check
No significant circularity; empirical comparison stands on external bibliometric data.
full rationale
The paper's core claim rests on direct observation of changes in international co-authorship shares across 27 Scopus categories from 2018-2025, compared against the qualitative deterrence prediction that security policies should produce selective contraction in high dual-use fields. No equations, fitted parameters, or self-referential definitions are used to generate the reported 3-6 pp vs. 8-24 pp differentials. References to invisible college theory and Ostrom knowledge commons function as post-hoc interpretive frames rather than premises that define or force the empirical pattern. The analysis draws on external Scopus data and is falsifiable against independent bibliometric sources; no load-bearing step reduces the result to the paper's own inputs by construction.
Axiom & Free-Parameter Ledger
free parameters (1)
- dual-use proximity classification
axioms (1)
- domain assumption Scopus publication counts and co-authorship flags accurately measure changes in international collaboration activity
Reference graph
Works this paper leans on
-
[1]
Balzacq, T. (Ed.). (2011). Securitization theory: How security problems emerge and dissolve. Routledge. Buzan, B., Wæver, O., & de Wilde, J. (1998). Security: A new framework for analysis. Lynne Rienner Publishers. Crane, D. (1972). Invisible colleges: Diffusion of knowledge in scientific communities. University of Chicago Press. Dao, T. K., Shannon, M. M...
-
[2]
Please confirm which.] Marginson, S
Stanford UP. Please confirm which.] Marginson, S. (2022). What drives global science? The four competing narratives. Studies in Higher Education, 47(8), 1566–1584. https://doi.org/10.1080/03075079.2021.1942822 Mulvey, B. (2025). Securitising the university: Lessons from China in the Xi Jinping era. Critical Studies in Education. https://doi.org/10.1080/17...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.