Source-linked AI summary
International Collaboration in Science and the Formation of a Core Group
Loet Leydesdorff, Caroline Wagner
TL;DR
The paper examines how international scientific collaboration forms a global network and whether its structure is expanding or concentrating. It analyzes collaboration patterns across selected years using network measures, finding a larger and denser network but a normalized core that does not grow as expected. The authors note that peripheral countries may be disadvantaged as stronger actors consolidate their position.
Problem
The paper asks how global scientific collaboration is structured and whether the European Union or elite national structures shape this network.
Method
The study adds a third year of data to an earlier analysis and examines global collaboration using network measures before and after normalization.
Results
The network grew larger and denser, but normalization did not support the expectation that its core structure had grown.
Takeaways & Limitations
Global collaboration appears to reinforce a smaller, tighter core within the wider network rather than producing an expanding normalized core.
Takeaways & Limitations
Developing countries may be disadvantaged because ideas can flow from their laboratories to larger actors better able to publish them.
Abstract
from arXiv · showhide
International collaboration as measured by co-authorship relations on refereed papers grew linearly from 1990 to 2005 in terms of the number of papers, but exponentially in terms of the number of international addresses. This confirms Persson et al.'s (2004) hypothesis of an inflation in international collaboration. Patterns in international collaboration in science can be considered as network effects, since there is no political institution mediating relationships at that level except for the initiatives of the European Commission. During the period 2000-2005, the network of global collaborations appears to have reinforced the formation of a core group of fourteen most cooperative countries. This core group can be expected to use knowledge from the global network with great efficiency, since these countries have strong national systems. Countries at the periphery may be disadvantaged by the increased strength of the core.
1. Introduction
International collaboration in science rose rapidly, with internationally co-authored papers increasing across fields. The paper frames global collaboration as a largely self-organizing network distinct from nationally mediated systems, while asking whether the EU or elite national structures shape it.
- An increasing share of scientific papers is co-authored by scientists from two or more countries.
- Internationally co-authored publications doubled during the 1990s and continued rising faster than nationally co-authored articles in the early 2000s.
- The increase in international collaboration occurred across all scientific fields at more or less the same rate.
- Global scientific collaboration is presented as a self-organizing communications network with no mediating political institution except European Union incentives.
- The introduction asks whether the EU emerges as an international actor or whether elite structures, including the USA, better describe collaboration patterns.
2. Data and methodology
The study analyzes international collaboration using Science Citation Index records from selected years and network measures based on country co-authorship. It uses integer counting, cosine normalization, and network analysis to identify collaboration structures while retaining document-level information.
- 734,750 articles, reviews, and letters from the Science Citation Index were considered for the selected years.
- Collaboration was measured as a document-level co-authorship event, with integer counting assigning one count to each country occurrence.
- Cosines were computed from the asymmetrical document-versus-country matrix, using cosine normalization because the distributions were not expected to be normal.
- Cosine values below or equal to 0.01 were treated as incidental variation and discarded from the relevant network structure.
- The analysis used both co-occurrence and normalized tables with UCINET and Pajek network-analysis software.
3. Results
International collaboration expanded through more addresses, links, and participating countries, making the global network larger and denser. The network also showed short distances, broad connectivity, and increasing opportunities for knowledge diffusion and small-world connections.
- Internationally co-authored publications grew linearly, while the number of international addresses grew exponentially from 1990 to 2005.
- The average number of addresses per internationally coauthored publication increased from 2.86 in 1990 to 3.61 in 2005.
- The number of participating countries increased by 20 between 1990 and 2000, then rose from 192 to 194 between 2000 and 2005.
- The number of links increased exponentially, and the k-core grew from 35 to 64 countries over the 15-year period.
- The network became denser, with higher average degree in 2000 and 2005 than in 1990, indicating more widely distributed influence across nations.
- Average distances between nodes were below two steps, and the network's small diameter supported dense connectivity and possible small-world structures.
- The global network formed a single connected component, while clustering was especially evident within the European Union.
4. The Effects of Normalization
Normalization reveals that the collaboration network’s central group is becoming smaller and more selective, while changing countries’ central positions and retaining a strong core structure.
- Normalization approach: The k-core measure reveals a latent network structure that cosine normalization uses to expose participation patterns beyond country-size effects.Normalization is needed to reveal the network’s structure, because country size alone affects participation.
- Core-group structure: The k-core shrank from 22 nations in 1990 to 21 in 2000 and 14 in 2005.Other countries remained linked to the core but were not structurally bound to it.
- Observed versus normalized networks: The observed network grew significantly and included an increasing number of countries in its core.Figures 3 and 4 compare k-core membership before and after normalization for 1990, 2000, and 2005.
- Core-group structure: Normalization shows the core remained broadly stable from 1990 to 2000 before shrinking between 2000 and 2005.This pattern suggests a more tightly ordered and self-selective core.
- Core-group membership: By 2005, the core no longer included all EU nations; the remaining non-EU members were the USA, Russia, and Switzerland.Denmark, Finland, and Portugal were excluded, while Ireland had never been included.
- Centrality changes: Before normalization, the USA was the most central country in every studied year; after normalization, France surpassed it in betweenness centrality during the 1990s.Russia also became increasingly important at the global level.
5. Discussion
The 2005 data show a larger and denser global collaboration network, while normalization does not support growth in the network core. Instead, collaboration appears to have produced a tighter, more coherent core through decentralized, self-organizing choices that favor scientifically advanced actors.
- The 2005 data satisfy the first two expectations: the collaboration network has grown larger and denser.
- After normalization, the third expectation—that the network core has grown—is not supported.The normalized network is smaller and tighter at the core.
- The core group appears to be becoming a more coherent cluster, possibly reflecting more deliberate partner choices by collaborators and policymakers.
- Scientifically advanced countries broadened their view of the global system while limiting partnerships to specific countries.These actors reorganized to exploit systemic changes and protect their positions.
- Global expansion increases knowledge-diffusion opportunities for peripheral scientists while further strengthening the core group's ability to use peripheral contributions.
- The global order emerges from researcher-level rules and collective action rather than from a single entity's plans.International collaborations now form part of an order spanning local, regional, national, and global levels.
6. Implications for Research Policy and Management
Globalized science requires policy and management approaches that account for distributed research, knowledge flows, evaluation, and accountability. Although the global system may gain efficiency through shared resources and complementary teams, its strengthening core may disadvantage peripheral countries.
- The global science system does not eliminate the nation-state’s influence, but requires governments to assess science at global, national, and regional levels.
- Policymakers must identify innovations developed elsewhere and make them locally available while evaluating distributed research and local absorptive capacity.
- Distributed tasking and resource sharing may improve system-wide efficiency by avoiding redundant national capacities and accelerating idea testing.
- As global links expand, tracking public spending against outputs and outcomes may become increasingly difficult.
- A strengthening core group may develop greater absorptive capacity, while peripheral countries risk losing ideas to larger actors better able to publish them.
- Benefits may reach peripheral countries more effectively only with deliberate policies and incentives that encourage knowledge flows and participation favoring them.