The Research Collaboration Index measures one thing precisely: how central a university is in the global research-collaboration network. It is objective, reproducible, and built entirely on open data.
What it measures: six pillars
For every university, six indicators are computed from the network of its co-authored research with the world's leading research universities (2021-2025). Publication credit is fractionally counted, shared across each paper's collaborating institutions, so a large university cannot score simply by being large, and three of the six are deliberately size-independent.
Network influence (0.22) · weighted eigenvector centrality on the fractional collaboration graph: a university scores highly when its partners are themselves well-connected, so this rewards partnering with central institutions, not merely partnering often.
Partner reach (0.16) · the number of distinct leading universities it co-publishes with, its degree in the network.
Partner-country diversity (0.16) · the number of distinct countries its partners span, the international breadth of its network. Size-independent: a small university can span many countries.
Collaboration impact (0.18) · the field-normalised citation impact (FWCI) of the co-authored output, so collaboration that produces influential work counts for more than collaboration for its own sake. Field-normalised, so a humanities partnership is not penalised against a biomedical one.
Sustained partnership (0.18) · the count of deep partnerships (five or more co-authored publications), durable alliances rather than one-off papers.
International intensity (0.10) · the share of output produced with international co-authors. Size-independent: it rewards a genuinely cross-border research culture, not volume.
The three-step score
Each of the six indicators is computed for every university from the open collaboration record.
Each is converted to a within-cohort percentile (0 to 100), so measures on different scales become directly comparable.
The Collaboration Score is the weighted sum of the six percentiles. The weights sum to one, so a weight is simply that measure's maximum share of the score, and every university's score decomposes exactly into the six contributions.
Robustness. The score combines six independent measures, three of them size-independent (diversity, impact, international intensity), so no single indicator and no amount of sheer size can dominate. We publish the score's rank correlation against full (non-fractional) counting, against equal weights, and against dropping the impact pillar; the ordering is stable under all three. Every input is open and auditable, and the entire pipeline can be independently rebuilt.
Brokerage: a reported diagnostic
Alongside the six scored pillars, each fact file reports a brokerage diagnostic. It is not part of the Collaboration Score (it correlates with reach and we would rather not double-count), but it answers a distinct and useful question: does a university merely collaborate widely, or does it genuinely bridge partners who are not otherwise connected to each other? Two universities can each have 150 partners; the one whose partners form separate, otherwise-unconnected clusters is the broker. We compute it as a genuine structural property of the network, combining two complementary published measures rather than a naive proxy (counting the distinct world regions a university's partners fall into saturates immediately, since almost every leading university already reaches every region):
Structural holes, 60% (Burt's effective size). Of a university's partners, how many are non-redundant, that is, not already collaborating heavily with each other? A university whose partners all co-publish among themselves adds little bridging; one whose partners would be disconnected without it spans a "structural hole." We report this on each fact file as the share of partners that are non-redundant.
Backbone betweenness, 40%. From the full weighted network we first extract its statistical backbone using the disparity filter (Serrano, Boguna & Vespignani, 2009), keeping only collaborations significantly stronger than chance for at least one partner. On that backbone we compute weighted betweenness: how often a university lies on the strongest path connecting two others, capturing global bridging between research communities.
The two are each converted to within-cohort percentiles and blended 60/40. The result separates all 1,195 universities (it takes 101 distinct values), and its leaders are the institutions one would expect to bridge the global system, such as Oxford, UCL, Harvard and Sorbonne. Every university's brokerage inputs (its non-redundant-partner share and backbone position) are shown on its fact file so the figure can be checked, not merely trusted.
The five-year window
The index uses a rolling five-year window of co-authored research (currently 2021 to 2025). Participation is recency-weighted: an institution's recent activity (its share of output in the most recent years) counts for more than older activity, so the index reflects current collaboration. Each annual edition rolls the window forward one year.
Subject areas and subject-level indices
Alongside the overall index, every university is scored within four broad research areas from the global scholarly taxonomy: Life Sciences, Physical Sciences, Health Sciences and Social Sciences. Each area is scored on its own collaboration network, so a specialist can lead in its field even if it is smaller overall.
Beneath the four areas, the index is also computed for every individual subject with enough data to rank at least 50 universities, 26 subject indices in the current edition. Each subject is scored on its own worldwide co-authorship network using the same six pillars as the overall index (with citation impact and international intensity field-normalised within the subject), so a leader in one subject is a genuine global comparison and not a separate exercise, and no large multi-disciplinary university can top a field it is not genuinely strong in. The 50-university threshold ensures every published index is statistically credible; subjects that do not clear it are not published rather than shown thinly. As coverage grows, more subjects qualify each edition.
Data, scope and identity
Source: the open global scholarly record of publications, authors and institutions (CC0, openly licensed). Publications means peer-reviewed journal articles and reviews. No surveys, no citations, no licensed commercial data enter the score.
Cohort: the world's leading research universities (currently 1,195 institutions across 81 countries). The network is measured among these institutions, so the index reflects collaboration with the world's foremost research universities.
Identity: institutions are resolved through the Research Organization Registry (ROR). Federated umbrella bodies are handled separately so their member universities are ranked in their own right.
Reproducible: the entire pipeline runs on open data and can be independently rebuilt.
Funding data and its sources
The index does not rely on a single source. Funding is drawn from each authority where it is definitive, and every funder is named from the funders' own records, not inferred:
European Commission (CORDIS) · verified euro amounts and project counts for Horizon Europe and Horizon 2020, mapped to each university. Over €30 billion of EU research funding is analysed.
Crossref Funder Registry, CORDIS & the funders' own award systems · grant records from funders' own systems worldwide: the US National Institutes of Health and National Science Foundation, UK Research and Innovation, the European Research Council, the German DFG, Japan's JSPS, China's NSFC, Wellcome and thousands more. In total, 19 million grants worth $6.1 trillion from 29,851 funders are analysed.
National funder systems (NIH RePORTER, NSF, UKRI Gateway to Research, the Australian Research Council, Canada's tri-agency and others) are being integrated directly to deepen coverage where the open record is thin.
Funding coverage is uneven by region in the open record, so we show each university's funding only from the source that is authoritative for it (European Commission figures for EU universities, funder-registry grants elsewhere) rather than presenting an incomplete or misattributed total. Amounts in non-USD currencies are shown either in their native currency (EU) or converted at indicative rates for comparison; they measure analysed grant funding, not a quality verdict.
Data currency. Collaboration is measured over 2021-2025 from the open scholarly record as at late June 2026; funding records reflect the same June 2026 data cut. The window and data refresh each edition.
What it does, and does not, do
It does
Measure how central a university is in the global research-collaboration network.
Compare every university on the identical, open, reproducible method.
Show who a university collaborates with, how widely, and how deeply.
It does not
Measure reputation, teaching quality, or overall university quality.
Rank citations or research impact; those are different questions.
Claim to move or predict any other ranking or reputation score.
The index is a measure of collaboration standing. It is a lens, not a verdict on a university's worth.
See where your university stands
The full index launches 15 September 2026. Partners verify their data, feature their academics, onboard 500 members and compete for the Subject Awards.