The National Research Index is built from raw scholarly records to comparable national and institutional measures. The index adopts no external indicator, score or ranking: every figure is computed in-house from first principles, under an open methodology, on a global multilingual corpus that GINC assembles and maintains.
Approach
We take scholarly records, assign each work to the authors, institutions and nations that produced it, and build every measure from that base — the same way for every nation and every field, so that any two readings can be set against each other on a common footing.
The whole method reduces to three steps: assemble a global corpus, resolve the entities and the network within it, then compute the bibliometrics and impact indices on top. The rest of this page works through each in turn.
- Global Corpus. Index all known institutional research — every public index and repository, direct national and institutional submissions, and continuous scanning of newly released work. Sources include DOI registries, OpenAlex, arXiv, Crossref, PubMed, PubMed Central, DOAJ, ORCID, ROR, ISSN, Unpaywall, the Internet Archive and OAI-PMH-harvested institutional repositories.
- Entity Resolution & Network Integrity. Paper, author and institution disambiguation and de-duplication; citation-network integrity checks; and a paper-by-paper topic classification rubric applied across the whole corpus.
- Bibliometrics & Impact Indices. Fractional attribution by author share, field-weighted impact read from the citation network rather than raw counts, and rolling, rebased time series — the foundation for a family of impact indices and bibliometric signals.
1 · Global Corpus
The index rests on a corpus we assemble and maintain ourselves, drawn from every primary source we can reach. There is no single database, and no finished metric is taken from any provider — each contributes records, which we process from there.
- Public indices and repositories — broad open coverage of works, venues and the citation reference graph, led by OpenAlex and the sources it draws on: Crossref, PubMed and PubMed Central, arXiv, DOAJ, ORCID, ROR, ISSN, Unpaywall and OAI-PMH-harvested institutional repositories.
- Specialist research datasets — discipline- and region-specific collections that improve coverage where general indexes are thin, particularly outside the established Anglophone venues.
- Direct submissions — records contributed by institutions and research bodies, used to fill gaps and correct attribution at source.
- Continuous scanning — newly released work is ingested on an ongoing basis, so the corpus tracks the live record rather than a fixed snapshot.
No source is treated as authoritative on its own. Records are reconciled across sources and the same work appearing in more than one place is matched and merged to a single canonical record, with conflicts resolved against the most complete and verifiable version rather than by precedence of provider.
2 · Entity Resolution & Network Integrity
Before any work can count toward a nation it must be tied to the right people, places and references — and the citation graph it sits in must be sound. This is the most consequential stage in the method; every downstream national figure depends on it.
2A
The record is first cleaned. Disambiguation resolves authors, institutions and venues to single identities across the whole corpus; author disambiguation carries the most weight, since national attribution flows directly from it, and is treated accordingly. De-duplication then collapses the same work arriving from multiple sources into one canonical record, so nothing is counted twice and every measure starts from a clean denominator.
Citation-network integrity. Citations are linked work-to-work across the reconciled corpus to form the reference graph that all citation measures rest on. The graph is checked for citation rings and reciprocal clusters against a null model, so that engineered citation is separated from genuine influence — structure, not accusation.
2B
Topic classification. Each work is then mapped to one or more research fields by a consistent paper-by-paper rubric, applied the same way across the whole corpus, so every national measure can be read whole or sliced to a single field on the same basis.
3 · Bibliometrics & Impact Indices
From the resolved corpus and its citation network we compute every national measure, the same way for every nation and field. Four conventions govern how the figures are counted.
Fractional credit. All counts are fractional by author share, not whole-counted per nation. A paper with four authors contributes a quarter to each, and those fractions roll up to institutions and nations; credit sums to one across all contributors. This avoids inflating multi-national collaborations and keeps national totals additive.
Field weighting. Measures are normalised by field so that a work in a heavily cited discipline does not outweigh one in a sparsely cited discipline. Comparison is always like with like.
Rolling windows. Volume and movement are read over rolling multi-year windows rather than single years, so trajectory can be read alongside scale without single-year noise.
Rebasing. Time series are rebased to a common reference point. Where a measure is a share of a global total this is stated explicitly, since a share can fall while the absolute level rises.
On this footing, impact is read from the structure of the citation network, not the size of the pile. A citation from a highly cited work counts for more than one from an obscure one; presence in the global top decile counts for more than the long tail; and work that displaces what it cites is told apart from work that merely adds to it.
The indices are organised into six families — Scale, Authority, Excellence, Integrity, Connectivity and Trajectory — each a pair of readings of the same record, from raw output and field-normalised citation impact through to eigenvector-weighted authority, top-decile leadership, field-shifting disruption, inward self-citation, cross-border collaboration and the direction of travel over time. The full set is laid out on the Index page.
Every index is produced at two resolutions — national and institutional — and across the full field taxonomy, so the same readings of Scale, Excellence and Connectivity can be taken for a country, a university or a single research group. The series span the twenty-first century, from 2000 to the present, so current standing is always set against a quarter-century of trajectory rather than a single year.
Research fields and the framework
The fields a work is classified into are GINC’s own list, set out on the Fields page: the union of OpenAlex’s hierarchy of domains, fields, subfields and topics with eight standard and national classifications (ANZSRC 2020, the OECD fields of research, Science-Metrix, China’s GB/T 13745, the United States’ NCSES Taxonomy of Disciplines, the United Kingdom’s REF 2021 units of assessment and HECoS, and India’s AISHE disciplines). Where two or more name the same field it is one GINC field carrying each of their codes; the Mapping page shows how each classification maps onto the list, and what each side has that the other lacks.
Each of the finest fields is tied to the capabilities of the National Capability Framework it informs. A link carries a strength, the share of the field’s papers that are about the capability, and from it a weight, the field’s share of the capability’s total strength, so a capability’s weights sum to 100. The links are a judgement, not a computation: a first full pass, made under a written rubric and not yet reviewed link by link. A field is linked only where its research bears directly on a capability; nothing is linked by default to fill a gap, and the links of one capability, Physical Health & Longevity, each carry the reason they were made.
Limits and caveats
The measures are only as good as the record they are built on, and the record is incomplete in known ways.
Coverage is uneven. Bibliographic coverage varies by language, discipline and region. Specialist datasets and direct submissions narrow the gaps but do not close them. Counts are indicative, not exhaustive.
Disambiguation carries error. Author and institution resolution is imperfect at scale. Because attribution is load-bearing, residual error propagates into national figures; we treat this as the primary validity risk and report at resolutions where it is tolerable.
Field boundaries are choices. How a field is delineated affects its results. For field-level indices the definition is published alongside the figures, so the boundary can be inspected rather than assumed.
Open records under-measure closed research. Work that is classified, commercial or otherwise unpublished does not enter the corpus. In strategically sensitive fields this can understate real national capability, and the measures should be read as covering the open research record, not the whole of national activity.
Small slices are volatile. At narrow field or short time slices, counts fall and figures grow sensitive to a handful of works. Series below a minimum volume are shown with wider bands or withheld.
Openness
The method and index definitions are published in full; the intent is that any figure can be traced back to the way it was built. Underlying data is made available to the extent that coverage and licensing allow, sufficient for the published measures to be checked. Full documentation lands alongside each data release.
Citation
The content and data of this site are open under CC BY 4.0, with credit; the code is MIT. Cite as:
Global Institute for National Capability (GINC), GINC Research, research.ginc.org, 2026.