← Back to Rankings

Rankings Methodology

How the Winnow Research Index calculates rankings for countries, institutions, authors, and journals.

Paper Selection

The full index contains 82,000+ papers sourced from OpenAlex and PubMed. After quality filtering, approximately 71,000 papers are used for rankings. The filters applied are:

  • Relevance threshold: Papers must score 30 or higher on our 0–100 relevance scale. The score is assigned by an AI classifier trained to distinguish genuine microplastics research from adjacent fields (e.g., materials science papers about "microplasticity" of metals).
  • Must have at least one author: Papers with no recorded authors are excluded. These are typically journal housekeeping records (covers, indexes) that databases like OpenAlex index alongside actual research.
  • Author count limit: Papers with more than 50 co-authors are excluded. These are typically conference proceedings, multi-consortium announcements, or bulk-indexed journal volumes — not individual research studies. Including them would inflate the paper counts of any institution that had even a single contributor, making rankings misleading.
  • Housekeeping records excluded by title: Papers whose titles match patterns like "Table of Contents", "Editorial Board", "Issue Information", "Front Matter", or "Contents List" are excluded regardless of other scores. OpenAlex sometimes classifies these as research articles — they are not.

Evidence Tiers

Each paper is classified into one of three evidence tiers based on its study design. The "Tier 1 Papers" column in rankings reflects the strongest evidence type.

Tier 1
Systematic reviews, meta-analyses, clinical trials — highest quality evidence; synthesizes or prospectively tests across multiple studies or human subjects.
Tier 2
Original research articles and narrative reviews — primary research on specific topics; the majority of the index.
Tier 3
Commentaries, editorials, letters, conference abstracts — opinion pieces or brief reports; counted in total papers but not in Tier 1 columns.

Institution Attribution

A paper counts toward an institution's total if any of its authors listed that institution as their affiliation in the paper's metadata. This follows the standard bibliometric approach used by databases like Web of Science and Scopus.

As a result, a highly collaborative paper with authors from 10 different institutions will appear in all 10 institutions' paper counts. This is intentional — each institution made a contribution to the work. The "Top Authors" section on institution pages shows the specific researchers driving that institution's output.

Institution assignments come from OpenAlex , an open catalog of scholarly works. Affiliations are resolved from author profiles and per-paper affiliation strings where available.

Author Attribution

Author paper counts reflect papers linked to that researcher in the OpenAlex database. Where possible, authors are disambiguated using ORCID identifiers, which provide a reliable cross-paper identity. Authors without ORCID are resolved by name and institutional affiliation, which may occasionally link papers from different people with similar names.

The resolution confidence score (visible on author detail pages) reflects how certain the identity match is: 1.0 for ORCID-verified, 0.9 for name match, 0.7 for ID-only resolution.

Country Attribution

A paper counts toward a country if any of its authors is affiliated with an institution in that country. Like institution attribution, a paper can be credited to multiple countries simultaneously. Country assignments are derived from institution records, which include ISO 3166-1 alpha-2 country codes from OpenAlex.

Data Sources

OpenAlex — primary source for paper metadata, author profiles, institution affiliations, and citation counts.

PubMed — supplementary source for biomedical papers not yet indexed in OpenAlex.

VoyageAI — 1024-dimension embeddings used for semantic search and relevance scoring.

Last updated March 2026. Methodology may evolve as the index grows. Questions? Contact us at hello@winnowlabs.com.