Methodology
How we calculate
every signal.
No black boxes. Every score on OpenRLens is derived from open scholarly data using reproducible logic. This page documents exactly what we measure, where the data comes from, and what the numbers mean.
Data sources
All data is sourced from publicly available, open scholarly databases. We do not use proprietary datasets — everything below is freely accessible to anyone.
OpenAlex
Primary source for papers, authors, affiliations, concepts, and citation counts. Fully open, updated continuously.
Crossref
DOI metadata and funder acknowledgement records. Used to cross-reference funding presence on individual papers.
Europe PMC
Funder and grant metadata for biomedical and life-science literature.
NIH RePORTER
US National Institutes of Health grant records linked to publications.
NSF Awards
US National Science Foundation grant data linked to funded papers.
UKRI
UK Research and Innovation grant records for UK-affiliated institutions.
OpenAIRE
European open-access funder metadata, covering EU Horizon and national funding bodies.
CORE
Open-access full-text repository. Used as a supplementary source for papers not indexed elsewhere.
How signals work
Each signal is scored on a 0–1 scale and classified as High, Moderate, or Low based on fixed thresholds derived from the distribution across all institutions in our dataset. Signals are computed over the year window you select — changing the window changes the result.
Higher is better — a high score on this signal is a positive indicator (e.g. Affiliation Stability).
Lower is better — a high score on this signal is a risk indicator (e.g. Collaboration Dependency).
Informational — no direction is better. The signal provides context, not a verdict (e.g. Gender Balance Index).
Structural Signals
These six signals describe the fundamentals of a research program — who leads the work, how stable the team is, and whether output is consistent.
Institutional Leadership Signal
What it measures
How often does this institution play a leading role on its own papers — as first author, corresponding author, or PI-equivalent last author?
How it's calculated
Each paper is scored: +0.5 for first author, +0.6 for corresponding author, +0.4 for last author (PI position in engineering fields), plus up to +0.4 for affiliation dominance (anchored at 35% share of authors). Scores are capped at 1.0 per paper. ILS = mean score across all papers in the window.
Corresponding author metadata is missing on a significant share of older papers. ILS may be slightly underestimated for institutions with strong pre-2015 output.
Research Concentration
What it measures
Is the citation output concentrated in a small group of star researchers, or spread across the wider faculty?
How it's calculated
Authors with ≥2 papers are ranked by total citations. RCR = the share of total citations held by the top 10% of these recurring authors. High concentration is neither good nor bad — it depends on whether the stars are tenured or likely to leave.
Name disambiguation is imperfect. Authors with common names may be merged or split, slightly inflating or deflating concentration scores.
Collaboration Dependency
What it measures
How much of the institution's output depends on a small number of external partner institutions? If those partners disappeared, how much would output drop?
How it's calculated
The top-3 external collaborating institutions are identified by co-authorship frequency. Dependency = max(paper drop%, citation drop%) when those three partners are excluded from the dataset. High = research is not self-sustaining.
Affiliation Stability
What it measures
Do researchers stay affiliated with this institution across multiple years — a proxy for faculty retention and program stability?
How it's calculated
Institutional authors are classified as 'recurring' (≥2 papers across ≥2 distinct years) and 'stable' (≥3 distinct years). ASI = stable ÷ recurring. Confidence is penalised when fewer than 50% of authorships carry a stable author ID.
Author ID coverage varies by country and era. Institutions with older or less-indexed literature may show lower apparent stability.
Output Stability
What it measures
Is annual publication volume consistent, or does it spike and crash? Consistent output suggests a healthy, sustained research culture.
How it's calculated
Stability score = 1 − (standard deviation ÷ mean) of annual paper counts, floored at 0. The current year is excluded when fewer than 6 months of data are available to prevent partial-year distortion.
Thematic Continuity
What it measures
Do the research topics stay focused across years, or does the program drift? A focused program is easier to evaluate and signals genuine expertise.
How it's calculated
Top-5 concept labels are extracted per year from scholarly metadata. Jaccard similarity (|A∩B| ÷ |A∪B|) is computed between every consecutive year pair, then averaged. Only concept-tagged papers contribute.
Funding Signals
These five signals look at whether the institution's research is externally funded, and crucially — whether it leads that funded work or merely contributes to it.
Funding Presence
What it measures
What share of the institution's papers carry at least one external funder acknowledgement?
How it's calculated
Funded papers ÷ total papers. Funder data is drawn from Crossref, Europe PMC, NIH RePORTER, NSF Awards, UKRI, and OpenAIRE.
Open databases significantly underreport funding acknowledgements — particularly for older papers and non-Western funding bodies. FPR is a lower bound, not an exact figure.
Funding Leadership
What it measures
On papers that carry external funding, does this institution hold a lead authorship position — or is it a supporting contributor?
How it's calculated
Identical calculation to ILS, but restricted only to papers with at least one funder record. Measures PI-level authorship presence specifically in funded-context work.
Funding Dependency Differential
What it measures
Does this institution lead its funded work more or less than it leads its general output? A positive gap means it tends to be the PI when money is involved.
How it's calculated
FDD = FLS rate − ILS rate. Positive = institution leads a larger share of funded-context papers than its overall output. Near zero = consistent behaviour. Negative = more often in a supporting role on funded work specifically.
Lead Role on Funded Papers
What it measures
Of papers with funder metadata, how many have no lead author from this institution? High = the institution rarely drives the research it is credited with funding.
How it's calculated
Funded papers with no lead institutional author ÷ total funded papers. Lower is better — a low EFAS means the institution typically holds PI-equivalent authorship when funder metadata is present.
Funding-Theme Alignment
What it measures
Does the funded research align with the institution's core research themes, or does it pull in a different direction?
How it's calculated
Top-5 concept labels from funded papers are compared to top-5 from unfunded papers using Jaccard similarity. High = funding reinforces core strengths. Low = funded work diverges significantly from the program's main research focus.
Mentorship & Diversity Signals
These two signals are informational — they don't contribute to the overall verdict but give additional context about who is doing the research.
Mentorship Diversity Index
What it measures
Is the institution consistently bringing new junior researchers into first-author roles, or is the same cohort producing everything?
How it's calculated
The analysis window is split in half. MDI = unique institutional first authors in the second half who did NOT appear in the first half ÷ total unique institutional first authors in the second half. High = institution is actively elevating new researchers.
Gender Balance Index
What it measures
Is the first-author cohort gender-balanced? This is an estimate, not a census.
How it's calculated
Estimated from author first names using an offline name-gender database. Only institutional first authors in the second half of the window are counted. Balance = min(p_male, p_female) × 2; 1.0 = perfect 50/50 split.
Binary inference only — non-binary identities are not captured. Accuracy is ~80–85% for Western names and lower for East Asian, Arabic, and other origins. Treat as indicative only.
The overall verdict
The overall verdict shown at the top of each results page is a weighted composite of the structural and funding signals. Informational signals (MDI, GBI) do not contribute to the verdict.
We do not publish the exact weighting coefficients. This is the one part we keep internal — not to obscure the methodology, but because weights are calibrated iteratively and publishing fixed numbers would create incentives to optimise for the score rather than the underlying research quality.
What we can say: positive-polarity signals contribute positively, negative-polarity signals contribute inversely, and no single signal dominates. An institution cannot achieve a Strong verdict without broad coverage across both structural and funding dimensions.
Known limitations
Funding underreporting
Open funding databases cover a fraction of actual grant activity. FPR and related signals are lower bounds. Institutions with less-indexed funding bodies (e.g. many Asian and African national funders) will appear less funded than they are.
Author disambiguation
OpenAlex uses name-matching and ORCID links to identify authors. Common names may be merged across different people, or a single person's record may be split. This affects RCR, ASI, and MDI.
Concept tagging
Research topics are inferred from paper titles, abstracts, and metadata by OpenAlex's ML classifiers. Niche or emerging fields may be tagged inconsistently, affecting TC and FTA.
Time window sensitivity
All signals are computed over your selected date range. A short window (e.g. 2 years) produces high-variance results. We recommend a minimum 3-year window for stable readings.
Non-English literature
English-language papers are better indexed in open scholarly databases. Institutions where a significant share of output is in other languages may appear less productive than they are.
Have a question about a specific signal or score?
Send us feedback