How journals are ranked

Release 2025.2 is computed from the OpenAlex snapshot of September 23, 2026. OpenAlex metadata is dedicated to the public domain under CC0. Every indicator year in the release, 2000 to 2025, is recomputed from that one snapshot, so all years use the same data and the same rules.

Ranking figures are computed by our code. Values taken directly from OpenAlex, such as its two-year mean citedness, are labelled as OpenAlex values.

The world-average baseline already includes all journals in OpenAlex, not just the ranked ones, so adding journals to the ranked set never shifts existing scores.

Subjects follow OpenAlex’s hierarchy of 4 domains, 26 fields and 252 subfields, 282 categories in all. Each paper takes the subfield of its primary topic.

Each journal is assigned one subfield, the same in every indicator year: the subfield holding the most of its classified citable papers in the latest three-year window. The journal is then ranked in that subfield, in the subfield’s field and domain, and in the global ranking.

A citation is counted only when the citing paper’s reference list was captured. Coverage is lower in older years, so every citation figure shows the coverage of its citing year. The reported rate is corpus-wide: the share of core-journal articles and reviews published in a year that have at least one matched reference. It is not a per-journal figure, reference capture can vary between journals and fields, and a paper with a matched reference may still have an incomplete list.

In this release, coverage is lowest in 2000 at 66.2% and highest in 2022 at 89.5%. In 2025, 84.3% of core-journal articles and reviews have at least one matched reference, corpus-wide.

Actual and expected citations are both counted in the same citing year, so a change in coverage that affects every field equally cancels out. Coverage that differs by field or journal does not, which is why the citing year’s coverage is shown with every score.

Only citable papers count: articles, reviews and conference papers, using the document types OpenAlex assigns. Editorials, letters, news items and other front matter are left out.

Document type is part of the comparison, so reviews are compared with reviews and articles with articles. Papers without a subfield are left out and reported as classification coverage.

• Journal self-citations are left out.
  A citation to one of the journal’s papers from another paper whose primary source is the same journal.

  Self-citations are left out of both the journal’s count and the world average. The self-citation rate is shown separately on every journal page.
• Unknown sources count. Citations from works without a known journal still count, since they cannot be confirmed as self-citations.
• Repositories are excluded. Citations from repositories, including preprint servers and data archives, are left out; see citing sources and citation audit.
• Each citing–cited pair counts once.

Citations count when the citing work is hosted by a journal, conference, book series or ebook platform, or has no known source. Works whose primary source is a repository are excluded before every citation-based calculation, including world-average expectations, journal self-citations and top-10% shares. This release excluded 28,910,621 repository window edges across all indicator years.

A journal is ranked for an indicator year when all of these hold:

• it is in the release’s ranked set, CWTS Core journals in this release;
• it has at least 30 citable documents in the three-year window;
• at least 80% of those citable documents have a subfield;
• its world-average expectation is positive, so OJR can be computed.

A journal that misses a rule keeps its page and statistics, and the page states which condition it failed.

OJR is a ratio: the citations a journal’s recent papers actually received, divided by the citations papers like them receive on average worldwide. The method is called NCI3, for 3-year normalized citation impact.

So a 2022 review in oncology is measured against 2022 oncology reviews, and a 2024 conference paper in software against 2024 software conference papers. That is what makes OJR comparable across fields with very different citation habits.

• 1.00 means the journal’s papers were cited as often as the world average for papers of the same subfield, publication year and type. 2.00 is twice the world average; 0.50 is half.
• OJR is a ratio of sums, total actual over total expected. It differs from a mean of per-paper ratios (MNCS in the Leiden Ranking) and from OpenAlex’s per-work FWCI, which uses the publication year plus three later citation years.
• OJR is not computed when the journal has no classified window papers or Expected is zero.

Within each category and year, journals are ranked by OJR rounded to four decimals, the published precision. Journals showing the same score share a rank.

Ties share a rank and the next rank is skipped: 1, 2, 2, 4.

Each journal has a rank in its subfield, field and domain and in the global ranking. Because OJR is normalized per paper, scores are comparable across categories, which is what makes field, domain and global rankings meaningful. When a category has fewer than 10 ranked journals, its rank and quartile are not published; the journal’s other levels remain.

Ranked journals in a category are split into four equal groups by OJR; Q1 is the top quarter. Each journal has a quartile in its subfield, field and domain. These quartiles describe this site’s population and method. They are not SCImago or Scopus quartiles.

Nor are they Journal Impact Factor (JIF) quartiles.

A range that crosses a quartile boundary is described as such, but the published quartile does not change. Earlier years have no range.

Ranking pages include a summary of the category for the page’s indicator year: ranked journals, median OJR, window papers and change over three years. A historical ranking page summarizes its own year.

Self-citations count; citations from repository-hosted works do not. It is not normalized by field: unlike OJR, it does not adjust for how often papers in different fields are cited.

Clarivate’s Journal Impact Factor™ uses the same basic formula but gives different numbers, and this site does not show it.

The values can differ because of citation coverage, document classification and publication dates; the size and direction of the difference have not been established across this release. Don’t compare the two directly.

Why our impact factor differsLooking for the JIF? Master Journal List

Each measure uses journals ranked in 2025 with a known value meeting its eligibility rule. Missing values and ineligible journals do not enter its population.

The group is the field of the journal’s fixed subfield. When fewer than 20 eligible journals belong to that field, the domain is used instead. If the domain also has fewer than 20, there is no comparison. Group membership is computed separately for every measure.

The median and 25th and 75th percentiles use continuous quantiles of per-journal values. The position is 100 × (journals with a lower value + half the journals with an equal value) / group journals, rounded to a whole number. The journal itself is included. Pooled rate measures use the sum of numerators divided by the sum of denominators, with their stated rate scale, and omit quantiles and positions.

Unless stated otherwise, measures use citable papers published in 2022–2024 and citations made in 2025.

These comparisons describe the ranked population, not all journals or research quality. Fields can contain different subjects and publishing models. Unknown OA status, affiliation countries and APC prices are excluded rather than treated as zero; citation counts follow each measure’s stated rules. Coverage differences can affect the comparisons. Release group summaries include groups below the minimum size, flagged as small. Shares are exported to six decimal places and USD to whole dollars; comparisons use unrounded values.

This release compares indicator years 2022 and 2025, using non-overlapping publication windows 2019–2021 and 2022–2024. It describes change in citation impact relative to the typical journal in the comparison group.

In both windows, a journal needs a positive OJR, at least 100 citable papers, at least 80% of them classified and a computable standard error. Guards then remove journals whose change would mostly reflect the data rather than the journal:

• Content change. Output in 2022–2024 must be between 0.4 and 2.5 times output in 2019–2021. A bigger swing usually means the indexed content changed. Excludes 476 journals.
• Low base. OJR must be at least 0.1 in both windows; lower values usually indicate citations missing from the source data. Excludes 384 journals.
• Single paper. No one paper may contribute more than 50% of the journal’s counted citations in either window. Excludes 32 journals.

We subtract the median log NCI3 change among eligible journals in the same field. If fewer than 20 journals qualify, we use the domain; if that also has fewer than 20, no trajectory is assigned. Relative change is the journal’s NCI3 ratio divided by the group’s median ratio, minus one.

• Rising: relative change of +10% or more, and z ≥ 2, with the low end of the two-standard-error range at zero or above.
• Falling: relative change of -9.1% or less (rounded), and z ≤ −2.
• In line: every other eligible journal.

11,451 journals have trajectories: 1,582 rising and 1,636 falling. Falling journals are listed in the Falling tab of the 2025 ranking pages and labelled on journal pages; rising journals also have their own lists: Rising journals

The publication windows do not overlap, but citing sources can overlap, so the noise estimate does not establish independence. Reference coverage drifts between years and can differ by journal. The journal’s static subfield may miss changes in its subjects. Uncertainty in the group median is ignored. An unusual citation peak in the earlier window makes the later year look lower: papers from 2019–2021, for example, include the COVID-19 citation surge, so a falling label in medicine or multidisciplinary science can reflect a return from that peak rather than a long-term decline. The 100-paper minimum excludes smaller journals and does not remove all sampling noise. Trajectory describes citation impact, not research quality or a forecast.

OpenAlex sets document types automatically, and its types can be incomplete or inconsistent. Two problems matter most:

• Reviews are undercounted. Many reviews are typed as articles. Type normalization then compares them with articles rather than with reviews, and the review-share measure understates the true share of reviews. When the indexed review share is at least 10%, journal pages show an articles-only OJR beside the score.
• Some front matter is typed as articles. News items, editorials and letters typed as articles enter the citable set. They are rarely cited, so they can pull down both OJR and the Impact factor (2-year).

A fix for both problems is planned for the next method version.

• A citation is counted only when the citing paper’s reference list was captured. Coverage is lower in older years and can vary between journals and fields; see reference coverage.
• OpenAlex links references for most but not all works. A work with references may still have an incomplete list.
• The citation audit cannot detect every reference error.
• Rank ranges do not account for errors in the underlying citation data.

• Papers without a subfield are left out of both the journal’s score and the world average. A journal needs at least 80% of its citable papers classified to be ranked.
• Each journal is ranked in one subfield, chosen from its recent papers. A fixed subfield may miss changes in a journal’s subjects, and multidisciplinary journals are still ranked in a single subfield.

• This release ranks CWTS Core journals only. National and regional journals, non-English journals and many arts and humanities journals fall outside that list, so they are not yet ranked.
• Journals need at least 30 citable documents in the window, and scores of small journals still depend on a few papers. That is why ranks carry a likely range.

OJR measures citation impact under the published NCI3 method. It is not a judgement of research quality or an endorsement. Its quartiles are this site’s, not SCImago, Scopus or Journal Impact Factor (JIF) quartiles.

Don’t use it to judge individual articles or researchers. See responsible use and integrity for intended use, how we handle citation manipulation, and corrections.

For how this method compares with the published evidence, and where that evidence is mixed, read What the research says about OJR.

Releases are annual and named by their last indicator year, the way JCR names its editions: release 2026 holds indicator years up to 2026, just as “the 2025 impact factor” is the JCR value for 2025.

The first release, 2026-Q4, was named under an earlier quarterly scheme, by its snapshot quarter, and holds indicator years up to 2025. Later releases use yearly names only. A fix that only re-renders a release, such as a translation fix, gets a new name, like 2026.1 (or 2026-Q4.1 for the first release), with identical computed tables.

This page describes release 2025.2, which holds indicator years 2000 to 2025.

A release is immutable. Each release recomputes every indicator year from its own snapshot, so as OpenAlex’s reference matching improves, historical values can change from one release to the next. Each release records coverage by year so the change is visible. Always quote a value with its release.

Each part of the method carries a version name, shown on the pages that use it.

Method nci3-v1, with changes: Meeting abstracts excluded from citable papers.

Changes applied to the method without a new version name, newest first. Each applies from the first release that records it; earlier releases keep the method they were computed with.

1. Meeting abstracts excluded from citable papers

   Recorded October 20, 2026

   OpenAlex records many meeting abstracts, news items and other short pieces as journal articles. They are rarely cited, so counting them as citable papers pulled down the scores of journals that print large abstract supplements, most of them in medicine. We now leave a paper out of the citable papers when OpenAlex has matched none of its references and either it belongs to an abstract supplement (its DOI names a supplement, or it appeared on the same day as a large set of papers without references in a journal whose other papers have references), or it is in the health sciences, by its own subject or by its journal's main subject, and its publisher has matched references for at least half of its articles and reviews that year. A paper with at least one matched reference always counts, and the same rule applies to the papers behind the world averages. About 6% of the papers of CWTS Core journals are left out, and in our checks about 93% of them were abstracts, news, editorials or other front matter. The typical journal's OJR moves by about 2%; journals with large abstract supplements rise, some of them several times over, and about 75 journals no longer reach the 30-paper minimum.

Name the project and the release, link to the licence, and say if you changed the data. Values are recomputed for every year in each release, so always give the release.

• Data and text published by each release are licensed under CC BY 4.0. OpenAlex records themselves are CC0.
• Code is licensed under the MIT License. It will be published under the OpenJournalRank organization; contact us for access meanwhile.
• Inputs are fixed. Each release names its snapshot date in its manifest, and the rank-range bootstrap uses a fixed seed, so the published numbers can be recomputed.

Every ranking figure is computed by our code from a frozen OpenAlex snapshot; provider values are labelled as OpenAlex values. Page text uses fixed templates filled with computed numbers. No page text is written by AI. This process makes scores reproducible and applies the same definitions to every journal.

Found something wrong? Write to contact@openjournalrank.com.

About OpenJournalRank