Peer Review in Numbers

Acceptance rates are the figures publishers are least consistent about publishing. Where a journal states one, it is here. Where a rate circulates widely but traces to nothing, it is in the last section instead.

Science

5.3% of research papers accepted from 15,088 submissions, with a 15.0% review rate, meaning 85% were never sent for external review. 2025, AAAS journal metrics.

Science Advances

8.8% accepted from 35,831 submissions, 16.4% review rate. 2025, same source.

Nature

about 8% of submissions accepted, and most declined without being sent for peer review. Undated, from Nature's editorial criteria page.

PNAS

14% acceptance rate, with more than half of submissions declined at initial evaluation. Undated, PNAS editorial policies.

PLOS ONE

26% acceptance rate on 18,300 articles published, against 4% at PLOS Medicine and 43% at PLOS Neglected Tropical Diseases. 2025, PLOS research metrics.

Frontiers

43% accepted and 57% rejected across all journals, with 65,544 manuscripts desk-rejected and 45% of all rejections made by its in-house research integrity team. 2024, Frontiers annual report.

MDPI

a rejection rate above 60% on more than 669,000 submissions publisher-wide. 2025, MDPI announcement.

eLife

27.7% of submissions sent for peer review under the Reviewed Preprint model, against 31.4% under the previous model. January 2023 to January 2024, eLife.

Two kinds of figure appear here and they are not comparable. Publisher medians measure one journal's internal process. The study figures measure what authors across many journals actually experienced, which is longer.

13 weeks on average to a first decision across all fields, with 19% inside one month and 10% beyond six months. 2013 to 2016, 3,500 review experiences, Huisman and Smits, Scientometrics 2017.

8 weeks in medicine against 16 to 18 weeks in economics, the social sciences, mathematics and the humanities. Same study.

17 weeks average total review duration for accepted papers, over an average of 2.03 rounds. Same study.

16.4 days median between a reviewer agreeing to review and delivering the report. 2016 to 2018, Global State of Peer Review 2018.

8 days median to a first editorial decision at Nature and 32 days at PLOS ONE. 2025.

321 days median from submission to acceptance at Nature and 242 days at Nature Communications. 2025.

About 100 days median submission to acceptance across 3,482 biomedical journals in PubMed, rising over time. 2015, Himmelstein.

The reviewer supply is the part of the system under most visible strain, and it is the best documented. The single report that most of these figures come from is the 2018 Global State of Peer Review, which has not been updated since.

2.4 invitations sent per completed review in 2017, up from 1.9 in 2013. ScholarOne cross-publisher data, Global State of Peer Review 2018.

Invitations sent grew 9.8% a year while accepted articles grew 4.9% a year, 2013 to 2017. Same report.

70.6% of reviewers who decline cite the paper being outside their expertise, and 42% cite their own workload. 2018 survey of 11,838 reviewers, same report.

About 10% of reviewers perform about 50% of all reviews. 2018, same report.

20% of researchers perform between 69% and 94% of reviews in a simulation over 1.1 million MEDLINE articles, with the top 5% supplying 30% of all reviewer hours. 2015, Kovanis et al., PLOS ONE.

The United States supplied 32.9% of reviews against 25.4% of published articles, while China supplied 8.8% of reviews against 13.8% of articles. 2013 to 2017, Global State of Peer Review 2018.

38% of reviews paired an editor and a reviewer from the same region, against 14% expected by chance. Same report.

Invitation acceptance at one family-medicine journal fell from 44.7% in 2017 to 38.6% by July 2025 across 2,951 invitations. Morley et al., PRiMER 2025.

Peer review is donated labour, and the only rigorous estimate of its scale models the total from article counts and self-reported review times rather than measuring it directly. The authors are explicit that the six hours per review is an assumption.

130.8 million hours of reviewer time on journal peer review in 2020, from 21.8 million reviews at an assumed 6 hours each. That is roughly 15,000 years of work. Aczel, Szaszi and Holcombe, Research Integrity and Peer Review 2021.

USD 1.51 billion of salary-equivalent value for US-based reviewers alone, with USD 626 million for China and USD 391 million for the United Kingdom. Same study, 2020.

4.73 reviews per reviewer per year in the same model.

5 hours median self-reported per review. 2016, Global State of Peer Review 2018.

About 47% more articles published in 2022 than in 2016, against limited growth in the number of practising scientists. Hanson et al., Quantitative Science Studies 2024.

25,648 active journals in Scopus as of October 2020, and more than 48,000 active peer-reviewed journals in Ulrich's as of September 2021. STM Global Brief 2021.

This is the best-replicated finding on this page, and it matters directly to anyone filling in the suggested reviewers field. Reviewers nominated by authors recommend acceptance more often than reviewers chosen by editors, across every study we could find that tested it, while the quality of the reports themselves does not differ measurably.

61% of journals let authors recommend reviewers both for and against, while 23% give authors no routine influence. 2019 survey of 322 editors, Hamilton et al., eLife 2020.

Author-nominated reviewers gave 47 accept and 10 reject recommendations where editor-chosen reviewers gave 35 accept and 23 reject on the same 100 submissions, with no difference in review quality. 2003 to 2004, Wager, Parkin and Tamber, BMC Medicine 2006.

Odds ratio of 1.64 for recommending acceptance, author-suggested against editor-suggested, over 788 reviews of 329 manuscripts at 10 biomedical journals. Schroter et al., JAMA 2006.

Author-suggested reviewers were more favourable on 22% of manuscripts against 11% the other way, over 1,145 reviews of 552 manuscripts submitted 2001 to 2006. Bornmann and Daniel, PLOS ONE 2010.

Manuscripts reviewed entirely by author-suggested reviewers were accepted 89% of the time against 67% for those reviewed entirely by editor-suggested reviewers, over 7,965 manuscripts and 23,964 reviews. Author-suggested reviewers also accepted invitations less often, 45.6% against 59.2%. 2011 to 2012, Acuna et al., PLOS ONE 2022.

Editors used 70% of author-suggested names at one journal, where authors suggested reviewers on 73% of submissions. 2008 to 2009, Moore, Neilson and Siegel, JASN 2011.

What follows from this

editors know about the effect, which is the reason your list is screened rather than used. Naming people you can defend is the response, and the conflict of interest guide covers what defensible means.

The figures below were computed from the full Retraction Watch database, released publicly through Crossref, as it stood on 13 August 2026 (71,778 records). They are counts of retraction records carrying a given reason, not of confirmed misconduct findings.

21,212 retractions, 32.0% of all retractions ever recorded, carry a reason of compromised peer review or concerns about peer review. Retraction Watch database via Crossref.

11,710 retractions carry a paper mill reason, and 779 carry a conflict of interest reason. Same source.

13,214 retractions were issued in 2023 alone, of which 78.8% carried a peer-review manipulation reason. Same source. More than 10,000 papers were retracted in calendar 2023, a record, per Nature news.

9,673 of those 2023 retractions, 73% of the year's total, came from a single publisher, Hindawi. Same dataset. Wiley and Hindawi announced 511 retractions in September 2022 and about 1,200 more in April 2023 for compromised peer review, per Retraction Watch.

Between 2% and 46% of submitted papers were flagged as suspected paper-mill output depending on the publisher, with 2% the baseline seen by essentially every journal, across more than 53,000 papers at six publishers. 2019 to 2021 data, COPE and STM paper mills report, 2022.

The mechanism in most of these cases is a fabricated reviewer identity, usually a real researcher's name attached to an email address the submitter controls. That is why editors check the email domain against the stated affiliation.

Several widely quoted figures could not be traced to a primary source, so they are absent. Listing them is more useful than repeating them.

Portfolio-wide acceptance or desk rejection rates for Wiley, Elsevier, Springer Nature and Taylor & Francis. They do not publish them, and the numbers that circulate have no traceable origin.

The BMJ acceptance rate, often quoted as 7%, which we could trace only to search-optimised blog posts.

The figure of 68.5 million hours a year of reviewing attributed to Publons, which does not appear anywhere in that report. The Aczel et al. estimate above is the sourced alternative.

How much editor time is spent finding reviewers. Only qualitative statements exist, such as the 75% of editors who called it the hardest part of the job.

How often author-suggested reviewers are invited across publishers. Only single-journal data exists, given above.

Publisher-scale data on the number of review rounds. The only round-count data we found is the single-journal PLOS ONE sample cited in the timeline guide.