Scientific retractions reached more than 10,000 research articles in 2023, according to Nature’s analysis. The figures below show how retractions vary by cause, country, field, publication type, and the time papers remain in the literature. The studies cover different populations and periods, so their percentages should be read within their stated scope rather than treated as one universal retraction rate.
Contents
- Global volume and time trends
- Why scientific papers are retracted
- Countries and regional patterns
- Fields, journals, and article types
- Time from publication to retraction
- Retraction notices and record visibility
Global volume and time trends
Nature reported more than 10,000 research articles retracted during 2023, a record annual total. Its analysis also found that the share of papers eventually retracted became more than three times higher over the decade preceding 2023, approximately 2013–2023. These are global research figures and should not be confused with a percentage of all papers published in 2023.
The long historical series in the medical Retraction Watch analysis covered retracted publications dated from 3 July 1975 through 30 December 2024. Within that broader record, a study of papers containing the word “statistics” in Scopus identified 680 retracted publications in the 2005–2024 publication window. That statistics-keyword dataset peaked at 149 retracted papers in 2023.
Some fast-growing research areas show their own distinctive pattern. An AI-focused PubMed analysis identified 764 retracted publications, with retractions occurring from 2003 through 2025. Its highest annual count was 667 retractions in 2023. Because this is an AI-focused dataset, it describes that subject-specific collection rather than all AI research or all scientific publications.
An older study by Fang and colleagues found that the percentage of published articles retracted for fraud or suspected fraud increased nearly tenfold from 1975 to the late-2000s study period. Its sample analyzed retractions through 2010, so it provides historical context rather than a current estimate.
Sources: Nature, “More than 10,000 research papers were retracted in 2023 — a new record”; Fifty Years of Retracted Medical Publications From 1975 to 2024; Retracted Publications Related to Statistics; Analysis of Retracted Publications on Artificial Intelligence; Misconduct accounts for the majority of retracted scientific publications.
Why scientific papers are retracted
The 1975–2024 medical dataset recorded 48,649 reason instances. Data concerns were the largest category, with 15,311 instances, or 31.47%. Within that category, unreliable results accounted for 6,016 instances, or 39.29%, while concerns about data accounted for 4,895, or 31.97% of data-concern reasons.
Fraud represented 5,531 recorded reason instances, or 11.37%. Paper-mill cases made up 3,476 of those fraud instances, or 63.08%. Peer-review issues represented 11.21% of recorded reasons, followed by referencing issues at 7.54% and ethical issues at 7.09%. These are reason instances, not necessarily mutually exclusive papers; one retraction may have more than one recorded reason.
The historical sample from Fang and colleagues reported fraud or suspected fraud as the stated reason for 43.4% of retractions analyzed through 2010. Duplicate publication accounted for 14.2%, and plagiarism for 9.8%. The difference between these figures and the later medical database categories reflects differences in study design, time period, source records, and classification.
Human-participant research shows another view of the problem. Among 1,339 retracted human-participant articles from 1980 through February 2016, misconduct was the reason for 51% and error for 14%. In an otolaryngology sample through 2024, intentional data misconduct was recorded for 176 of 471 papers, or 37.4%. Authors did not respond to or did not agree with the retraction in 103 cases, or 35% of that sample.
Sources: Fifty Years of Retracted Medical Publications From 1975 to 2024; Misconduct accounts for the majority of retracted scientific publications; Human research participant retractions survey; Retractions in Otolaryngology Publications.
Countries and regional patterns
Country comparisons depend heavily on whether they count total retractions or retractions relative to publication output. Across 100 countries in a 1996–2023 study, researchers counted 37,858 retractions among 79,645,579 publications, an overall retraction proportion of 0.048%. For medicine across the same countries and period, there were 10,890 retractions among 23,175,369 publications, or 0.047%.
| Comparison | Count or rate | Period and scope |
|---|---|---|
| China, overall misconduct-related retractions | 17,541 | 1996–2023 |
| United States, overall misconduct-related retractions | 3,006 | 1996–2023 |
| India, overall misconduct-related retractions | 2,950 | 1996–2023 |
| Highest overall proportion: Ethiopia | 0.35% | 1996–2023 |
| Highest medical proportion: Bahrain | 0.48% | 1996–2023 |
The same country analysis placed Kazakhstan at 0.22% and Saudi Arabia at 0.19% for overall retraction proportions. Pakistan’s overall proportion was 0.18%, and China’s was 0.17%. For medicine, Saudi Arabia reached 0.25%, Sudan 0.24%, Iraq 0.24%, and Egypt 0.21%.
The medical Retraction Watch analysis, covering 1975–2024, attributed 7,264 retracted publications to China, or 45.28% of country-attributed records. India accounted for 1,096, or 6.83%. The AI-focused dataset attributed 551 retracted publications to China, 40 to India, and 23 to Bangladesh. These totals are not directly interchangeable because the datasets have different subject boundaries and dates.
In the human-participant survey covering 1980 through February 2016, Asia represented 36% of retracted articles, Europe 32%, and North America 26%. The United States accounted for 26% of the articles, Japan 11%, and Germany 10%. Separately, the rate of retractions per 10,000 papers in Latin American institutions rose from 0.60 during 1987–2010 to 1.00 during 2011–2024.
Source: Geographical Disparities in Research Misconduct; Fifty Years of Retracted Medical Publications From 1975 to 2024; Human research participant retractions survey; Latin American biomedical retractions.
Fields, journals, and article types
The medical Retraction Watch analysis found that oncology contributed 3,187 retracted records, or 19.87% of the medical dataset. Cardiovascular medicine contributed 2,506, or 15.62%; pharmacology contributed 2,324, or 14.49%; and surgery contributed 1,983, or 12.36%.
The statistics-keyword study had a different subject mix. Engineering was the largest subject area with 232 retracted papers, followed by computer science with 212. It counted 168 in medicine, 88 in mathematics, and 81 in biochemistry, genetics, and molecular biology. These figures describe papers containing “statistics,” not all retractions in those disciplines.
Retractions were also concentrated among some journals. The ten leading journals in the 1975–2024 medical dataset accounted for 16,041 retractions, or 29.87% of the total. Computational and Mathematical Methods in Medicine led that journal list with 948 retracted publications, or 5.91%, followed by Journal of Healthcare Engineering with 938, or 5.85%.
Article type matters as well. In the otolaryngology sample through 2024, research articles made up 372 retractions, or 79.0%, while reviews and meta-analyses made up 63, or 13.4%. In the human-participant survey, the median five-year journal impact factor was 2.94, with an interquartile range of 1.81–4.69.
Sources: Fifty Years of Retracted Medical Publications From 1975 to 2024; Retracted Publications Related to Statistics; Retractions in Otolaryngology Publications; Human research participant retractions survey.
Time from publication to retraction
Retraction statistics measure not only how many papers are removed or marked, but also how long questionable findings may remain available. In the 1975–2024 medical analysis, the median time that articles remained published before retraction was 562 days. The range was 1 to 29,622 days, showing a very wide spread.
The AI-focused dataset reported a median publication-to-retraction interval of 510 days, with a range of 28 to 4,200 days. Its retracted papers had a median citation count of 2, with an observed range of 0–167 citations; citation counts were collected through 2025.
The human-participant survey reported a median author count of 4, with an interquartile range of 3–6 authors. Author count is not a delay measure, but it helps describe the papers in that sample and reinforces why comparisons should retain the study population and date range.
Sources: Fifty Years of Retracted Medical Publications From 1975 to 2024; Analysis of Retracted Publications on Artificial Intelligence; Human research participant retractions survey.
Retraction notices and record visibility
A retraction does not always mean that an article disappears. In the human-participant survey, editors initiated retraction in 59% of articles, or 794 cases. Only 14%, or 181 papers, were completely withdrawn. Among still-available retracted articles, 34% had no indication of retraction, representing 395 papers.
A retraction watermark appeared on 52% of still-available articles, or 599 papers. A separate retraction note appeared on 8%, or 91 papers, and 5% had both a note and watermark, or 63 papers. Among post-August-2010 human-participant retractions, 63%, or 567 cases, were reported by Retraction Watch.
A separate notice-improvement study examined 123 retraction notices from 2008–2016. It found that 99.2% stated a clear reason and 0.8% did not. The original paper mentioned the retraction in 81.3% of cases, did not mention it in 12.2%, and was deleted in 6.5%. In that sample, mistakes and fraud each represented 26.0% of stated reasons, while plagiarism represented 20.3%.
Together, these figures make visibility a measurable part of scientific integrity. The percentages come from specific historical samples, not a current audit of every publisher or database, but they show why readers need to check the article page, the retraction notice, and the indexing record together.
Sources: Human research participant retractions survey; Correcting the literature: Improvement trends seen in contents of retraction notices.