Research Paper Retraction Statistics 2026: Rates, Causes, and Global Trends
The research paper retraction statistics of the past decade tell a story of an accelerating crisis in scholarly publishing. The Retraction Watch database — now integrated with Crossref — crossed 63,000 total logged retractions by the end of 2025, and annual retraction volumes have grown at a compound rate of 22.09% since 2000, more than three times faster than the overall growth in published research. Understanding what drives these numbers matters for every researcher, thesis writer, and institution that relies on the scientific record.
The peak came in 2023, when more than 14,000 retraction notices were issued in a single year — a figure that would have been unimaginable in 2000, when fewer than 200 retractions appeared annually. While some of this growth reflects improved detection mechanisms and greater willingness by journals to act, the dominant drivers are systemic: paper mills, fake peer review networks, and large-scale data manipulation. This article consolidates the best available data across these causes, geographies, disciplines, and timelines.
For students writing theses or dissertations, these statistics carry direct implications — every source you cite may one day require re-verification, and understanding how retractions happen is part of maintaining your own scholarly integrity. Understanding what peer review is supposed to do — and how long it takes — is equally important context; see our journal peer review time statistics for 2026 for the full picture of how manuscripts move through the editorial process before any retraction risk arises.
Database Size and Annual Retraction Volume
Video: Why Is There So Much Fraud in Academia? — Freakonomics Radio Network (Ep. 572)
The Retraction Watch database is the most comprehensive public repository of retracted scholarly works. As of December 2025, it contained approximately 63,000 retraction records — up from just shy of 55,000 entries reported in early 2024. The database is jointly maintained by the Center for Scientific Integrity and integrated with Crossref metadata infrastructure, meaning retraction notices now propagate automatically to bibliographic platforms.
Annual retraction volumes show a clear upward trajectory, though with notable year-to-year variation:
| Year | Retraction notices issued (approx.) | Notable driver |
|---|---|---|
| 2000 | ~140 | Baseline tracking |
| 2013 | ~1,600 | Expanded monitoring |
| 2022 | ~12,000 | Paper mill wave |
| 2023 | >14,000 (record) | Coordinated mass retractions |
| 2024 | >9,000 | Ongoing mill cleanup |
| 2025 (to Aug) | >5,000 | Continued detection |
Sources: Prevalence and Trends in Global Retractions (arXiv, 2025); Retraction Watch 2024 year-in-review.
The Retraction Watch Leaderboard
Live database: researchers ranked by retraction count — updated continuously by the Center for Scientific Integrity
The 22.09% compound annual growth rate for retracted papers (2000–2021) versus 6.25% for all published papers is the single most alarming summary statistic in this field. Retractions are not simply keeping pace with publication volume — they are outrunning it by a factor of more than three.
Causes: Fraud, Error, and Misconduct Breakdown
A foundational 2012 study published in PNAS analysed 2,047 biomedical and life-science articles retracted from PubMed-indexed journals and found that only 21.3% were attributable to honest error. The remaining 67.4% involved misconduct of some kind:
| Cause | Share of retractions |
|---|---|
| Fraud or suspected fraud | 43.4% |
| Duplicate publication | 14.2% |
| Plagiarism | 9.8% |
| Honest error | 21.3% |
| Unknown / other | ~11.3% |
A more recent 50-year analysis of medical retractions (1975–2024) covering 16,041 documents identified a somewhat different breakdown, reflecting the rise of newer misconduct categories. Using 68 primary cause categories, the leading reasons were:
- Data concerns — 31.47%
- Fraud — 11.37%
- Peer review issues (including fake peer review) — 11.21%
- Referencing issues — 7.54%
- Ethical violations — 7.09%
Critically, the same 50-year study found that data integrity issues and fraud approximately double every 5.5 and 5.2 years respectively. This trajectory suggests that the causes of retraction are not static — each decade introduces new vectors, from digital image manipulation in the 2000s to AI-assisted fabrication emerging in the 2020s.
Paper Mills and Fake Peer Review
The single largest structural driver of the post-2015 retraction surge is the paper mill industry. Paper mills are commercial operations that sell authorships, fabricate data, and systematically manipulate the peer review process. As of May 2024, Retraction Watch had recorded 7,275 retractions attributable to paper mills — out of approximately 44,000 total retractions at that snapshot date, representing roughly one in six retractions in the database.
The scale of the underlying problem is far larger than the retraction count suggests. Researchers estimate that up to 400,000 paper mill articles have infiltrated the scientific literature over the past two decades, the vast majority of which remain unretracted and continue to be cited.
The volume of paper-mill-related retractions grew steeply from a low base: from approximately 10 in 2019 to over 2,099 in 2023 — a 200-fold increase in four years. Source: Phys.org / The Conversation, “Paper mills: the ‘cartel-like’ companies behind fraudulent scientific journals” (2024).
Journal-level data on paper mill exposure (cross-sectional study, 2020–2022):
| Journals with paper mill retractions | Count | Share |
|---|---|---|
| 1 retraction | 71 | 50% |
| 2–9 retractions | 36 | 25.4% |
| 10 or more retractions | 35 | 24.6% |
| More than 100 retractions | 4 | 2.8% |
Fake peer review — where paper mills recommend fictitious reviewers or real experts with falsified contact emails — has grown at an average rate 5.5 times faster than the baseline retraction rate. This single mechanism now accounts for 11.21% of all medical retractions, according to the 50-year JKMS analysis cited above. The COPE (Committee on Publication Ethics) updated its retraction guidelines in 2025 specifically to address paper mills, fake reviewers, and third-party submission manipulation. Understanding where AI-generated content fits into this landscape is also relevant — see AI detection accuracy statistics 2026 for the data on how tools currently perform when trying to identify AI-written or manipulated text.
The same fake-peer-review ecosystem feeds directly into the predatory journal sector. For a quantified overview of how many such journals exist and who publishes in them, see our predatory journal statistics for 2026, which documents the 20,274 titles flagged by Cabells and the ~420,000 articles published in non-peer-reviewed outlets annually.
Retraction Rates by Academic Discipline
Retractions are not evenly distributed across fields. Analysis of 55,000+ retracted papers (as of August 2025) against their Web of Science topic assignments reveals sharp disciplinary variation in retractions per 10,000 publications:
| Discipline | Retractions per 10,000 pubs | Risk level |
|---|---|---|
| Electrical Engineering & Computer Science | 31.97 | Highest |
| Clinical & Life Sciences | 13.85 | Elevated |
| Global average (all fields) | 11.04 | — |
| Physics | 3.25 | Lower |
Source: “Prevalence and Trends in Global Retractions Explored Through a Topic Lens,” arXiv (2025).
The highest-volume retracted meso-topics within the database are:
- Micro and long noncoding RNA — 3,123 retractions
- Human-Computer Interaction — 2,137 retractions (emerging hotspot; publication growth rate of 48.7% per year drove a retraction rate of 3,201 per 10,000 — 78 times the global average)
- Artificial Intelligence and Machine Learning — 1,389 retractions
Within medicine, the most-affected disciplines over the 50-year period (1975–2024) were oncology (19.87% of medical retractions), cardiovascular medicine (15.62%), and pharmacology (14.49%). Source: JKMS 2025 analysis cited above.
For researchers building bibliographies, this disciplinary breakdown matters. When you cite rapidly-growing technical fields, the probability that any individual paper was later retracted is meaningfully higher. Tools that verify citations at the point of writing — see our overview of best AI citation checkers for catching fake and hallucinated references — can flag retracted sources before they enter your thesis.
Geographic Distribution of Retractions
Country-level analysis of retractions reveals two distinct patterns that must be interpreted carefully: absolute counts (which correlate heavily with research output volume) and misconduct-adjusted rates (retractions as a proportion of total publications from each country).
A 2025 study published in the Journal of Medical Internet Research analysed 79,645,579 publications across 100 countries from 1996 to 2023, cross-referencing against 37,858 retractions in the Retraction Watch database.
| Country | Total retractions (all fields) | Medical retractions |
|---|---|---|
| China | 17,541 | 3,997 |
| United States | 3,006 | 1,216 |
| India | 2,950 | 769 |
China’s dominance in absolute counts partly reflects its position as the world’s largest producer of scientific papers. The 50-year JKMS analysis found that nearly 45.28% of all retracted medical publications had at least one Chinese author — a figure that must be read alongside the disproportionate growth of Chinese research output during the same period.
When proportional rates are calculated against publication volume, a different set of countries emerges. Countries with the highest misconduct-adjusted retraction rates overall include Ethiopia (0.35%), Kazakhstan (0.22%), and Saudi Arabia (0.19%). In medicine specifically, Bahrain, Saudi Arabia, and Sudan rank highest. These patterns point to structural pressures in certain academic systems — “publish or perish” incentives, limited institutional oversight, and in some regions, the concentration of paper mill activity.
Time Lag: How Long Before a Paper Is Retracted?
One of the most consequential aspects of the retraction problem is the delay between publication and retraction. The 50-year medical retraction study reported a median time to retraction of 562 days (approximately 18.7 months), with a range from 1 day to over 80 years. This median masks enormous variation by cause:
- Fraud-related retractions take significantly longer — the 2012 PNAS study reported a mean of 46.8 months for fraud versus 20.0 months for error-based retractions. Fraud is harder to confirm, often requiring institutional investigation.
- AI-related paper retractions show a median of 510 days (approximately 17 months) to retraction, according to a 2024 analysis of AI-topic retractions (PMC12624210).
- Fake peer review averaged 15.08 months from publication to retraction notice.
- Authorship disputes averaged 12.63 months.
The practical implication: a paper that appears clean in a literature search today may be retracted within the next one to three years. For thesis writers, this is a real risk — particularly in fast-moving fields such as biomedicine, machine learning, and engineering. Verifying the current retraction status of key citations is part of good research practice. Our data on thesis citation counts by degree level and discipline provides context for how deeply-sourced graduate work typically is, and therefore how exposed it may be to uncorrected retracted sources.
Post-Retraction Citations: The Zombie Paper Problem
Retraction does not stop a paper from being cited. Multiple studies have documented widespread post-retraction citation, often with no acknowledgement of the retraction in the citing text.
Key findings from the literature:
- Among 13,252 post-retraction citation contexts in biomedical literature, only 722 (5.4%) acknowledged that the cited paper had been retracted. Source: “Continued use of retracted papers,” Quantitative Science Studies (2021).
- One analysis found that 23% of 34,487 citations to 460 retracted genetics papers accrued post-retraction. Source: Dal-Ré and Ayuso (2021), cited in Teixeira da Silva, “The Citation of Retracted Papers and Impact on the Integrity of the Scientific Biomedical Literature,” Learned Publishing (2025).
- Brown, Bakker, and Theis-Mahon (2022) found that 283 of 1,396 retracted papers were cited 1,096 times in systematic reviews, with 65% of those citations occurring before retraction. Source: Retracted systematic reviews continued to be frequently cited, Journal of Clinical Epidemiology (2022).
- On average, retraction reduces the annual citation frequency of a paper from approximately 5 to 2 per year — but does not eliminate it.
These “zombie papers” pose a direct threat to the integrity of any literature review. A retracted paper cited in 100 subsequent studies can propagate a flawed finding through hundreds of downstream publications before the error is noticed and corrected.
The problem is compounded by inconsistent metadata propagation. Even though Crossref and PubMed now flag retracted articles, many researchers access papers through institutional repositories, ResearchGate, or direct Google Scholar links that do not display retraction notices prominently. Understanding what is in the citation count data behind your bibliography — and what may already be compromised — is a practical skill every graduate researcher needs. The broader landscape of what researchers actually include in academic work is covered in our analysis of AI in academic writing statistics 2026.
Publisher-Level Data: Springer Nature in 2024
Springer Nature, the world’s largest academic publisher by journal count, retracted 2,923 papers in 2024 from its portfolio of over 3,000 journals — which published more than 482,000 articles that year. This represents a retraction rate of approximately 0.61% of published output.
The breakdown by publication timing is instructive:
- 61.5% (1,797 articles) were originally published before January 2023 — older content being cleaned up retroactively
- 38.5% (1,126 articles) were published after January 2023 — suggesting ongoing contamination despite tighter post-pandemic screening
- 41% of retractions for post-January 2023 content involved open access articles
Flagged journals within the Springer Nature portfolio included Environmental Science and Pollution Research (cited for “suspicious citations, tortured phrases and undisclosed use of AI”), Scientific Reports, and Applied Nanoscience. Source: Retraction Watch, “Springer Nature retracted 2,923 papers last year” (February 2025).
“Tortured phrases” — recognisable algorithmic paraphrasing used to evade plagiarism detection — are now a documented marker of paper mill output. The phrase “breast cancer” becomes “bosom malignant growth”; “artificial neural network” becomes “counterfeit neural organisation.” Detecting this kind of content is now an active area of journal screening. For context on how AI detection accuracy compares across tools, see AI detection accuracy statistics 2026.
The open-access angle is worth noting separately: 41% of post-2023 Springer Nature retractions involved OA articles. This aligns with the finding that some paper mills specifically target open access journals with article processing charge (APC) models, exploiting the pay-to-publish structure. The broader economics of open access publishing are covered in our data report on open access publishing statistics 2026.
What These Statistics Mean for Thesis Writers
For students writing graduate theses or dissertations, research paper retraction statistics translate into concrete, practical risks.
1. Verify the status of every primary source. Check your cited papers in the Retraction Watch database or through PubMed’s retraction filter before submission. Given that 5.4% of post-retraction citation contexts show any awareness of the retraction, most citing authors do not check.
2. Be more cautious in high-retraction disciplines. Electrical engineering, computer science, and the biomedical sciences carry retraction rates 2–3 times the global average. The faster a field is publishing, the more likely individual papers are unvetted paper mill products or contain undetected errors.
3. Time lag creates a hidden risk window. Papers published in the last 18–24 months in your field may be retracted before your thesis is published or examined. In fields where retraction takes 46 months on average for fraud cases, a paper you cite today could be retracted well after your viva.
4. Literature review tools matter. A tool like Tesify — Write Your Thesis with AI helps you structure and critically evaluate sources, which is the first line of defence against inadvertently building arguments on retracted or unreliable foundations.
5. Understand plagiarism as a retraction cause. 9.8% of biomedical retractions involve plagiarism, making it the third most common form of misconduct in the PNAS dataset. Institutions increasingly screen thesis submissions with the same tools used for journals. The standards now applied to submitted theses are not materially different from those applied to published articles.
Frequently Asked Questions
How many research papers have been retracted in total?
The Retraction Watch database contained approximately 63,000 retraction records as of late 2025. This figure grows continuously as new retractions are issued and historical cases are identified. The total number of retracted papers ever published is almost certainly higher, as many retractions — particularly from earlier decades — were never indexed in any centralised database.
What is the most common reason for a research paper to be retracted?
The most common single cause in the biomedical literature is fraud or suspected fraud (43.4% of retractions, PNAS 2012 data). In more recent and broader analyses, “data concerns” (including fabrication, manipulation, and integrity failures) is the leading category at 31.47%. Fake peer review and plagiarism have grown substantially as secondary causes since 2015, with paper mills now driving a significant proportion of annual retraction volumes.
Which country has the most research paper retractions?
China leads in absolute terms with 17,541 retractions recorded across all disciplines in a 1996–2023 JMIR dataset, followed by the United States (3,006) and India (2,950). However, when rates are calculated relative to total publication output, China’s rate (approximately 0.17% of publications) is not the world’s highest — countries including Ethiopia (0.35%), Kazakhstan (0.22%), and Saudi Arabia (0.19%) show higher proportional rates.
How long does it typically take for a paper to be retracted after publication?
The median time to retraction in medical literature is approximately 562 days (18.7 months), but varies significantly by cause. Fraud-related retractions take a mean of 46.8 months because they require formal investigation. Error-based retractions average around 20 months. Fake peer review cases average 15 months. AI-related retractions have a median of 510 days (17 months). The practical implication: any paper published within the past two to four years in a high-risk discipline may still be retracted after you cite it.
Are retracted papers still cited after retraction?
Yes, extensively. Only 5.4% of post-retraction citation contexts in biomedical literature acknowledge that the cited work was retracted. Around 23% of all citations to a set of 460 retracted genetics papers accrued after the retraction notice was published. Annual citations to retracted papers roughly halve after retraction — from about 5 per year to about 2 — but do not drop to zero. This means the scientific record continues to carry the influence of retracted work for many years.
What are paper mills and how do they cause retractions?
Paper mills are commercial services that sell authorship slots, fabricate datasets, and systematically manipulate the peer review process — typically by recommending fictitious reviewers or real researchers whose contact emails have been intercepted. As of 2024, paper mills had generated an estimated 400,000 articles now embedded in the scientific literature, with confirmed paper-mill retractions growing from approximately 10 in 2019 to over 2,099 in 2023. The COPE retraction guidelines were updated in 2025 to specifically address paper mill activity and third-party submission manipulation.
References
- Fang FC, Steen RG, Casadevall A. “Misconduct accounts for the majority of retracted scientific publications.” PNAS, 2012. PMC3479492
- “Prevalence and Trends in Global Retractions Explored Through a Topic Lens.” arXiv, 2025. arXiv:2511.21176
- “Fifty Years of Retracted Medical Publications From 1975 to 2024.” Journal of Korean Medical Science, 2025. JKMS 2025.40.e300
- “Geographical Disparities in Research Misconduct: Analyzing Retraction Patterns by Country.” JMIR, 2025. PMC11775486
- “Springer Nature retracted 2,923 papers last year.” Retraction Watch, February 2025. retractionwatch.com
- “A look back at 2024 at Retraction Watch.” Retraction Watch, December 2024. retractionwatch.com
- “Continued use of retracted papers: Temporal trends in citations in biomedicine.” Quantitative Science Studies, 2021. MIT Press
- Teixeira da Silva JA. “The Citation of Retracted Papers and Impact on the Integrity of the Scientific Biomedical Literature.” Learned Publishing, 2025. Wiley
- “Retracted systematic reviews continued to be frequently cited.” Journal of Clinical Epidemiology, 2022. ScienceDirect
- “Identifying common patterns in journals that retracted papers from paper mills.” Research Integrity and Peer Review, 2025. Springer Nature
- “Paper mills: the ‘cartel-like’ companies behind fraudulent scientific journals.” The Conversation / Phys.org, 2024. Phys.org
Write your thesis with AI
Structure, draft, cite, and format your thesis faster with Tesify’s AI writing tools, automatic bibliography, and plagiarism checker. Free to start, no credit card required.
Leave a Reply