A sharp rise in higher education retractions

The earliest article in the dataset was published in 2010 and retracted in 2013 after an analytical error. Retractions remained relatively uncommon during most of the 2010s, averaging approximately three per year. The pattern changed dramatically after 2020.

The number reached 19 in 2021 and climbed to 158 in 2023 before falling to 52 in 2024. This does not necessarily mean that misconduct suddenly appeared in 2023. Retraction is a delayed institutional response. Papers must first be published, questioned, investigated and formally withdrawn.

The study found that the growth of retractions followed an earlier expansion in higher education publications. Publication output rose strongly in 2021 and 2022, while the corresponding retraction surge appeared roughly one year later. This sequence suggests that unusually rapid growth placed additional pressure on editorial screening and later generated waves of integrity investigations.

A rise in retractions can indicate more misconduct, better detection or both. Retraction statistics measure the failures that institutions found, not every failure that entered the literature.

This distinction is essential when interpreting national or journal-level comparisons. A publisher that performs a rigorous retrospective investigation may report more retractions than a publisher that rarely examines its archive. A high count can therefore reflect both vulnerability and a willingness to correct the record.

Why China dominates the dataset

China accounted for 238 of the 283 retracted articles, or approximately 84 percent of the dataset. The remaining papers were distributed across countries including Kazakhstan, Saudi Arabia, Iran, Singapore and Thailand.

The scale of the Chinese figure is striking, but it requires careful interpretation. China is also one of Asia’s largest producers of academic research. A country that publishes far more papers than its neighbours will normally generate a larger absolute number of retractions, even before differences in misconduct rates are considered.

The study also discusses institutional pressures that may increase risk. Universities across parts of Asia have used publication counts, journal rankings and cash rewards as indicators of performance. When employment, promotion, graduation or financial bonuses depend heavily on publication, the demand for rapid and prestigious output can become stronger than the capacity to produce it responsibly.

Paper mills thrive in this gap. They provide an illicit route for researchers who need a publication but lack the time, training, data or incentive to complete genuine research. The resulting problem should not be reduced to national culture. It is better understood as a market created by institutional demand and enabled by weaknesses in academic publishing.

Paper mills were the leading recorded cause

Paper mills were associated with 124 articles, making them the most frequently recorded reason for retraction. A paper mill produces or arranges manuscripts for sale. The person who purchases the paper may be listed as an author despite having made little or no genuine intellectual contribution.

These operations can manufacture data, recycle text, manipulate images, invent participants, insert paying authors or submit large numbers of formulaic papers to journals with vulnerable review systems. A paper-mill article may look superficially complete because the abstract, statistical language and reference list have been designed to imitate legitimate research.

Research ethics violations were recorded in 83 papers. These included 44 cases involving absent or unverified institutional ethics approval, 37 involving missing or withdrawn informed consent and two involving missing third-party approval.

Approximately 29 percent of the papers associated with paper mills also had a research ethics problem. This overlap is logical. A manufactured paper may claim to have studied students, patients or other participants even though no legitimate ethics process was completed.

Peer review became part of the misconduct

Seventy-seven papers were associated with manipulation of peer review. The study divided these cases into 52 involving fake peer review and 25 involving a rogue editor.

Fake peer review can occur when authors provide fabricated reviewer identities, false email addresses or recommended reviewers controlled by themselves or their associates. The journal believes that it is receiving an independent assessment, while the manuscript is effectively being reviewed by someone connected to its authors.

A rogue editor can compromise the process from inside the journal by assigning unsuitable reviewers, bypassing normal scrutiny or coordinating the acceptance of papers that should not have passed peer review. These cases show why research integrity cannot be protected by checking the manuscript alone. Journals must also verify the people and communications surrounding it.

Interestingly, only six paper-mill cases were also coded as fake peer review, while no paper was simultaneously categorized as both paper-mill output and plagiarism. The limited overlap may indicate genuinely different misconduct pathways. It may also reflect the way publishers select a principal reason when several warning signs are present.

Retraction notices are not always detailed enough to reveal the complete chain of misconduct. A paper may be categorized under the clearest documented failure even when other irregularities are suspected but not formally proven.

Retractions were concentrated in a small group of journals

The 283 articles appeared across 70 journals. The distribution was not even. A small number of titles accounted for a large share of the retractions, with many published by Hindawi.

Mathematical Problems in Engineering accounted for 34 papers, while the Journal of Sensors accounted for 27. Journal of Healthcare Engineering contributed 19, Arabian Journal of Geosciences contributed 18 and Discrete Dynamics in Nature and Society contributed 14.

These titles demonstrate the interdisciplinary character of higher education research. Many retracted studies were not published in journals devoted primarily to education. They appeared in mathematics, engineering, computing, healthcare and geoscience journals while using a university setting or educational topic.

Q3 journals contained 151 papers, approximately 53 percent of all articles in the dataset. Q2 journals contained 87 and Q1 journals contained 45. No Q4 articles were identified.

The absence of Q4 papers should not be interpreted as evidence that lower-ranked journals are free from misconduct. The review was limited to Scopus and Web of Science, which do not include every journal. Some lower-ranked or locally published titles may therefore fall outside the databases used to construct the sample.

The review also found an equal split between open-access and subscription-based journals among the 70 titles. This is important because it challenges a common assumption that research misconduct is primarily an open-access problem. The vulnerabilities concern editorial systems, incentives and oversight rather than access models alone.

Only seven retracted papers appeared across six specialist higher education journals. None was identified in the six leading English-language higher education journals highlighted by the study. Most problematic papers entered the field indirectly through journals belonging to other disciplines.

Retracted research continued to be cited

The most troubling result concerns what happened after retraction. The 283 papers accumulated 2,067 citations after being withdrawn. Their average citation count increased from 5.38 before retraction to 7.30 afterward. The difference was statistically significant.

One open-access article on e-learning and student outcomes received only four citations before it was retracted for plagiarism in 2013. It later accumulated more than 160 additional citations and continued to appear in research published during the 2020s.

This does not mean that every later citation endorsed the article. Some researchers may have cited it while discussing misconduct or retraction. Nevertheless, the scale and persistence of post-retraction citation show that a publisher’s warning does not automatically remove a paper from the intellectual supply chain.

PDFs circulate through personal libraries, institutional repositories, academic social networks, email attachments and reference-management systems. Once downloaded, they may no longer display a live warning. Researchers can import the citation from an old bibliography without visiting the journal’s current landing page.

Continued citation also has a cumulative effect. A later paper may repeat a claim from a retracted source without citing the original directly. Future researchers then encounter the claim through an apparently valid secondary source. The unreliable finding survives even when the original paper is correctly labelled.

What the data reveal beyond the headline numbers

1. Retraction data measure detection as much as misconduct

A journal with an active research-integrity team may retract more papers because it investigates more papers. A journal with weak oversight may report fewer cases while leaving problematic work untouched. Retraction counts should therefore never be treated as a simple league table of honesty.

2. The system concentrates scrutiny on recent publications

Approximately 54.8 percent of papers were retracted one year after publication, while 99.3 percent were retracted within five years. Rapid detection is valuable, but this concentration suggests a possible legacy blind spot. Older papers may attract less scrutiny even when similar problems remain in the archive.

Journals often investigate batches of recent papers after discovering a compromised editor, reviewer network or paper-mill pattern. Articles published before the investigated period may escape examination because records are incomplete or because the cost of reviewing an entire archive is too high.

3. Paper mills are an organizational problem

Traditional accounts of misconduct focus on an individual who falsifies a result or copies a passage. Paper mills transform misconduct into a service industry. They divide tasks across writers, brokers, data manufacturers, image producers and authorship sellers. Prevention therefore requires network detection, not only individual punishment.

4. Journal concentration can distort regional conclusions

A large share of the papers came from journals affected by coordinated integrity investigations. The resulting geographic pattern may partly reflect the submission markets of those journals and the scope of the publisher’s audit. It should not be generalized automatically to every higher education researcher or institution in Asia.

5. Retraction is not the final stage of correction

The substantial number of post-retraction citations demonstrates that issuing a notice is only the beginning. Correction must also include metadata updates, visible PDF watermarks, repository notifications, citation alerts and examination of later work that relied heavily on the retracted findings.

Research topics most frequently affected

The researchers analysed article titles because author-supplied keywords were unavailable for some papers. Terms connected to education, universities, colleges, teaching and students appeared most frequently, as expected.

More revealing was the concentration of technology-related language. Terms such as data, model, intelligent, network, algorithm, artificial and technology appeared repeatedly. Ideological and political education was the most common specific subject area, represented by 29 papers. It was followed by physical education with 17, environmental education with nine and entrepreneurship education with eight.

Highly technical language can give a manuscript the appearance of sophistication. A paper may combine a fashionable algorithm with a university-related topic even when the research question, data and method have little substantive connection. Editors need reviewers who can evaluate both sides of such submissions: the technical method and the higher education context.

Interdisciplinary work is most vulnerable when each reviewer assumes that another expert has checked the unfamiliar part. An education reviewer may trust the machine-learning analysis, while a technical reviewer may trust the educational theory and data collection. Paper mills can exploit that divided responsibility.

The study’s limitations

The findings are significant, but they do not represent every retracted higher education publication in Asia. The review included peer-reviewed journal articles indexed in Scopus or Web of Science. It excluded books, conference papers, reports and non-journal publications.

Relevant research may also appear in databases such as CNKI, PubMed or discipline-specific indexes. Coverage differences can affect which countries, languages and journal types become visible.

The study classified a paper geographically according to the first author’s institutional affiliation. Internationally co-authored research may therefore be associated with one country even when researchers from several regions contributed.

Thirty-six papers could not be matched to a reason in the Retraction Watch Database. All were retracted in 2023 or 2024, suggesting that database updating delays may partly explain the missing information.

Earlier retraction notices were also less informative. One 2015 case had no official notice, while several notices issued between 2015 and 2019 provided little explanation. The study found improvement during the 2020s, but transparent notices remain necessary for researchers seeking to understand why a paper was removed.

What universities, journals and researchers should change

1. Reform publication incentives. Universities should assess research quality, openness, social value and methodological contribution rather than relying mainly on paper counts and journal quartiles.
2. Verify reviewer identities. Journals should use institutional addresses, reviewer databases, identity checks and conflict-of-interest screening before accepting recommended reviewers.
3. Screen submissions as networks. Repeated templates, shared language, recycled images, unusual authorship changes and clusters of submissions should be examined together rather than as isolated manuscripts.
4. Audit special issues and editorial activity. Sudden increases in acceptance, extremely short review times and repeated use of the same reviewers should trigger investigation.
5. Check ethical approval directly. Journals should require identifiable approval numbers and reserve the right to contact ethics committees when documents appear inconsistent.
6. Improve retraction metadata. Notices should state the reason, responsible party, date, evidence and effect on the findings in a form that indexing services can process.
7. Flag retracted references during submission. Editorial systems should warn authors and reviewers when a reference has been retracted and require an explanation when it remains relevant.
8. Examine older literature. Integrity programmes should not end with papers published during the previous five years. Risk-based historical audits can identify older work that escaped earlier detection.

Frequently asked questions

Conclusion

The review of 283 retracted papers reveals a higher education research system facing more than a collection of isolated ethical failures. Paper mills have industrialized manuscript production. Manipulated peer review has weakened the gatekeeping process. Institutional reward structures have created demand for rapid publication, while fragmented databases allow withdrawn research to remain influential.

The high number of cases associated with China deserves attention, but responsible interpretation requires publication volume, database coverage and publisher investigations to be considered. Retraction counts identify documented failures. They do not provide a complete measurement of national research cultures.

The most important warning may be the 2,067 citations recorded after retraction. Academic publishing has developed mechanisms for withdrawing a paper, but it has not yet created an equally effective mechanism for withdrawing that paper’s influence.

Genuine reform must therefore work at three levels. Universities must reduce incentives that reward quantity without sufficient attention to quality. Journals must strengthen identity, ethics and peer-review verification. Research databases and citation tools must ensure that a retraction warning follows the paper wherever it travels.

A retraction corrects a journal record. Research integrity is restored only when the correction reaches the wider system of knowledge.