A tide of retractions: The retraction rate of high-momentum papers in China, and the implications for the research-integrity ecosystem
There is significant variation in the data that is highlighted by scatter charts. The show indicates that Chinese universities tend to have a lower rate of admitted students than do Chinese hospitals.
In 2023, as Nature reported, more than 10,000 retraction notices were issued (Nature 624, 479–481; 2023). Most of these were from Hindawi, a now-closed London subsidiary of the publisher Wiley, which found that Hindawi journals were affected by a blizzard of peer-review fraud and sham papers. (Wiley told Nature at the time that it had scaled up its research-integrity teams, put in place more rigorous processes to oversee manuscripts and removed “hundreds” of bad actors, including some guest editors, from its systems). During the past decade, the annual retraction rate — the proportion of published articles in a particular year that have been retracted — has trebled (although fewer retraction notices were issued overall in 2024 than in 2022 or 2023; see ‘A tide of retractions’). As more articles are withdrawn, the proportion will rise to around 1% for papers published in 2022.
The small universities that top the lists are often the most prestigious. In Ghazi University’s case, around half of its total retracted papers are by just four authors.
Some sleuths have speculated that authors from Ethiopia and some other African countries that publish relatively few articles might sometimes have been added to paper-mill products that originated elsewhere, to take advantage of Hindawi’s waived open-access fee for scholars from low- or lower-middle-income countries. A spokesman for a publisher said that they are aware of the use of authorship for sale and waiver manipulation schemes by paper mills.
The retraction data provided to Nature come from research-integrity tools that technology firms have launched over the past two years, which aim to help publishers stem a surge in fake and significantly flawed research. Among a flurry of these software products are Argos, from Scitility in Sparks, Nevada; Signals, from the firm Research Signals in London; and Dimensions Author Check, from the London-based company Digital Science. The latter is part of the majority shareholder in Springer Nature. Nature’s news and features team is editorially independent of its publisher.) Nature was provided with the data from these three firms, as well as countries and journals associated with retracted articles.
But this is not quite as great as the rates in Saudi Arabia and Ethiopia, and also less than those in Iraq and Pakistan. Many Russian retractions do not have a global database and are omitted from the Retraction Watch data. Many countries have lowertraction rates compared to the global average, for example, the United States and the United Kingdom have rates of around 0.04%, much lower than the global average.
Analysing Institutional Affiliations by Creating a Database of Retracted Articles: A Study of Incentivities for Scientific Publishing
Analysing institutions is difficult because there are both data errors and different approaches in the underlying databases. To map institutional affiliations, Dimensions uses a private Global Research Identifier Database (GRID), whereas OpenAlex uses the public Research Organization Registry (ROR). Both contain quirks — affiliations can be missing or wrongly attributed (a particular issue for some smaller institutes) or the database curators might simply have made different choices about how to assign an affiliation. Accordingly, the firms’ analyses vary.
Researchers in public universities and government institutes in India face fewer pressures to publish than do those in private universities and colleges, says Agrawal. Private institutions, he says, push students and researchers to publish many articles, and in some cases pay bonuses for papers published.
Over the past decade, the number of retracted articles by the three firms is more than the Retraction Watch data set they build on. The data might not have been 100% accurate, because articles can be recorded wrong in CrossRef and other online sources, according to a researcher at the University of Illinois.
Their tools aim to alert users to potential ‘red flags’ in research articles or submitted manuscripts, such as authors who might have high numbers of misconduct-associated retractions. To build them, the firms created internal data sets of papers. These are mostly based on Retraction Watch’s database that was acquired byCrossref in the year 2023 in order to public distribute it. It was easy for others to use and analyze the information.
The firm, Scitility, says it will make public its figures later this year. Jan-Erik de Boer, the firm’s co- founder and a former chief information officer at Springer Nature, thinks that scientific publishing will be helped if there is more transparency around this.
“It’s tempting to think about whether differences are related to varying incentives for researchers in different institutions,” says Ivan Oransky, co-founder of the website Retraction Watch, which maintains a public database of retractions on which the companies contacted by Nature partly depend.
Some young physicians at hospitals bought fake manuscripts from paper mills because they were concerned with fraudulent scientific reports to order. Doctors were under pressure to publish papers to get jobs or earn promotions because they had to, says integrity sleuth Elisabeth Bik in California. Sleuths noticed that there were duplicated images in large numbers of papers. They made the issue public and then made multiple retractions.
In general, there is little consistency in how publishers record and communicate retractions — although last year, publishers did agree, under the auspices of the US National Information Standards Organization, on a unified technical standard for doing this, which might help to improve matters.
There are messy online records of work done in the past. For instance, one heavily cited paper in The Lancet is currently categorized as retracted in CrossRef, a site that records data about published articles, when it has in fact only been corrected. This was because of an error by the paper’s publisher, Elsevier (Elsevier did not respond to a request for comment).
New mountain of retractions data should not be ignored. Retractions data are a reminder that quality, as well as quantity, count in science.