For years, whistle-blowers have warned that fake results are sneaking into the scientific literature at an increasing pace. A new statistical analysis backs up the concern.
A team of researchers found evidence of shady organizations churning out fake or low-quality studies on an industrial scale. And their output is rising fast, threatening the integrity of many fields.
“If these trends are not stopped, science is going to be destroyed,” said Luís A. Nunes Amaral, a data scientist at Northwestern University and an author of the study, which was published in the Proceedings of the National Academy of Sciences on Monday.
Science has made huge advances over the past few centuries only because new generations of scientists could read about the accomplishments of previous ones. Each time a new paper is published, other scientists can explore the findings and think about how to make their own discoveries.
“Science relies on trusting what others did, so you do not have to repeat everything,” Dr. Amaral said.
By the 2010s, journal editors and watchdog organizations were warning that this trust was under threat. They flagged a growing number of papers with fabricated data and doctored images. In the years that followed, the factors driving this increase grew more intense.
As more graduate students were trained in labs, the competition for a limited number of research jobs sharpened. High-profile papers became essential for success, not just for landing a job, but also for getting promotions and grants.
Academic publishers have responded to the demand by opening thousands of new scientific journals every year. “All of the incentives are for publishers to publish more and more,” said Dr. Ivan Oransky, the executive director of the Center for Scientific Integrity.
Organizations known as paper mills are now turning scientific fraud into a lucrative business. Scientists eager to pad out their resumes can pay hundreds to thousands of dollars to be named as an author of a paper that they had nothing to do with, according to Anna Abalkina, a social scientist at Free University of Berlin who studies paper mills.
The manuscript might be provided to the paper mill by a dishonest scientist for a price; in other cases, it might be generated in house. To ensure the papers get published, paper mills sometimes offer bribes to corrupt editors, according to an investigation by the Center for Scientific Integrity.
Dr. Abalkina said that such papers are typically riddled with fraud — everything from doctored images to plagiarized text. To avoid plagiarism detectors, paper mills often use artificial intelligence to alter the text they lift from other papers, sometimes introducing bizarre phrasing such as “bogus upside” instead of “false positive.”
Even as paper mills have worked to keep their efforts hidden, Dr. Abalkina has traced the output of companies in Russia, Iran and other countries, and found thousands of their papers in print. “You learn to see the patterns,” she said.
Dr. Amaral and his colleagues have now analyzed those patterns, using network theory and other statistical techniques. “We tried to give a picture of what’s below the surface,” said Reese Richardson, a postdoctoral researcher at Northwestern University and an author of the new study.
For their analysis, the scientists built a database of more than a million scientific papers. They searched for the papers in online forums where sleuths share duplicated images and tortured phrases, as well as the Retraction Watch Database, maintained by the Center for Scientific Integrity.
The researchers compiled a list of 30,000 papers that have either been retracted or show signs of having come from a paper mill. They discovered connections between the papers that strongly hinted that they were the product of large-scale fraud. Many of these connections linked clusters of editors and authors who often worked together.
“There are huge networks that are very densely connected, where they’re all sending their papers to one another,” Dr. Richardson said. “If that’s not collusion, I don’t know what is.”
The scientists found more evidence of paper mills by looking at duplicated images. Some papers contained images copied from more than one other paper. Mapping the connections between them, the researchers charted networks of thousands of papers. The papers in a cluster all tended to be published in the same short window of time, often in journals put out by a single publisher.
The best explanation Dr. Amaral sees for the pattern of this network is that paper mills are creating banks of images that they use to create entire batches of papers, which they then peddle to certain corrupt editors. After a while, the paper mills make new images and find new targets.
The papers that Dr. Amaral and his colleagues could study came to light only because of the work of independent sleuths. To estimate how many paper mill papers have yet to be exposed, Dr. Amaral’s team created a statistical model that accurately predicted the rate at which suspicious papers surfaced. They estimate that the number of paper mill products may be 100 times greater than the ones they have identified.
Elisabeth Bik, a California-based expert on scientific fraud who was not involved in the study, said that it confirmed her early suspicions. “It’s fantastic to see all the work we’ve done now solidified into a much higher-level analysis,” she said.
Dr. Amaral and his colleagues warn that fraud is growing exponentially. In their new study, they calculated that the number of suspicious new papers appearing each year was doubling every 1.5 years. That’s far faster than the increase of scientific papers overall, which is doubling every 15 years.
“It’s already a problem, and I foresee a crisis,” Dr. Abalkina said.
Dr. Amaral and his colleagues found evidence that paper mills are selectively targeting certain fields to publish dubious papers. The team compared research on different versions of RNA, a molecule that has many roles in the cell. Papers on a form of RNA called microRNA and its role in cancer were much more likely to show signs of possible fraud than other RNA-related fields, such as the gene editing technology CRISPR.
But Dr. Amaral suspects that paper mills will eventually turn their attention to other fields as well.
“The risk is for more and more areas of science to become poisoned, so that no respectable scientist will enter them because the field is so flooded with junk,” Dr. Amaral said.
Artificial intelligence is likely to make things worse, Dr. Abalkina predicted. Instead of doctoring an existing image, paper mills can now use A.I. tools to create images on demand. Dr. Abalkina and her colleagues have tested some of these tools and found that they can already fabricate images that can’t be recognized as fake.
“It’s really scary,” she said.
In an executive order in May on “gold-standard science,” President Trump drew attention to the problem of scientific fraud. “The falsification of data by leading researchers has led to high-profile retractions of federally funded research,” the order stated.
But the administration has not offered any new initiatives to address the problem. Thousands of scientists have protested the order, arguing that it would lead to the political muzzling of genuine scientific findings.
Dr. Richardson and his colleagues said that the administration’s cuts to science funding in the United States could even drive up the rate of scientific fraud by leaving scientists scrambling for limited resources and dwindling jobs.
“The stuff that the Trump administration is doing under the name of upholding scientific integrity is going to make this competition worse,” Dr. Richardson said.
The National Cancer Institute, for example, announced last month that it will fund only the top 4 percent of grant applications for the rest of the fiscal year, instead of the top 7 percent. The Trump administration has proposed even more sweeping cuts that Congress is now considering.
Dr. Richardson warned that these cuts would make it harder for American scientists to put in the long-term work that genuine scientific discoveries require.
“When people are under immense pressure, that’s when these things happen,” Dr. Richardson said.
Ordinary measures, such as retracting papers, are not sufficient to stem the tide, Dr. Amaral said. “It’s like I have a huge faucet putting out water, and I have a little cup that I’m using to take the water out,” he said. “It’s just not going to work.”
Dr. Bik proposed that scientific publishers dedicate more of their profits to monitoring manuscripts for fraud, similar to how credit card companies check for suspicious purchases.
Dr. Amaral suggested that scientists who commit misconduct be temporarily banned from publishing. That would include prominent scientists who put their names on shoddy papers that they had not checked carefully for errors.
“Making authors fully responsible for the research published under their name would decrease the publication rate,” Dr. Amaral said. “That’s something that needs to happen, so we can focus on quality instead of quantity.”
Dr. Oransky said that the way scientists are rewarded for their work would have to change as well. “To paraphrase James Carville, it’s the incentives, stupid,” he said. “We need to stop making it profitable to game the system.”