Scientific fraud is often presented as a problem of individual researchers crossing the line. Think of cases such as Diederik Stapel, more recently Dan Ariely, or even Oliver Sacks. Someone fabricates data. Someone manipulates an image. And then someone writes an article that simply isn’t correct.
But a recent PNAS study by Richardson et al. suggests that the problem may be much, much larger. Not only because fraud may be more common than we think, but because it sometimes happens on an organised, industrial scale through so-called paper mills. These are organisations that produce and sell scientific papers, sometimes with fabricated data or recycled results.
The study starts from a simple observation. Science is a system that largely runs on trust. Researchers build on each other’s work. Journals trust reviewers and editors. Universities and funders assume that published results are reported in good faith.
But the system is increasingly driven by measurable indicators: the number of publications, citations, impact factors, and rankings. That creates pressure to publish and simultaneously makes those indicators attractive targets for manipulation.
A first indication
The researchers, therefore, looked for patterns that are difficult to explain by chance. An analysis of the journal PLOS ONE, for example, showed that a small group of editors had accepted a striking number of papers that were later retracted or criticised on platforms such as PubPeer. Although they represented only a tiny fraction of all editors, they were involved in a large share of the retractions.
The researchers also found that certain authors appeared disproportionately often with those same editors. That could still be a coincidence, but it could also indicate forms of coordination or steering within the publication process.
More
A second indication came from image analyses. In biomedical research, papers often include photographs or microscopic images. The researchers identified thousands of cases where exactly the same images appeared in different publications.
When they mapped those papers, another pattern emerged. The publications tended to cluster: groups of papers appearing around the same time, often with the same publisher.
This is exactly the kind of pattern one would expect if papers were not produced individually, but in larger batches.
Journal hopping
The study also shows something else. Organisations offering such papers adapt their strategies as journals tighten their controls. When a journal disappears from major databases, paper mills simply move their activities to other journals. The authors refer to this as “journal hopping”.
Fraud also appears to be unevenly distributed across fields. In some research areas, retraction rates are clearly higher than in others, even when those fields are closely related.
Perhaps the most striking result concerns the speed at which these publications are growing. According to the study, the number of suspected paper-mill papers is increasing much faster than the overall number of scientific publications.
So what now?
This does not mean that science has suddenly become unreliable. Science has always had mechanisms to correct errors: peer review, replication and, eventually, retractions. But these systems were mainly designed to deal with individual mistakes.
When fraud becomes organised on a larger scale, those mechanisms begin to reach their limits. If the problem becomes more systematic, the way we detect and address it will also need to become more systematic. Long live a site such as Retraction Watch, but clearly more will be needed.
The authors therefore argue that three tasks should be clearly separated: detecting fraud, investigating it, and sanctioning it. At the moment, these roles often overlap, for example, when journals or universities are asked to investigate their own problems.