Mind-blowing, outrageous criminal fraud is rare in scientific research, but fraudulent practices such as turning a blind eye to contradictory data or failing to report anomalies are commonplace in laboratories worldwide. The building tide of retractions and the growing need to see all data, including anomalies and negative results, for big data experiments has brought these activities into the limelight, and preventing fraud in research has become an important issue for modern science.
Medical trials are frequently brought up in discussions about scientific fraud. It is easy to see why they are publicised, as inaccurate trials can cause thousands of people to have unnecessary or even harmful treatment. But fraud occurs in all science, and in non-medical research there are less emotive, but still serious, consequences of fraudulent activity.
I may be naïve, but I think in plant science and other areas, where the stakes are not as high as in medical research, most scientists do not deliberately reject data and only publish the minority of results that fit a favourite hypothesis. They do actively bury projects that just don’t go anywhere, as evidenced by many frustrated PhD students with a thesis full of negative results but no publications. As Ed Yong put it in the SpotOn London session ‘Fixing the Fraud,’ negative results are becoming an endangered species.
Causing someone to unwittingly replicate doomed experiments because you did not publish your perceived failed experiment may lead to wasted time and effort, but I would hesitate to call it fraud. An excellent example is described by Jim Caryl in his SciLogs blog The Gene Gym. Jim recently published his finding that a class of tetracyclin resistant genes identified in 1996 was actually a plasmid replication gene, and did not confer any kind of antibiotic resistance. The original authors did not set out to deceive, and scientists who used the gene in their research must have had negative results, which they did not publish.
Another bad practice which I suspect is fairly common, to varying extents, is biased analysis of results. While this is definitely fraud, it has been overlooked up until now because of the need for conclusive, statistically significant data for papers. Again, Jim Caryl is an example – he struggled to get his important negative result published. Over the last decade, big data experiments requiring raw datasets have become the norm, and authors are usually obligated by their funders or by publishers to provide all their raw data. Yet frequently, datasets are not added to a suitable open access repository (Piowar, 2011), and a 2009 study showed that the rules governing open data are poorly enforced (Savage and Vickers, 2009).
At the SpotOn London session on Fixing the Fraud, the panelists briefly discussed the causes of fraud, pointing at the usual, and probably completely responsible, culprits: the pressure on researchers to publish in the most respected journal they can manage, and the requirements of journals who publish only positive results that point to significant, clear conclusions. (more…)