“Why Do Scientists Fabricate And Falsify Data?” That’s the start of the title of a new preprint posted on bioRxiv this week by researchers whose names Retraction Watch readers will likely find familiar. Daniele Fanelli, Rodrigo Costas, Ferric Fang (a member of the board of directors of our parent non-profit organization), Arturo Casadevall, and Elisabeth Bik have all studied misconduct, retractions, and bias. In the new preprint, they used a set of papers from PLOS ONE shown in earlier research to have included manipulated images to test what factors were linked to such misconduct. The results confirmed some earlier work, but also provided some evidence contradicting previous findings. We spoke to Fanelli by email.
Retraction Watch (RW): This paper builds on a previous study by three of your co-authors, on the rate of inappropriate image manipulation in the literature. Can you explain how it took advantage of those findings, and why that was an important data set?
Daniele Fanelli (DF): The data set in question is unique in offering a virtually unbiased proxy of the rate of scientific misconduct. Most data that we have about misconduct comes either from anonymous surveys or from retracted publications. Both of these sources have important limitations. Surveys are by definition reports of what people think or admit to have done, and usually come from a self-selected group of voluntary respondents. Retractions result from complex sociological processes and therefore their occurrence is determined by multiple uncontrollable factors, such as the policies of retracting journals, the policies of the country in which authors are working, the level of scrutiny that a journal or a field is subject to, the willingness of research institutions to cooperate in investigations, etc.