Reply to “Likelihood of False-Positive Results in High-Impact Journals Publishing Groundbreaking Research”

REPLY

We thank Drs. Agrawal and Sharma for their thoughtful letter, which suggests that papers in journals with high-impact factors are more likely to be retracted because they have a higher prior probability of being false as a consequence of their novelty. This is a good point, and we concur with the authors' reasoning. However, we would also emphasize that this factor alone cannot account for the strong positive correlation between impact factor and retraction rate. Bayes' theorem applies only to manuscripts retracted due to error rather than to misconduct. We have reviewed the specific reasons for retraction of manuscripts by seven high-impact journals (Cell, Journal of Clinical Investigation, Journal of Experimental Medicine, Nature, New England Journal of Medicine, Proceedings of the National Academy of Sciences, and Science) from 2000 to 2010. Mishandling of data (e.g., fabrication, falsification) was cited in 40% of cases, with possible misconduct suggested in an additional 12%. Over the same time interval, Steen found that only 28% of all 742 retracted papers indexed by PubMed were due to data fabrication or falsification (3). The higher rate of manuscript retraction in high-impact journals as a result of dishonest behavior suggests that potent incentives for cheating are likely to be involved (1, 2).