Should a scientific paper be retracted because it is mistaken?
We’re not talking here about misconduct, or deliberate fraud. We’re talking about a result that is, for whatever reason, wrong: a false positive or a miss, or an overly enthusiastic interpretation, or a good old honest mistake.
At the Retraction Watch blog, Tom DeCoursey argues that papers that are wrong should be retracted from the scientific record. His main argument is that people waste a lot of time trying to reproduce results that later papers have been unable to confirm.
This may be a rather different view of retraction than has typically existed. My impression is that previously, retraction occurred primarily when there was scientific misconduct: fabricated data, or an editor doing an end run around the peer review process. In the medical literature, retracting could also occur in case of an error that might kill people from mistreatment (“lethal error”; Horton 1995).
I get the impression that papers are getting retraction for a much wider range of reasons than ever before, for reasons that have nothing to do with the quality of the science (e.g., authorship squabbles; embargo violations; pressure from bloggers).
I went looking for whether anyone has conducted research on the reasons for retraction. A quick search turned up work by Snodgrass and Pfeifer (1982). Of the papers they looked at, 94% of the retractions they studied had a reason given, but they don’t break down those reasons into any categories.
Six percent of retractions were not explained at all. I’ve seen papers much more recently with no reason given, so the practice hasn’t stopped.
If we widen the use of retractions, all sort of questions are raised. What level of evidence, of failure to replicate, should be enough to warrant a retraction? One of my own recent papers was an extended attempt to replicate another experiment, without success. But it would be presumptuous of me to demand the original paper be retracted.
In the case of honest mistakes, how should a retracted paper factor into promotion, tenure, or funding decisions? And how should journals seek to notify readers? Remove the paper from the record, which is how retraction was supposed to work? Label it as retracted?
Another take on this from Biochem Belle.
Additional: An abstract describes a study (which I don’t think has been published yet) that ranks reasons for retraction: Misconduct makes up 15% (data fabrication, 5%; data falsification 4%; plagiarism, 16%); “mistakes” make up a larger component of about 29% (honest research errors; 28%, non-replicable findings, 11%). Then you have issues that are harder to categorize: redundant publication, 17%; disputed authorship/data ownership, 5%; inaccurate/ misleading reporting, 4% (not sure what that could be); and the kicker, no stated reason at 9%.
Horton R. 1995. Revising the research record The Lancet 346(8990): 1610-1611. DOI: 10.1016/S0140-6736(95)91935-X
Snodgrass GL, Pfeifer MP. 1982. The characteristics of medical retraction notices Bull Med Libr Assoc 80(4): 328-334.