Thursday, October 04, 2012

Why are research papers retracted?

According to a study out of Yeshiva University, it's usually because of fraud.  Now to be fair, the study is not linked, and the article does not suggest what percentage of papers are retracted overall, but they do suggest that the number, as a percentage of papers published, has grown by an order of magnitude since 1975.  The study is also published by the NAS, which I would assume would usually be respectably peer reviewed.

Love to hear from Brian on this if he reads this site.  My take is that if the methodology of the study can be defended, it is intuitive that we'd see a fair amount of fraud.  After all, nobody gets tenure because they do work that defends the null hypothesis.  There is a very real pressure to "show something new."

I have to wonder if it has a lot to do with requiring a Ph.D. to teach in most colleges (even freshman level courses that a high school teacher could lead), and a general "publish or perish" mood.  If the study holds, I would not be surprised. 

Another possibility is the strong push for full acceptance of evolution in bioscience; I do not have numbers, but I would have to believe the vast majority accept the hypothesis without reservation, and that a large portion of these have followed that with Richard Dawkins to atheism, or at least agnosticism.

At which point John Paul Sartre's comment about the nature of French existentialism comes to mind:

If there is no God, everything is permitted.

(in reference to Ivan Karamazov of the Brothers Karamazov)

And if you think that I'm enjoying quoting Sartre, as cited by "infidels.org", about the importance of Exodus 20:16 as it related to the scientific process, you would be entirely correct.  It is absolutely delicious irony!

2 comments:

Brian said...

I have the PNAS paper in front of me right now. A few quick points:

1. The methodology is fine; they did a comprehensive search on Pubmed. Anyone with enough time on their hands and internet connection could do it.

2. The growth of retracted papers by an order of magnitude since 1975 is from ~0.001% to ~0.009%, so while that is an order of magnitude, we're talking about a very small number becoming a slightly less small number. I think both the authors of the study and the article covering it are overstating the significance of the increase. Which brings me to...

3. The data around the trendline from which those numbers are extrapolated is pretty noisy...and actually, the trend is essentially flat from the 70's until the late 90's. Then it ticks up a bit. Astute readers will note that this is right around the time journals and indexes moved to the internet on a broad scale, and I doubt that this is a coincidence.

So...my take is that a very small level of fraud is with us, but I don't think that it is a great deal more than there was at some arbitrary point in the past. Considering how quickly the enterprise of science has grown, that is actually pretty remarkable, given the incentives to fraud which the authors discuss.

None of which is to say that fraud isn't a problem.

However, I would argue that the fact that studies are retracted is evidence that the system is working, not that it is broken. Detecting fraud at the level of peer review is damn near impossible--most people outside of science don't appreciate the fact that ALL peer review is done by us, for no compensation whatsoever, on our own time. If I'm lucky, I might be able to spend a few hours in total with a paper to read it in detail, look up any background info I'm lacking, and write a decent critique. The only way I'm likely to detect outright fraud at this stage is if either 1) they are really bad at it, as in they clearly photoshopped an image or something, or 2) I happen to be privy to directly contradictory data, because I have done the exact same experiment and found a different result. And even in the latter case, that doesn't necessarily mean fraud.

More likely, I might encounter a result that just seems too good to be true. At that point, I can raise pointed questions about methodology: Was the study blinded? How were replicates managed? Etc. I may even express some concern confidentially to the editor, if I really think I smell something fishy. But an outright accusation of fraud is not something you do (here, or anywhere) without some hard, concrete evidence. So at some point, we either decide to take the authors at their word, or not.

Frauds are caught when the study is published, and someone else tries to replicate key findings independently, and cannot. That doesn't happen unless the study is published in the first place. (And even then, it might not be fraud.)

To your other points: virtually no one in science gets a PhD because they want to teach. We teach because we have to. And your equation of atheism with amorality is as baseless as it is insulting, Sartre quote notwithstanding.

Bike Bubba said...

Glad it's not worse, Brian.

And regarding "they teach because they have to"; what an inversion of the old system of apprenticeship, no, where masters saw passing on the craft as part of the craft?

Regarding Sartre's quote; I don't view it as baseless or insulting at all, but rather the natural outgrowth of what happens when a man doesn't fear punishment after death. For a religious counterpart, what happened with the Pope told Crusaders that as they strove to liberate the Holy Land, all their sins would be forgiven?

The stocked up on sins to be forgiven, of course, and Sartre notes the same thing for the French existentialist.