Cheating in Science

[jimmydasaint]

Reading mooeypoo's article about the confirmation bias involved in locating an Australopithecine skeleton, the suggested 'Missing Link' of human ancestry (mooeypoo ) inspired me to think of examples where scientists had cheated to get ahead. Certainly, I had witnesses one or two 'dodgy' goings on as a research scientist over twenty years ago (I quit research 20 years ago) and I wondered how much of an issue it had become.

 

I found two reasonably relevant articles:

 

"With apologies to Charles Dickens, in the world of biomedical publications, "It is the best of times, it is the worst of times". Scientific productivity, as measured by scholarly publication rates, is at an all-time high1. However, high-profile cases of scientific misconduct remind us that not all those publications are to be trusted — but how many and which papers? Given the pressure to publish, it is important to be aware of the ways in which community standards can be subverted. Our concern here is with the three major sins of modern publishing: duplication, co-submission and plagiarism. It is our belief that without knowing whether these sins are becoming more widespread, the scientific community cannot hope to effectively deter or catch future unethical behaviour."

 

Nature Article

 

 

In the competitive world of academia, a person's worth is often ostensibly gauged by one's scientific contribution, wherein the 'article count' has become the simplistic measure of this contribution. The number and frequency of publications reflect an academic's stature in the scientific community and hence the race to publish and increase this 'article count' has become an end unto itself. Sadly though, the overriding desire to publish sometimes defeats the very purpose of scientific contribution as, not unsurprisingly, even the learned may cheat.

Saudi Journal

 

But I have to admit that everything is not black and white here, and that my idealism as a young man to do the ethically correct thing has now given way to feeling sympathy for those that feel that they can only gain grant funding or achieve a permanent contract through adding to their publication count. Of course, tweaking the data is unethical and deserves disapprobation.

 

How do others feel?

Edited March 22, 2011 by jimmydasaint

[imatfaal]

JImmy - I think you make very good points, and maybe this is an issue that might be brought more into the popular view with the very public stripping of Guttenberg's PhD in Germany and other investigations. Personally, I think a greater danger to academic integrity is the non-publishing of adverse data in the medical/pharmacological arena. I believe there are moves to address this problem but it cannot be right to carry out multiple drug trials and only allow release of data from those that suit the drug company. Back on topic - is this form of data manipulation not one of the very reasons that we have peer review? I think a very strong distinction also needs to be drawn between the dedicated researcher that over-reaches himself/herself and massages data to fit a strongly held conviction; and the more pernicious cynical manipulation of the knowing fraud who lies in order to further a career.

[swansont]

From a recent analysis - less than 200 cases out of more than six million papers, or a 0.0032% rate of fraud.

 

http://blogs.scienceforums.net/swansont/archives/7117

 

I disagree with moo about the extent of confirmation bias in the finding of Lucy; if looking for data where theory tells you where it should be is bias, one has to argue that building the LHC (or any accelerator) is a massive act of confirmation bias. Why are we looking for the Higgs at 7 TeV (and eventually up to 14 TeV)? Why don't we look for electron/positron pair production when the available energy is less than 1.02 MeV?

 

But more to the point, mistakes in methodology or execution of an experiment are not the same as fraud (falsification).

[JohnB]

Not having done research I don't know, but from the laymans perspective.....

 

There might sometimes be a fine line between "testing a hypothesis" and "confirmation bias". At one end of the spectrum the researcher is looking through a limited and known dataset for evidence to either confirm or falsify a very specific prediction. Swansonts example of the LHC is a good case for this end of the spectrum as you either find the Higgs, or you don't. There is little choice in which dataset you use and the results are quite specific and replicable. At the other end we have something like Lucy. The dataset is large and completely unknown as to size and content. It's the researchers opinion as to what data from that set is "relevent" to the research. The wider the freedom to pick and choose the data, the more likely that confirmation bias will be a factor. This is simple human nature, we pay more attention to those things that confirm our previous beliefs than we do to those that falsify them. I simply assume that researchers are human too.

 

This would mean that somewhere on that spectrum "confirming the hypothesis" must lie beside "confirmation bias", exactly where I don't know, but that is where the line is thin. As we move from zero choice in data to total choice in data the probability of confirmation bias increases from zero to "very likely". I must echo swansont one this point though, a case of confirmation bias is an example of poor methodology and nothing more, it is not fraud or anything untoward, it's a rectifiable mistake.

 

In many ways this is exactly the problem I have with dendroclimatology. We have two theories, one is that tree ring size is an indicator of temperature and the second is that tree ring density is also an indicator temperature. So we take 100 tree cores and find that for 12 of them the ring size tracks temps over the calibration period so we can use them in the paleo reconstruction. We then run them again for density and find a further 10 that thrack the temps so we can use them too. We now have 22 series that we can use for the reconstruction of paleo temperature, and they track the temps in the calibration period, so they must be reasonably accurate and a good indicator. What about the 78 that don't track the temps? Wouldn't their existence imply that there was something wrong with initial theories? There are 3079 datasets listed in the NOAA ITRDB database, yet when we look at paleo reconstructions the same 15 or 20 keep getting used. What's wrong with the other 3,000 datasets? (Granted that many won't go back as far as we would like, but I hope people see my point here.)