Why do scientists commit fraud? If you are active in the scientific community, you are aware of many cases of research misconduct. It happens with significant, internationally recognised research [dias-paper], and it can happen at your university, too  . This is a fascinating question partly because of the sensational nature of the drama but also because it can make us think about the social way in which we practise science. In this blog post, I'd like to talk about a neat paper I found by Liam Kofi Bright [on-fraud], which provides a formal model for why scientists might commit fraud.
Take some of these high-profile cases: one that made a lot of noise around my research area was the case of Ranga Dias, who claimed to have discovered a room-temperature semiconductor earlier this year. This had been published in Nature and has since been retracted [dias-paper]. Because room temperature semiconductors would be a revolutionary material with wide-ranging applications from power transmission to digital electronics to medical imaging, it's almost certain that the researcher would be caught when other researchers. Given that the stakes and the probability of discovery are so high, why do it in the first place?
You might say that the expectation is that they won't get caught, but this is also a puzzling belief suggests either:
- they don't see it don't see that outcome (cognitive blindness),
- they think that their work won't be scrutinised (deliberate deception),
- or that the underlying phenomenon is true, and their data gives the wrong result (noble lie).
The first explanation is a psychological one, which might be true. Still, it doesn't really explain the why of committing fraud. Much consideration is given to the first deliberate deception scenario: scientific fraudsters are bad apples or con artists who lack morals. However, the truth is likely much messier.
There are many models of human behaviour, but one of the most prominent is the "rational choice theory". Rational choice models, as formalised by game-theoretic games, are rarely fun to play but can explain why people make the choices they do. Bright's model works within this framework; it is a game-theoretic account of the decision-making process by the scientific community. You might be familiar with well-known game-theoretic models such as the prisoner's dilemma or matching pennies. These are examples of zero-sum, two-player games with perfect information, which are also well-known and well-studied. Bright's model, by contrast, is not zero-sum, with incomplete information and many players.
I'll briefly describe the premise of the model. The model assumes that the scientific community consists of scientists who are interested in theoretical statements that can be true or false (e.g. can a particular method of synthesis yield room temperature stable semiconductors). For simplicity, assume there are two possibilities, one of which Nature picks. Scientists privately conduct unreliable experiments that reveal evidence about the theory. Then, in light of the evidence, the scientists update their beliefs and publish their results; that is, they signal their beliefs about the theory.
Periodically, the scientists gather at a conference to resolve the theoretical questions bugging them. The position that most scientists agree to become the received wisdom in the field and the scientists who have published in favour of that position are rewarded. Note that the scientific community may decide on a theory that does not agree with the state of affairs picked by Nature (e.g. phlogiston). So far, this is similar to other models of the reputation economy of science (e.g. Kitcher 1993).
Bright introduces a wrinkle to assign scientists different motives: they may be pure credit seekers, pure truth seekers, or mixed credit/truth seekers. Truth seekers only want the scientific community and Nature to be in agreement, pure credit seekers only want to be seen to be on the right side of the community, and mixed credit/truth seekers would like for the scientific community to arrive at the truth but also to be on the right side of the scientific community. Moreover, the scientists also have a set of beliefs about what other scientists will publish. Scientists are motivated to fraud when they are motivated to publish a paper that supports an outcome that is the opposite of what was revealed by their experiments.
Bright then looks at how changing the motive types would increase or decrease the incentives to commit fraud. Bright also looks at how behaviour assumptions about the cost of fraud and self-confidence would also affect incentives to commit fraud. Putting all this together, he derives several interesting (if pessimistic) results.
The first result is that turning a credit-seeker into a mixed credit/truth seeker only dissuades people from committing fraud in a setting where they believe their opinion is irrelevant. Further, turning credit seekers into mixed credit/truth seekers can incentivise people to commit fraud in new situations. Pure credit seekers will always be incentivised to commit fraud when they believe their experiments are 100% accurate. Finally, even pure truth seekers can be motivated to commit fraud when they believe their results are incorrect.
On the positive side, however, he can show that pure truth seekers are incentivised to lie in a strict subset of situations of the mixed truth/credit seekers. He also shows if there is a high price for committing fraud, then pure credit seekers will honestly report their experiments when certain conditions are met: they think their beliefs will affect the community consensus, they believe in the true state of Nature, and they think the community will also eventually arrive at the correct theory. Bright has further interesting things to say about the scientific community in that paper. I encourage people to read it because he's a very clear writer and sometimes quite funny ( a rare trait in scholarly writing!).
This model is attractive because formal analysis suggests counter-intuitive points. A standard theory is that the scientific credit system and the contemporary career pressures, are to blame . While this is undoubtedly true for some cases, the motivation question is worth scrutinising.
Bright shows that this is not necessarily the case; even people with ideal motives can be incentivised to commit fraud because of "noble lies". To me, this suggests that some problems about fraud are intrinsic to the scientific community because of the community nature of the decision-making. It's worth considering that truth-seeking is not always noble, nor is credit-seeking always villainous.
When this is all formalised as a game-theoretic model and in this context, the relation to the celebrated results on voting theory like Arrows' theorem becomes more apparent. The scientific consensus is a process of plurality voting. In plural voting (like all other forms of voting), it is vulnerable to strategic voting (the Gibbard–Satterthwaite theorem); a voter receives a better payoff if their vote is not a true representative of their true beliefs. Scientists would be better off lying about their beliefs if that would cause the community to arrive at what they honestly think. I think it is also quite interesting to use these game-theoretic/economic models to understand the process of science. Moreover, these hypotheses are testable in a reasonably precise way. From observational evidence, it would be interesting to see how often this happens in practice.
I'm also left wondering how often "noble lie"-type fraud happens. In particular, I keep thinking about two well-known and controversial experiments: the Stanford prison experiment and the Millikan oil drop experiment. The Millikan oil drop experiment, a canonical experiment to determine the charge of an electron, has been criticised for removing data points from his dataset: he published results based on only 58 out of 175 recorded points. The result is extraordinarily accurate, with 0.5% per cent of the accepted value nearly 100 years later. Yet the case is not clear cut because of the complexity of the experimental setup [goodstein-in-defence]: one person's exploratory data analysis and posthoc analysis can be another person's data dredging and p-hacking.
I hope you'll agree there is much excellent work being done in the history and philosophy of science which is relevant to practising scientists. I think scientific misconduct is interesting for the same reason that a sociologist might be interested in criminal behaviour or an evolutionary biologist might be interested in the Galapagos islands because they represent an extreme. This extreme informs about the practice of science.
[on-fraud] Bright, L. K. (2017). On fraud. Philosophical Studies, 174, 291-310.
[goodstein-in-defense] "In Defense of Robert Andrews Millikan" by David Goodstein (American Scientist, January-February 2001).