A modest (in the Steven Jay Gould sense) proposal for the undermining of scientific fraud.

I've recently been hearing a great deal about published and unconscious fraud in peer reviewed scientific publications. Fraud can range from not publishing negative results, to deliberate obfuscation in the interpretation of data, to falsification of data. I've heard it suggested that all studies should be pre-registered or results won't be published. Seems reasonable, but it would undermine the peer review process. The popular press would become the route to publication. This is sticky- cold fusion still has a significant following.

In the comments section above I see a suggestion to retract any paper for which data aren't forthcoming. Great. Let's move the fraud to data fabrication. Another commenter pointed out that with fabricated published data, further studies can perpetuate falsity.

My proposition is to use prediction markets (https://en.m.wikipedia.org/wiki/Prediction_market) to forecast the veracity of scientific papers.

Prediction markets could be created for papers appearing in important journals (Nature, Science, The Uri Geller Journal of Spoon Bending). Liquidity would act as a proxy for importance (and therefore become a substitute for Citation Analysis). Since the likely validity of a paper depends on both the quality of data and analysis, and also the Bayesian prior probability of the contention made by the study before the study was conducted, the betting odds (proxy for the public estimate of the truth of its assertions) will reflect both the likelihood of the base study and its novelty (more significant departures from received wisdom are inherently less likely to be valid, and thus more important).

I make my living out of gambling on prediction markets of another kind: horse racing. Winning is a highly mathematical endeavour: assertion of personal biases or faulty statistical processes leads indubitably to poverty. Robert Trivers has created a credible body of literature concerning the evolutionarily supported propensity to self-deceive. Cognitive biases produced in this way (and other ways) show up in prediction markets resulting in inefficiencies that can be exploited by those with no biases, those possessing undisclosed knowledge, or those possessing analytical techniques for correcting the market for biases. The market trends toward efficiency more efficiently than any other known statistical process. It bites the bum of those engaging in deliberate or unconscious deception. And it certainly isn't a good look if your study shows odds of 100 to 1 of being correct.