3 ms·
It seems like the solution to this is fairly simple. Use some statistical or machine learning method to figure out the probability that a certain thing is true
by Houshalter 13y ago
It seems like the solution to this is fairly simple. Use some statistical or machine learning method to figure out the probability that a certain thing is true using the information we know about it, like what journal it was published in, the results of replications, maybe even stuff like how crazy the result seems or the experience/reputation of the scientists, etc. There is a ton of data to work with, on top of the actual data itself.
You could predict with decent accuracy how probable a study is to turn out to be true or false. Then you can use that information to decide whether it would be worthwhile to do more studies.
- eli_gottlieb 13y agoIf you can work out what learning problem you've got here, and what methods you can apply towards a solution, you should go and pitch that as a start-up.
- Houshalter 13y agoI'm not sure there would be anyway to monetize it but I considered trying it as a personal project. It would be way to much work to manually enter the data of thousands of papers into the computer though. I would also need objective data on which studies actually turned out to be true or false. Or at least which ones could be successfully replicated.