3 ms·
Yes. I think few non-medical researchers realize how limited the value of a particular correlation generally proves to be. The more you work with biomedical d
by randcraw 10y ago
Yes. I think few non-medical researchers realize how limited the value of a particular correlation generally proves to be. The more you work with biomedical data, the more you appreciate the limited value of big R values when no plausible mechanism of action accompanies it. Given enough metrics, it's common to find one with perfect correspondence to a disease's biomarkers or symptoms. Lab biologists and chemists have long ago lost interest in exploring all the miraculous sure-fire yet inexplicable signals found by data scientists. The solution in this case is for the paper's authors to spend THEIR OWN grant money underwriting the clinical trials to explore their 'discovery'.
- nonbel 10y ago>"how limited the value of a particular correlation generally proves to be" I would go further. In my medical research work I saw dozens, probably hundreds of "statistically significant" correlations between different variables in a rather small dataset. I found them just by taking different subsets or summarizing the same data in a few different ways. I didn't find any interesting and so published nothing about them, but I am sure others could build entire careers out of coming up with narratives to explain those correlations. I have to think that people who have trouble finding correlations are either clueless about exploring the data (eg have only ever looked at dynamite plots of their results) or are measuring something so swamped by instrumental noise it is meaningless.
- pge 10y agoObligatory XKCD reference: https://www.xkcd.com/882/ https://www.xkcd.com/882/