I am working on a GWAS project, and I wanted to estimate the gain in power that I get by using a quantitative variable instead of a categorical.
I have already run the GWAS analysis for my quantitative variable. If I where to "transform" it into a categorical one, say by binning it, and run the GWAS, is there a way to estimate the gain in power? Can I do that using my data? Or would it be better to try to run some sort of simulation?
I think a way to visualize the difference would be doing a QQ-plot? Also I could compare the P-values distribution, but since there are not normal what would be the best way?
Thank you.
Not a GWAS expert, but it's generally best to run a simulation. You'd then compare some sort of "distance from correct value" metric. I'd expect the quantitative run to produce better results, but that's just a guess.
So a distance between the two P-values? A P-value difference? Also of the beta coefficient?
Sure? I'd probably use the difference of p-values, which might have the problem of being very skewed, but it's a start.