normality assumption for GWAS QT

Question

[kris]

Hi

Planning to perform GWAS on quantitative trait using linear reg in plink. From my basic knowledge of statistics, i know that the assumption of normality (residuals) should be only approximately true. But i have seen so many papers on GWAS transforming the data for normalization. My personal view is that " no real data set is perfectly normal". So do i need to transform my data with the following plot ? So far i have received mixed feedbacks from my group members (some say it's ok to go ahead with linear reg, while others want the data transformed)

TIA

Answer 1

[Lemire]

First, you can't judge if your distribution is normal by looking at a histogram. Always use a qq-plot against a normal to see if your tails are heavy or thin. The tails matter more than the shape.

Second, there has never been a time in history where simple regression was relying on your trait being normal. Inference is based on the central limit theorem and that is fairly robust as long as you have a decent sample size.

Third, people who are anal about transforming the data don't understand that they are sacrificing the interpretability of the effect sizes. They're doing research, yes, but they're just in it for the p-values...