Effect size for SKAT-O

[Jacqueline Otto]

Hello,

I am running SKAT-O within EMMAX in EPACTS 3.2.6. I know that the test uses rho coefficients in the computations, but is there any way to estimate the effect size based on the output it gives me? I see the Q statistic, p-value, a range of columns for variant counts, but nothing that would allow me to calculate effect size for each set tested. Has anyone been able to find a way to do this based on SKAT-O output? 

Thanks!

Jacqueline

[Mandar Karhade]

HI Jacqueline,

Disclaimer: I am working in genetics and I am no expert. 

But looking at the construction of any of the rare variant test statistics (or any test statistic over a region or group of SNPs that has been proposed before) , these are constructed using the distribution of the variants over multiple SNPs. The test is score based (most of the times) , which will not have a "effect size". We can consider some entity as the test statistic before generating P-values associated with it as a proxy for effect size but that is not a real effect size either. 

I would consider going through how these gene based tests are generated. That will clarify that these tests are not supposed to have estimates of effect size such as beta, OR, RR or anything like that. I appreciate the question because of which I am going to read SKAT paper once again myself.  :)

I hope I am helping. To all others, if you find something wrong in my logic or even the way I am expressing, please feel free to explain. 

Thank you

-MAndar