OmicABEL estimation of standard error for effect size of SNP doesn't match GWFGLS estimation for some SNPs

[Sodbo]

An output of pipeline OmicABEL::reshuffle contains beta_SNP (which effect size of SNP) and se_SNP (which is standard error for beta_SNP).
I applied OmicABEL and GWFGLS for same data with same procedure.
Report is in attached file : FindingNemo.pdf
FindingNemo.pdf
I compared results and found, that standard error (se_SNP) of beta_SNP is differ between OmicABEL and GWFGLS outputs. Meanwhile beta_SNP estimations are same across OmicA and GWFGLS
Looking deeper I found that differences between standard errors in increasing with significance of beta_SNP (page #2 in attached report).
Crossplots for beta_SNP, se_SNP and minus log10 (pvalue) between 3 methods: OmicABEL, GWFGLS, EMMAX are attached. On each plot you can see 15 x 266K dots, which are either beta,se of -log10(pvalue) of SNP. In total 15 traits and 266K SNPs were analysed.

[lckarssen]

Thanks for the extensive bug report!