Very high Beta coefficients in --glm output in plink2

Question

[putty]

Hi, I am not sure if this post is related to bioinformatics or stats, but posting here. i did GLM in plink2 with a quantitative phenotype (the range of measurement of the phenotype is from 300 microns - 900 microns). i got a number of significant hits, but i am confused as to why the Beta and 95% CI are so high ? Is it because the Beta calculated by plink2 is unstandardized Beta ? I am running association analysis using genotyping data (thoroughly QCed). The phenotype is thickness measurement (measured in microns).

Any one could please throw some light as to why the Beta values and corresponding 95 % CI are so high ?

Thank you all.

[LauferVA]

For those who may have come to this post for different reasons, keep in mind there are several fairly common reasons other than units that could produce this kind of issue other than the issue of units (which is discussed below).

One is collinearity between your predictors. This can be calculated simply by deriving a correlation coefficient for each covariate that is included in the GLM model. If you have significantly collinear variables, this can/will produce numerical instability in the Beta estimates produced by the model. This could manifest as an impractically large Beta coefficient, or as an impossible huge Wald statistic, etc.

Answer 1

[Lemire]

The betas are always on same scale as your trait, as such they represent the number of "micron" increase per A1 allele. If you want standardized betas then standardize the trait.

[putty]

Thank you, it makes sense