Specifying null model w/out covariates?
pmagwene opened this issue · comments
I'm interested in using OrdinalGWAS to simply fit the ordinal regression on SNPs withoutout additional covariates. What's the right way to specify the null model in this case when calling the ordinalgwas function?
I tried creating a dummy "sex" covariate (with all samples set to have the same value), but unsurprisingly this raises an error if the covariate is non-variable.
@pmagwene Have you tried ordinalgwas(@formula(trait ~ sex), datadir * "covariate.txt", datadir * "hapmap3"), where the @formula(trait ~ sex) specifies an intercept-only null model?
For my usecase, sex is not an appropriate covariate.
Thinking about this a bit more, perhaps I could fit the model repeatedly, substituting a normally distributed random variable as a covariate. Sort of a permutation based approach with a non-informative covariate...
@pmagwene I hope I'm not too late to weigh in. Actually, you can do ordinalgwas(@formula(trait ~ 0), datadir * "covariate.txt", datadir * "hapmap3")
Updated the docs. Marking as resolved.