Conditionally palinear returns results for

[mightyphil2000]

No description provided.

[mightyphil2000]

Using palinear, a colleague of mine ran a GWAS on a quantitative trait conditioning on the top hit from the marginal results. For practical reasons, they ran the conditional model for all SNPs, including the top hit. Therefore the tested SNP will be perfectly collinear with the covariate/adjusted SNP in one of the models. In the latter model, where tested and adjusted SNP are the same, palinear returns a result for the SNP. Many statistical software packages would typically drop one of the collinear SNPs, returning a result for the SNP that is identical to its result in the marginal results (i.e. unadjusted for any SNPs). However, this does not happen with palinear, which returns a result for the SNP in question that is completely different from its marginal result. i.e. the slope (b1) in model1 (Y = b1SNP1) is completely different from b1 in model2 ( Y=b1SNP1 + b1*SNP1).

Could you comment on how palinear handles perfectly colinear variables in the same model?