Saddlepoint approximation

When analysing binary traits, we recommend running single-SNP analysis using a saddlepoint approximation (SPA). Classical logistic regression may result in inflation of type 1 error when testing phenotypes with case:control imbalance. The SPA implemented in LDAK guards against false positives and can robustly test SNPs for association, even when the case:control ratio is highly imbalanced.

The SPA is included in LDAK by default when running --logistic, or when running --kvik-step2 on binary traits, for example:

./ldak7.01.linux --logistic out --bfile data --pheno phenofile --covar covfile

The SPA can be turned off by adding --spa-test NO, in which case LDAK uses classical logistic regression.