Whether genomic control can correctly capture the population substructures is still under debate (Devlin et al., 2004), with reports of either under- or over-correcting the correlation depending on the number of markers used (Marchini et al., 2004; Köhler and Bickeböller, 2006), and its performance perhaps also depends on whether the markers used to estimate λ are ancestral-informative or not. For whole genome association studies with a large number of markers, it is recommended to use a Bayesian version of the genomic control (Devlin et al., 2004). In our situation, we are correcting the known relatedness between samples, and there is no issue of under- and over-correcting the test statistic.
