As suggested in ref. 9, we explored the effect of including a variable number of principal components in the association tests. Although including two or five principal components are included has a considerable effect on the λ values, further augmenting the number of principal components does not substantially decrease the genomic control parameter (Fig. 2). It is often suggested that only principal components having predictive power for the phenotype should be included in the regression11. We identified principal components for each phenotype that have a t-test P < 0.005 as predictors; the results of their inclusion in the association tests are reported in Figure 2.
