In PCHAT [4] (http://www.wpic.pitt.edu/wpiccompgen/PCHAT/PCHAT.htm) the sample is split in a training set, which is used to construct the optimal linear combination of traits from a heritability point of view, and a test set, which is used for association testing between genotype and the optimal linear combination of traits. In this way, use of the same data for both estimation of the optimal linear combination of traits and association testing is avoided. In addition, so called ‘bagging’ is performed, in which bootstrap samples are drawn from the training sample and the optimal linear combination of traits is averaged across bootstrap samples. The null distribution of the test statistic is obtained in the same way, using permutation of the data. We applied the additive model and set input parameters to values recommended by the authors: 50 subsets and bagging subsets for the determination of the distribution of the PCHAT test statistic under the null hypothesis; 200 and 50 subsets and bagging subsets, respectively, for testing the association of a genetic variant with the trait; 150 individuals for the subsets; and 1000 simulations
