Under the null hypothesis of no marker-phenotype associations, the intensity distribution at each marker is independent of the family status (i.e. father, mother, child) in the family-based samples and independent of the affection status (i.e. case, control) in the case–control sample. Thus, in order to simulate the test statistics distributions under the null hypothesis, we permuted family status (and family labels) or affection status, respectively, several times; each time storing the corresponding test statistics. Following this procedure, marker-wise approximate exact P-values were then calculated by dividing the number of observed permutations achieving an equal or higher test statistic for this marker than the actually observed one by the number of permutations performed. The QQ plots in Supplementary Material, Figure S3 demonstrate the validity of this approach and of the asymptotical P-values by comparing the quantiles of the 8051 true (asymptotic) observed P-values in the 244 common CNVRs with those quantiles of approximate exact P-values based on 1000 permutations in the case–control sample and the family-based sample, respectively. Note that here, deviations from the diagonal line for the case–control sample result
