To compare cis-eQTL effect sizes, β, between the population and family, we sought to first correct for the overestimation of effect sizes (such discoveries exhibit characteristic regression to the mean). To address this in the population eQTLs, we divided the European-descended Geuvadis samples (n = 373) in half and partitioned them into discovery (n = 180) and replication (n = 193) panels. Within the discovery panel, we identified the strongest cis-associated variant per gene (by p value and within the same interval tested in the family). This allowed us to use the replication panel to more accurately measure the effect size of each cis-eQTL variant. However, to account for the difference in sample sizes between the replication panel (n = 193) and the family (n = 11), we further sought to estimate how much variance in effect-size measurements (β) could be obtained from sampling 11 people in the population at random. In this way we controlled for chance observations of larger effect sizes for some genes in the family. To achieve this, we repeatedly subsampled (100 times) 11 individuals from the replication panel while maintaining the exact same genotypes of the best associated variant between the subsample and the family. Figure S8 illustrates this subsample scheme. Effect sizes were then measured with the same regression formula, Ti ∼ μ + βjp + βkm, for both the family and the subsample; note that two regressors, p and m, match segregating patterns of both the haplotypes of the family and the best SNP of the population subsample. We note that estimation of β in the population was highly correlated independently of the use of a one- or two-regressor model (Figure S9). This allowed us to create a distribution of measured effect sizes that would be expected from randomly measuring the same number of individuals and genotypes in both the family and the population. Using this approach, we identified empirical p values representing how often measured effect sizes in the family were greater than that of the best associated SNP in the population. We also repeated this analysis by using fit (R2) given that we observed differences in the distribution of raw β values between the family and population and also observed higher variance in gene expression in Geuvadis overall (see Figure S17).