We applied the same subsampling method to identify large-effect splicing quantitative trait loci (sQTLs) and ASE. To compare ASE between the family and population, we focused on a subset of genes that had substantial data for the measurement and comparison of allelic ratios (n = 1,777 genes). For a gene to be included, allelic ratios at a single site had to be measurable for at least five siblings and at least 30 population samples. We tested each gene once and excluded genes that were not tested for eQTLs, such as pseudogenes or genes within high-complexity regions (human leukocyte antigen and immunoglobulin loci). For a site to be considered measurable, it needed to be covered by a minimum of 20 reads with at least five reads for each allele. We then took the maximum allelic ratio in the family and compared it with the maximum allelic ratio found in 1,000 subsamples of the Geuvadis; each subsample was matched to the number of heterozygous individuals found in the family for that site. This approach generated an empirical p value that we used to assess whether an ASE effect in the family was greater than that in the population. To account for ASE biases caused by differing read depths between the family and population, we downsampled (hypergeometric) Geuvadis reads by a factor of 1.97—we calculated this scaling factor by measuring the average level of read-depth differences between Geuvadis and family samples at those selected heterozygous sites for each gene. To exclude the possibility that large-effect ASE was due to technical artifacts such as mapping biases or sequencing errors, we also looked at ASE for the second-largest-effect siblings and IBD siblings (Figure S25).