Dickson et al. [4] consider up to nine rare causal variants within a 100 kb genomic region. We repeated their coalescent simulations (see Text S1), but using a quantitative trait for simplicity, although the same principles apply for disease outcomes. For a quantitative trait the simulated 100 kb haplotypes have quantitative values which are a function of the number of causal variants they carry, since we assume (like Dickson et al.) that all causal SNPs have the same effect size. We allowed for recombination at the standard rate of 1 cM/Mb across the whole 100 kb region and varied the number (k) of rare causal SNPs between 1 and 18. We investigated the frequency distribution of the most associated SNP allele. To make this representative of SNPs included in GWAS studies, we retained all SNPs with MAF>0.2 and a proportion of SNPs of lower MAF to generate an approximately uniform distribution of genotyped SNPs (Figure 2b–c). As the number of causal SNPs increased we found that the frequency of the most associated common variant changed, based on the most likely
