Using the same variance component framework, we developed a method to estimate the marker-specific inflation of test statistics using the correlation between each marker and the empirically estimated kinship matrix (described in the Supplementary Note). These estimates are concordant with the genome-wide genomic control inflation factor on average but show substantial differences across the SNPs (Fig. 5a). In the height phenotype, for example, the estimated marker-specific inflation factors have a mean of 1.107, s.d. of 0.090 and median value of 1.093. In light of this, we explored the relationship between marker-specific inflation factors and the overdispersion of test statistics with the uncorrected analysis. The distribution of height association P values for SNPs with inflation factors <1.05 shows a less marked departure from uniform distribution than does the distribution for SNPs with inflation factors >1.20 (Fig. 5b,c). Considering that SNPs with a higher inflation factor were identified without consideration of their possible association with the phenotype, it is reasonable to conclude that this excess of small P values reflects overdispersion of test statistics.
