Chunk #42 — Statistical Methodology for the Analysis of Association Trajectories — Identification of significant SNPs — Significant SNPs identified by concentration of effect sizes
Absolute values of effect sizes follow a folded normal distribution with variance 0.0025 corresponding to a normal distribution with variance 0.0069. However mean effect sizes averaged across correlated phenotypes and 3 year age ranges are neither uniformly nor normally distributed across SNPs as shown by the permutation test results. The distribution of the permutation test results, expressed in terms of z-scores, are not symmetrical but have a peak at about −2 standard deviations and a long rightward tail beginning about +2 standard deviations above the mean as shown in Figure 8. The fraction of effect sizes less than .1 among the SNP-age-phenotype combinations selected by having permutation scores greater than the inverse normal cdf determined by p-values in the range .05 to 10−4 was linearly related to the logarithm of the p-value, decreasing from about 30% for p = .05 to 15% for p = 10−4. For p < 10−4, the rate of decrease diminished from the rate over the range of .05 to 10−4. Further observation of the distribution suggests that the SNP-age combinations fall into three groups, non-significant, null, and significant.
