For validating the predictive power of the PGIs, we used five cohorts for which we had access to individual-level genetic and phenotypic data: the Health and Retirement Study, a representative sample of Americans over the age of 50; the Wisconsin Longitudinal Study, a sample of individuals who graduated from high school in Wisconsin in 1957; the Dunedin Multidisciplinary Health and Development Study, a sample of residents of Dunedin, New Zealand, born in 1972–1973; the Environmental Risk (E-Risk) Longitudinal Twin Study, a birth cohort of twins born in England and Wales in 1994–1995; and the UKB (our third partition). The top panel of Figure 3 shows the observed R2 and 95% confidence intervals for the single-trait PGIs in one or more validation cohorts, depending on which had a measure of the phenotype. Height, BMI, and educational attainment are shown separately because the y-axis scale is different. The bottom panel of Figure 3 shows the difference between the R2 of the single-trait Repository PGI and that of a PGI we constructed using the largest non-overlapping GWAS whose summary statistics are in the
