We conducted a validation exercise for our PGIs in the HRS (N = 10,978; 57% female; median age = 82), WLS (N = 8,937; 52% female; median age = 82), Dunedin (N = 887; 49% female; 1972 birth cohort), E-Risk (N = 1,968; 51% female; 1994 birth cohort), and UKB (third partition; N = 148,662; 54% female; median age = 71) cohorts. The HRS sample used in our validation exercise (2006–2010) is smaller than the HRS sample for which we are releasing PGIs (2006–2012) because we only had access to phenotype data in the former. Supplementary Table 12 describes the phenotypes used as outcomes in these analyses for all cohorts except UKB. The UKB phenotypes are described in Supplementary Table 5. (The UKB phenotypes used in the prediction exercise differ slightly from the GWAS phenotypes described in Supplementary Table 5 in that they were not residualized on the PCs and genotyping batch dummies. Instead, we have controlled for these covariates in the regressions when calculating incremental R2 as described below.) As a general rule, if a single measurement in time
