In order to estimate hSNP2 and R2, we first took the residualized phenotypes described in section “Prediction Analyses” and additionally residualized these on 20 PCs in HRS and WLS, and 40 PCs and batch effects in UKB3. We did the same for the PGIs. In HRS and WLS, we estimated hSNP2 with genomic-relatedness-matrix restricted maximum likelihood (GREML) implemented in GCTA v1.93.0beta29,33 using HapMap3 SNPs with MAF > 1%. Prior to the hSNP2 estimation, we dropped one individual from each pair with a relatedness greater than 0.025. We estimated R2 as the explained variance in a simple regression of the residualized phenotype on the residualized PGI. Standard errors for R2, hSNP2, and ρ were estimated with a 100-block jackknife procedure.
