Finally, we turn to the most difficult question: what causes differences in polygenic scores, as currently calculated, among different populations? Differences could be real or artifactual (i.e., due to bias in data and/or methods), and five categories of explanations are listed below.True differences due to driftTrue differences due to selectionTrue differences in genetic effects due to environmental differences (gene-environment interactions)Bias due to uncorrected population stratification in discovery and/or training samplesBias due to discovery/training population data and/or polygenic scoring methods. Specifically, linkage disequilibrium (LD) structure and variant frequency are captured imperfectly with current methods (including genotyping and imputation), and they vary across populations, and currently available data resources are unequally representative of diverse worldwide populations.Random error in the estimation of GWAS betas
