For both single- and multi-trait PGIs, we limited the initial set of PGIs released to those with an out-of-sample expected R2 above 1%. While the threshold itself is arbitrary, the decision to have a threshold was driven by two considerations: the value of a PGI for research is increasing in its predictive power, and we worried that a PGI with low predictive power could cause more harm than good if researchers are tempted to conduct underpowered studies.
