Skip to main content
. 2022 Mar 16;8(11):eabj1812. doi: 10.1126/sciadv.abj1812

Fig. 4. Predictive COD of AA and WA when no confounding variable was regressed during model building.

Fig. 4.

(A) HCP dataset. Each violin plot shows the various predictive COD across 40 data splits. (B) ABCD dataset. Each violin plot shows the various predictive COD across 120 training-test splits. Blue and green violins represent AA and WA, respectively. Red violins show the difference. Gray violins show the null distribution of difference generated by flipping the AA versus WA labels. Asterisk indicates that the difference in predictive COD between matched AA and WA is significant (FDR controlled at 5%). Gray dashed line indicates 0.