Balancing of participant demographics and genetic ancestry across hyperalignment training. The three sites differed in mean demographic attributes, including a four-level race model, parental education level, parental income, and Hispanic identity (A) and in mean genetic ancestry (B). To respect this diversity and sample the entire range of demographics in the normative hyperalignment model, we selected subjects to train the CHA model to reflect sample demographics along these four dimensions. Shown here are the distributions of participants’ scores, scaled for visualization, for the participants allocated for CHA train or test cohorts. Violins show training cohort on the left and testing cohort on the right, with quartiles of each distribution shown with a dashed line. Significance of differences along each dimension between CHA cohorts were tested with a two-tailed independent sample t test and * denotes p < 0.05 after Bonferroni’s correction.