FIGURE 6.

The results of CCA stability assessments using the HCP data set in the “main procedure” with 290 subject measures (i.e., the “strong correlation” scenario; the left column) and with 246 subject measures (i.e., the “moderate correlation” scenario; the right column). Panel (a) shows the magnitudes of CCCs obtained from 2,000 CCAs for all combinations of subject overlapping rate and data dimensionality. Panel (b) shows the absolute differences in CCCs of 1,000 pairs of CCA. Panel (c) shows the consistency of the statistical significance of CCCs between two subgroups of 1,000 pairs of CCAs. Panels (d) and (e) show the correlation coefficients of the loading vectors between two subgroups of 1,000 pairs of CCAs corresponding to brain imaging measures and those corresponding to subject measures, respectively. The abscissa of all subgraphs represents the dimensionality of imaging measures (i.e., the number of kept PCs, ranging from 35 to 315 with a step of 35) and the corresponding SVR (i.e., the ratio of the sample size to the dimensionality of the imaging measures, ranging from 10.0 to 1.11). The subject overlapping rates between two subgroups of each pair (ranging from 0 to 315 with a step of 35) are color coded. In all bar plots, the height of the bars indicates the mean and the error bars indicate the SD. IM, brain imaging measures; SM, subject measures