To deal with the false positives, we derived a distribution of correlation coefficient values by correlating randomly shuffled beta series 1,000 times and pooling across participants and all pairwise ROIs within a condition. All correlation values were transformed using Fisher's Z, or the arc hyperbolic tangent transform52. A functional connection, or edge, in the matrix was considered significant if the observed correlation value (correlation coefficient averaged across participants) was greater than the 90th percentile of the calculated distribution. This approach aimed to account for multiple comparisons and to fix the Type 1 error rate at 10%. This more liberal threshold preserves a higher network density thus providing enough connections to compute the graph theoretical metrics. Additional analyses using a bootstrap alpha level of 0.05 produced similar findings (Fig. S4) and suggest overall robust results. Additionally, the distribution of correlation values across all subjects and all pairwise regions was negatively skewed, so the average correlation value for all pairwise regions was greater than zero. Thus, we observed few negative correlations between regions, suggesting minimal “anti-correlated networks” within our findings. It is
