For the group level analyses, each individual participant's correlation coefficients were averaged to obtain a group correlation for each ROI pair, and then the group correlation matrix was thresholded as outlined above. For our behavioral analysis, we derived each individual participants network by thresholding that participant's correlation matrix again using bootstrap resampling, which resulted in 16 individual networks. The density of each participant's correct retrieval network was then computed and subsequently regressed against their behavioral accuracy for all task trials. Next, using stepwise regression, a final multiple linear model was produced to account for the variance in the density data. This final model was based on an initial model consisting of a subset of nodes whose percent contribution of the overall network density individually showed a significant (p < 0.05) linear relationship with performance across participants. The algorithm then stepped through all 26 regressors (or nodes) to construct a final model that accounted for the most variance (diagramed in Fig. 3).
