We subsequently considered how each individual node contributed to this significant linear relationship using a multivariate stepwise regression approach. Based on hubs that first showed a significant simple regression with performance, a stepwise regression using these brain areas (indicated with asterisks in Fig. 3) resulted in a final more parsimonious model with MFG, precuneus, hippocampus, and IFOr (shown in red) contributing significantly to the model (β = 4.05, 6.83, 6.76, 5.08, −8.83, p = 0.0001, R2 = 0.90). These findings argue that while overall network connectivity was a significant predictor of memory performance, specific hubs drove this effect most strongly. Consistent with our earlier analysis, this included connectivity to the MFG, hippocampus, precuneus (significant for uncorrected values), and IFOr as significant predictors of performance. Coefficients for MFG, hippocampus, and precuneus were positive, indicating a significant positive linear relationship. IFOr, in contrast, displayed a negative coefficient, indicating connectivity at this particular node was a negative predictor of performance. These findings largely mirror our earlier results, suggesting the importance of hippocampus, prefrontal cortex, and parts of parietal cortex to successful memory retrieval.
