We make a final technical point before turning to data. In rs-fcMRI networks, the similarity measures used to define edges (often Pearson or partial correlations) yield weights on the edges. For example, in networks formed from correlation matrices, all nodes relate to all other nodes with edges weighted −1 to 1. Negative edges are mathematically troublesome in many graph analyses, and edges with weights near zero are likely to be uninformative, and so a threshold is often applied to networks, eliminating edges with weights below the threshold. As thresholds rise, the density of edges in the network decreases, and at some point the network will begin to fragment into disconnected components. The actual structure of the network changes as edges are removed, and thus many network properties are functions of edge density. Additionally, there is no “correct” edge density or threshold at which to examine a network. For this reason, when investigating some property of interest, network studies should report results over a range of thresholds or edge densities to demonstrate this relationship, as well as the reliability of the
