Figure 3

GILs are scale free at varied partition sizes. Primary graph shows log-likelihood fit to a Kronecker scale-free graph model at k-means cluster sizes K. Clusters were randomized at 0% (solid bars) and 50% (striped bars) prior to network construction. Inset graph shows average scaling exponent (<γ>) for GILs constructed from K-means of cluster size K at 0% (solid line) and 50% (dashed line) randomization.