Figure 3:

The top panel shows the distribution of the difference between the MASE estimates and true eigenvalues of a single graph. The graphs in the sample are distributed according to a two-block multilayer SBM, block connection probability matrix B(i) = 0.3I + 0.111⊤, and a fixed sample size m = 3 but different number of vertices n. The estimation bias remains positive even with a large graph size (bottom panel), and for any fixed n the mean of the distributions (red solid line) is away from zero (blue dashed line), but the distributions appear to be Gaussian.