Fig. 4

Results from sampling SBM with varying levels of correlation between a binary covariate and the community membership. On the x-axis there is the ground truth correlation (the one used to generated the synthetic network). On the y-axis, we show the difference between this correlation and the one calculated after sampling. Standard deviations are computed over 10 different rounds of sampling 10% of all the nodes. Notice how the y-axis is restricted around the 0 values, denoting small deviations from the full network’s value. Notice that snowball sampling, being a deterministic method when starting from the same seed node (as in this experiment), presents no variance