Fig. 2
The impact of geography and industry across scales. The top-level communities of this network found through the Louvain method (see Methods) have a modularity of 0.47. a We recursively apply network community detection to discover the labor flow network’s hierarchical structure. See Methods for more details. b Both industry and geography affect job transition across all scales, but industry has a more important role in the middle of the hierarchy as seen by the proportion of communities with a greater reduction in industry entropy (ρind). c The average reduction of metadata entropy \(\ {\bar{{\!\!}\boldsymbol d}}\) (see Methods for the definition) at each level of the hierarchical community structure, calculated with respect to the whole network. The monotonic increase indicates that smaller communities are more homogeneous as expected. d This entropy reduction is greater than expected by a null model. The difference between the observed entropy reduction and the reduction in a randomized hierarchical null model is denoted as Δ. Positive Δ indicates that the homogeneity of clusters is stronger than expected
