Fig. 2

Communication and affiliation subnetworks of senators. Community detection algorithms based on leading eigenvector centrality show that the senators in communication (a) and affiliation (b) networks cluster based on party membership (red for Republican, blue for Democrat). The community detection algorithms correctly identified 100 % of Senate Democrats and 88 % of Senate Republicans in the direct communication network, and 97 % of Senate Democrats and 90 % of Senate Republicans in the affiliation network. Misidentified senators are plotted in purple. The size of the nodes is proportional to senators’ leading eigenvector centrality. The modularity scores, which describe the degree of compartmentalization in the network, are 0.165 for the community network and 0.093 for the affiliation network. These networks are significantly more modular than expected based on networks generated by randomly shuffling the accounts that each senator follows (p-value < 0.001)