To find out, we measure co-voting similarity networks in the US Senate and trace individual careers over time. Standard network visualization tools fail on dense highly clustered networks, so we used two aggregation strategies to clarify positional mobility over time. First, clusters of Senators who often vote the same way capture coalitions, and allow us to measure polarization quantitatively through modularity (Newman, 2006; Waugh et al., 2009; Poole, 2012). Second, we use role-based blockmodels (White et al., 1976) to identify role positions, identifying sets of Senators with highly similar tie patterns. Our partitioning threshold for roles is stringent, generating many roles occupied by single Senators. This combination allows us to identify movement between positions over time (specifically, we used the Kernighan–Lin improvement of a Louvain method greedy partitioning algorithm for modularity [Blondel et al., 2008], and CONCOR with an internal similarity threshold for roles; see Supplementary materials for details).