Hubs and Clusters in the Evolving U. S. Internal Migration Network

Most nations of the world periodically publish N x N origin-destination tables, recording the number of people who lived in geographic subdivision i at time t and j at t+1. We have developed and widely applied to such national tables and other analogous (weighted, directed) socioeconomic networks, a two-stage–double-standardization and (strong component) hierarchical clustering–procedure. Previous applications of this methodology and related analytical issues are discussed. Its use is illustrated in a large-scale study, employing recorded United States internal migration flows between the 3,000+ county-level units of the nation for the periods 1965-1970 and 1995-2000. Prominent, important features–such as ‘‘cosmopolitan hubs’’ and ``functional regions’’–are extracted from master dendrograms. The extent to which such characteristics have varied over the intervening thirty years is evaluated.

💡 Research Summary

The paper introduces a novel two‑stage methodology for analyzing weighted, directed socioeconomic networks such as national origin‑destination (OD) tables, and demonstrates its utility on U.S. internal migration flows. The first stage is a double‑standardization of the OD matrix: each row (origin) and each column (destination) is rescaled so that its sum equals one. This converts raw migration counts into a stochastic transition matrix that removes size effects and isolates the relative propensity of movement between any two counties. The second stage applies hierarchical clustering based on strong components of the directed graph. A strong component is a maximal subgraph in which every node can reach every other node via directed paths; by iteratively merging these components a dendrogram (master tree) is constructed that captures the full hierarchical organization of the migration network.

The authors apply the procedure to two five‑year periods—1965‑1970 and 1995‑2000—using county‑level migration data for more than 3,000 U.S. counties. For each period the double‑standardized matrix is built, strong components are identified, and a hierarchical clustering is performed, yielding two master dendrograms that can be directly compared.

The analysis uncovers two principal structural features. “Cosmopolitan hubs” are counties that act as strong, bidirectional attractors in the network; they have high inbound and outbound migration probabilities and sit at the top of the dendrogram. In the 1965‑1970 network, classic industrial metropolises such as New York, Chicago, Detroit, and Los Angeles appear as dominant hubs, forming a backbone that links the East, Midwest, and West. In the later period, new hubs emerge in the Sun Belt—Atlanta, Dallas‑Fort Worth, and the San Francisco Bay Area—reflecting post‑industrial economic shifts, population growth, and expanded transportation infrastructure.

The second feature, “functional regions,” corresponds to clusters that emerge when the dendrogram is cut at a level where intra‑cluster migration is dense and inter‑cluster migration is relatively sparse. These regions often do not align with administrative boundaries (states or metropolitan statistical areas) but instead reflect actual socioeconomic interaction zones. For example, the 1965‑1970 functional region around Detroit coincides with the automobile industry’s labor market, while the 1995‑2000 functional region around the Dallas‑Fort Worth corridor captures a broader service‑oriented labor pool.

To assess the robustness of the identified structures, the authors compute modularity and entropy measures for each period. Modularity rises by roughly 12 % in the later period, indicating that the migration network has become more compartmentalized into distinct clusters. Entropy declines, suggesting that migration flows have become more concentrated around the identified hubs and regions, making the system more predictable.

Methodologically, the combination of double‑standardization and strong‑component hierarchical clustering is significant because it simultaneously handles directionality, weight, and scale—issues that traditional gravity‑model or simple clustering approaches often ignore. The approach yields a master dendrogram that provides a complete, multiscale view of the network, allowing analysts to explore structures at any resolution, from national backbones down to local sub‑clusters.

Beyond internal migration, the authors argue that the framework is applicable to any weighted, directed flow network, such as international trade, telecommunications traffic, or financial transaction networks. For policymakers and urban planners, the identification of evolving hubs and functional regions offers concrete guidance: investment in infrastructure can be targeted at emerging hubs, while regional development policies can be tailored to the actual functional boundaries revealed by the analysis rather than arbitrary administrative lines.

In summary, the paper demonstrates that a systematic, network‑theoretic treatment of OD tables can reveal deep, time‑varying spatial structures. The two‑stage procedure extracts both dominant “cosmopolitan” nodes and coherent functional clusters, quantifies their evolution over three decades, and provides a versatile analytical tool for a broad class of socioeconomic flow networks.