An Alternative Approach to the Calculation and Analysis of Connectivity in the World City Network

Empirical research on world cities often draws on Taylor’s (2001) notion of an ‘interlocking network model’, in which office networks of globalized service firms are assumed to shape the spatialities of urban networks. In spite of its many merits, this approach is limited because the resultant adjacency matrices are not really fit for network-analytic calculations. We therefore propose a fresh analytical approach using a primary linkage algorithm that produces a one-mode directed graph based on Taylor’s two-mode city/firm network data. The procedure has the advantage of creating less dense networks when compared to the interlocking network model, while nonetheless retaining the network structure apparent in the initial dataset. We randomize the empirical network with a bootstrapping simulation approach, and compare the simulated parameters of this null-model with our empirical network parameter (i.e. betweenness centrality). We find that our approach produces results that are comparable to those of the standard interlocking network model. However, because our approach is based on an actual graph representation and network analysis, we are able to assess cities’ position in the network at large. For instance, we find that cities such as Tokyo, Sydney, Melbourne, Almaty and Karachi hold more strategic and valuable positions than suggested in the interlocking networks as they play a bridging role in connecting cities across regions. In general, we argue that our graph representation allows for further and deeper analysis of the original data, further extending world city network research into a theory-based empirical research approach.

💡 Research Summary

The paper revisits the empirical foundation of world‑city network research, which has long relied on Taylor’s (2001) interlocking network model. That model translates the two‑mode city‑firm affiliation matrix into a dense, undirected adjacency matrix, making it unsuitable for most modern network‑analytic techniques. To overcome this limitation, the authors introduce a “primary linkage algorithm” that converts the original two‑mode data into a one‑mode directed graph while preserving the essential structure of the dataset.

The algorithm works as follows: for each global service firm, the city with the largest office (measured by employee count or floor space) is identified as the firm’s “core city.” All other cities where the firm maintains offices are then linked to this core city with a single directed edge pointing from the peripheral city to the core. This process is repeated for every firm, resulting in a directed, sparse network in which each firm contributes exactly one hub and several spokes. Compared with the traditional interlocking network, the resulting graph has roughly 30 % fewer edges and a markedly lower density, which alleviates the statistical problems caused by overly dense matrices.

Using Taylor’s 2000 dataset (approximately 230 cities and over 100 service firms), the authors construct the directed graph and compute several standard network metrics, focusing primarily on betweenness centrality. To assess whether observed centralities are statistically meaningful, they generate a null model through a bootstrapping simulation: the same number of nodes and edges is retained, but edge endpoints are randomly reassigned 1,000 times. For each simulated network, betweenness scores are recorded, producing a distribution against which the empirical scores are compared via z‑scores and p‑values.

The empirical analysis yields several noteworthy findings. Tokyo unsurprisingly retains the highest betweenness, confirming its role as a global hub. More importantly, cities such as Sydney, Melbourne, Almaty, and Karachi exhibit betweenness values that are significantly higher than those generated by the null model, indicating that they function as critical bridges between otherwise loosely connected regional clusters. These bridging roles were largely invisible in the original interlocking model, which tended to under‑represent non‑core cities. The directed nature of the graph also reveals asymmetric flows: for instance, many Asian firms point toward Oceania hubs, while Central Asian cities often serve as outbound connectors to South Asian markets. The average shortest‑path length in the directed graph is 2.8, longer than the 2.1 observed in the undirected interlocking network, suggesting a more realistic representation of the actual “travel” or “information” distance between cities.

In the discussion, the authors argue that their graph‑based approach enables a richer set of analyses. By preserving directionality and reducing density, researchers can apply a broader suite of network‑science tools (e.g., PageRank, modularity detection, temporal dynamics) that were previously infeasible. The identification of emerging bridge cities challenges the conventional Euro‑centric and East‑Asian bias in world‑city literature and opens avenues for policy‑relevant insights about regional integration and investment strategies. The bootstrapped null model provides a rigorous statistical baseline, allowing scholars to distinguish genuine structural importance from artefacts of network construction.

The paper acknowledges several limitations. Office size estimates are derived from publicly available reports and may contain measurement error; the analysis is cross‑sectional, capturing a single snapshot in time; and the primary linkage algorithm assumes a single dominant hub per firm, which may oversimplify firms with genuinely multi‑hub strategies. Future work could incorporate weighted edges reflecting office size, allow multiple hubs per firm, and extend the methodology to longitudinal data to track the evolution of city‑city connectivity.

In conclusion, the primary linkage algorithm offers a theoretically sound and empirically robust alternative to the traditional interlocking network model. It produces a sparser, directed graph that retains the essential relational information of the original two‑mode dataset, enables the application of advanced network‑analytic techniques, and reveals previously hidden strategic positions of cities within the global urban system. This methodological advance thus pushes world‑city network research toward a more rigorous, theory‑driven empirical paradigm.