Not all paths lead to Rome: Analysing the network of sister cities

Andreas Kaltenbrunner, Pablo Aragón, David Laniado, Yana Volkovich

Abstract

This work analyses the practice of sister city pairing. We investigate structural properties of the resulting city and country networks and present rankings of the most central nodes in these networks. We identify different country clusters and find that the practice of sister city pairing is not influenced by geographical proximity but results in highly assortative networks.

Introduction

Sister city partnerships, often referred to as "twin towns," are formal agreements between cities intended to foster cultural and economic exchange. While social network analysis (SNA) has extensively mapped how individuals interact, we know much less about how larger institutional entities like cities and countries self-organize. This study presents the first large-scale structural analysis of the global sister city network, investigating whether these institutional relationships follow the same patterns—such as "distance decay"—as person-to-person networks.

Methods

The researchers employed a multi-step computational approach to analyze the network:

  1. Network Construction: Two distinct networks were built: a city network (nodes are cities, edges are partnerships) and a country network (edges weighted by the number of partnerships between cities in those countries).
  2. Data Acquisition: 15,225 pairs of sister cities involving 11,618 unique nodes were scraped from English Wikipedia. 11,483 cities were geocoded using the Google Maps API. Economic and social indices, including GDP, the Human Development Index (HDI), and a Political Stability Index, were integrated to analyze mixing preferences.
  3. Analysis Metrics: Standard SNA properties (degree distribution, clustering coefficients, path lengths) were calculated. Assortativity was tested against 100 randomized null models, and community detection was performed using the Louvain method.

Results

The "Death of Distance"

In most social networks, the probability of connection decreases as geographical distance increases. Sister cities are a radical outlier to this rule. The distribution of distances between sister cities is nearly identical to a random distribution of any two cities on Earth. The average distance between sister cities is 9,981 km, compared to 10,006 km for two randomly chosen cities—a negligible difference.

Geographical distance distribution of sister-city pairs vs. random city pairs.

Network Topology and Centrality

The city network exhibits "small world" properties with an average path length of 6.7. Connectivity is highly skewed: over 60% of cities have only one sister city, while a small number of "hubs" maintain many connections. Saint Petersburg (Russia) emerged as the most central city in the network, acting as a primary bridge between Eastern and Western cultural spheres. At the country level, the USA, France, and the UK lead in total connections.

Assortative Mixing and Global Clusters

The network is highly assortative, meaning cities seek out partners with similar economic health (GDP per capita), status (degree), and political environments. The Louvain method identified four main global clusters based on these twinning tendencies:

  • Western Europe and former colonies (Red): Including Germany, UK, France, and the Netherlands.
  • Eastern Europe and the Balkans (Green): Including Russia, Turkey, and Poland.
  • Nordic countries (Cyan): A distinct cluster of Northern European nations.
  • Global/Intercontinental (Violet): A broad group including the USA, Spain, and much of South America, Asia, and Africa.
Visualization of the global country network and its four main clusters.

Conclusion

Institutional diplomacy through sister city pairing appears to be driven by shared economic and political characteristics rather than geographical proximity. While the reliance on English Wikipedia introduces a potential Anglophone bias, this analysis provides a foundational map of global city-to-city relations. Future research could benefit from comparing this diplomatic network with functional networks, such as international trade or air traffic data, to see how formal twinning correlates with the actual flow of people and goods.

Summary updated

Abstract

This work analyses the practice of sister city pairing. We investigate structural properties of the resulting city and country networks and present rankings of the most central nodes in these networks. We identify different country clusters and find that the practice of sister city pairing is not influenced by geographical proximity but results in highly assortative networks.
BibTeX
@misc{kaltenbrunner2013not,
  author = {Andreas Kaltenbrunner and Pablo Aragón and David Laniado and Yana Volkovich},
  title = {{Not all paths lead to Rome: Analysing the network of sister cities}},
  year = {2013},
  eprint = {1301.6900},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1301.6900},
}
Figure