Betweenness Centrality as a Driver of Preferential Attachment in the Evolution of Research Collaboration Networks

We analyze whether preferential attachment in scientific coauthorship networks is different for authors with different forms of centrality. Using a complete database for the scientific specialty of research about “steel structures,” we show that betweenness centrality of an existing node is a significantly better predictor of preferential attachment by new entrants than degree or closeness centrality. During the growth of a network, preferential attachment shifts from (local) degree centrality to betweenness centrality as a global measure. An interpretation is that supervisors of PhD projects and postdocs broker between new entrants and the already existing network, and thus become focal to preferential attachment. Because of this mediation, scholarly networks can be expected to develop differently from networks which are predicated on preferential attachment to nodes with high degree centrality.

💡 Research Summary

This paper investigates whether preferential attachment in scientific co‑authorship networks depends on different forms of node centrality, focusing on the “steel structures” research specialty. Using a longitudinal dataset extracted from Scopus (1999‑2009) that comprises 1,869 papers, 3,004 authors, 1,324 institutions across 77 countries, the authors construct yearly undirected weighted co‑authorship graphs. For each year‑end snapshot they compute three classic centrality measures for every existing author: degree (local popularity), closeness (global proximity), and betweenness (brokerage). They then track which existing authors attract new entrants (authors who appear for the first time in the following year) and model the probability of a new link forming as a function of the three centralities using logistic regression and event‑rate models.

The results reveal a clear temporal shift in the driver of preferential attachment. In the early phase (1999‑2002) degree centrality is the strongest predictor: new authors tend to co‑author with already highly connected scholars. However, from around 2005 onward, betweenness centrality becomes the dominant factor, while degree and closeness lose statistical significance or even show negative effects. This pattern suggests that as the network matures and sub‑communities emerge, the role of “brokers” – scholars who sit on many shortest paths between otherwise distant groups – supersedes simple popularity. The authors interpret high‑betweenness individuals as supervisors of PhD projects and post‑doctoral researchers who mediate between newcomers and the established community, thereby acting as gatekeepers of information and collaboration opportunities.

The study challenges the classic Barabási‑Albert model, which assumes that new nodes attach preferentially to high‑degree nodes, by demonstrating that in scholarly collaboration networks a global structural position (betweenness) can be more influential. The authors propose a refined preferential‑attachment model that incorporates brokerage centrality. They also discuss policy implications: supporting high‑betweenness researchers (e.g., senior mentors, project leaders) could accelerate the integration of newcomers and enhance overall network cohesion.

Limitations include the focus on a single discipline, a ten‑year window, and the exclusive use of co‑authorship as the proxy for collaboration, which may overlook other forms of scientific interaction. Future work should test the findings across multiple fields, longer time spans, and richer collaboration data (e.g., joint grants, conference co‑attendance).

In sum, the paper provides robust empirical evidence that betweenness centrality, rather than degree, drives preferential attachment in the evolution of research collaboration networks, highlighting the pivotal brokerage role of senior scholars in shaping scientific community structure.