The large-scale structure of journal citation networks

We analyse the large-scale structure of the journal citation network built from information contained in the Thomson-Reuters Journal Citation Reports. To this end, we take advantage of the network science paraphernalia and explore network properties like density, percolation robustness, average and largest node distances, reciprocity, incoming and outgoing degree distributions, as well as assortative mixing by node degrees. We discover that the journal citation network is a dense, robust, small, and reciprocal world. Furthermore, in and out node degree distributions display long-tails, with few vital journals and many trivial ones, and they are strongly positively correlated.

💡 Research Summary

The paper presents a comprehensive network‑science investigation of the large‑scale structure of journal citation relations using data from the Thomson‑Reuters Journal Citation Reports. The authors construct a directed, unweighted graph where each node represents a journal indexed in the JCR (total = 6,708) and a directed edge from journal A to journal B exists if a 2008 article in A cites a paper published in B between 2003 and 2007. Only article and review document types are considered. The analysis focuses on the largest strongly connected component, which contains 1,315,238 edges, guaranteeing a directed path between any pair of journals.

Key structural metrics are examined:

1. Density – The overall edge density is 3 %, indicating a relatively dense citation environment. When restricting to the top‑30 journals by total degree, the subgraph becomes almost complete (93 % density), and even the top‑1,000 journals retain a high density of 32 %.

2. Robustness (Percolation) – Four node‑removal strategies are simulated: total degree, eigenvector centrality, closeness centrality, and betweenness centrality. Betweenness‑based removal most efficiently fragments the giant component, while eigenvector‑based removal is the least effective. To reduce the giant component to less than half of its original size, roughly 82 % of nodes must be removed under the optimal (betweenness) strategy, demonstrating strong resilience.

3. Path Lengths – The average geodesic distance between any two journals is 2.4 edges, and the network diameter is 6. This confirms a “small‑world” topology: most journals are reachable within two or three citation hops.

4. Reciprocity – The proportion of mutual citation pairs (2‑node cycles) is 0.29, meaning that nearly 30 % of citation events are reciprocated. This high reciprocity reflects tightly knit disciplinary clusters.

5. Degree Distributions – Both in‑degree (number of citing journals) and out‑degree (number of cited journals) exhibit long‑tailed distributions. The average degree is 196; the median out‑degree is 126 (max = 2,193 for PNAS) and the median in‑degree is 109 (max = 3,697 for Science). The in‑degree distribution is more skewed (skewness = 3.5, Gini = 0.55) than the out‑degree distribution (skewness = 1.9, Gini = 0.49). Neither follows a pure power law, so the network is not strictly scale‑free.

6. Degree Correlation – In‑ and out‑degrees are strongly positively correlated (Spearman ≈ 0.90, Pearson ≈ 0.87). High‑impact “authority” journals tend also to be “hubs” that cite many others. Correlations with journal size (number of published papers) are moderate (≈ 0.72 for in‑degree, ≈ 0.55 for out‑degree), indicating that structural prominence is not solely a function of volume.

7. Assortative Mixing – Positive assortativity is observed for both in‑ and out‑degree, meaning high‑degree journals preferentially connect to other high‑degree journals.

Overall, the journal citation network is dense, robust, and exhibits small‑world characteristics. A relatively small set of interdisciplinary journals with high betweenness centrality act as critical bridges that sustain global connectivity, while the majority of journals form tightly clustered, reciprocally citing communities. These findings illuminate how scholarly communication is organized at the macro level and suggest that interventions targeting bridge journals could significantly affect the flow of scientific knowledge across fields.