A small world of citations? The influence of collaboration networks on citation practices

1. Introduction

• Earlier work combining surveys and bibliometrics found some correlation between collaboration and citation patterns, but no strong or reliable link overall.That study covered only 16 individuals.Source: §1 · pp. 1–2
• Citation-network analyses reveal intellectual cohesiveness but do not establish how social networks affect referencing.Prior work also combined social and semantic networks to study knowledge production in epistemic communities.Source: §1 · p. 2
• The study asks whether researchers’ social networks affect which references appear in their papers, using co-authorship as an indicator of social proximity.The authors argue that co-author ties are likely stronger than ties to non-co-authors, even though co-authorship captures only part of a scientist’s social network.Source: §1 · p. 2
• The paper conducts a large-scale study using more than 2.6 million papers and 50 million references spanning over 50 years.It examines four levels of social proximity in eight specialties across the natural and medical sciences and the social sciences and humanities.Source: §1 · p. 2

2. Methods

• The study uses Web of Science records from 1945–2008 and examines eight specialties, including research articles, notes, and reviews.The data include SCIE, SSCI, and AHCI indexes.Source: §2 · p. 2
• The analysis identifies specialty authors and their collaborators, builds an unweighted, undirected co-author network, and classifies each source article’s references.References are limited to identifiable database source items published within the previous 10 years; collaboration is counted within a two-year window.Source: §2 · p. 3
• References are assigned to mutually exclusive categories: self-citations, level-1 co-author citations, level-2 co-author citations, or distant citations.When a reference fits multiple categories, it receives the category closest to self-citation; levels 1 and 2 represent decreasing social proximity.Source: §2 · p. 3
• The method tracks proximity only through two co-author levels, treating more distant authors as having no social connection within a specialty.The authors cite both that sociological boundary and the data-mining costs of deeper searches as reasons for stopping at level 2.Source: §2 · p. 3
• The analysis does not disambiguate authors sharing surnames and initials, which may overestimate self-citations in specialties with high co-authorship.The authors estimate homonyms at less than 5% of total positives based on small samples.Source: §2 · p. 3
• Citing items and authors are restricted to a specialty, but cited items are not; the authors checked that results were similar with or without restricting cited items to that specialty.They note that a closed-system restriction could introduce an artefact, particularly in interdisciplinary specialties such as biochemistry.Source: §2 · p. 4

A) Macroscopic properties of scientific specialties

• The benchmark combines measures of field size and collaboration with reference volume, intradisciplinary citation, and reference age.The authors use these variables to characterize scientific fields and contextualize proximity between citing and cited authors.Source: A) Macroscopic properties of… · p. 4
• Figure 2 reports six measures for each specialty: paper counts, references per paper, authors per paper, author productivity, within-specialty references, and recent references.Recent references are defined as less than 10 years older than the source item.Source: Figure 2 · p. 5

• There is no strong correlation between the tendency to cite recent literature and the share of that literature that is socially proximate.Source: B) Empirical evidence for… · p. 6
• 20% of references in NMS specialties and 10% in SSH are direct self-citations, with these proportions relatively constant across fields and over time.Source: B) Empirical evidence for… · p. 6
• The three SSH fields show no co-authorship citations, largely because co-authorship is less frequent and leaves fewer co-authors in researchers’ networks to cite.The authors therefore focus most of their analysis on NMS disciplines.Source: B) Empirical evidence for… · p. 6
• Restricting NMS source articles to five authors or fewer increases distant citations at the expense of level-1 and level-2 co-author citations, without affecting self-citations.The comparison uses dashed gray lines in Figure 3.Source: B) Empirical evidence for… · p. 6, Figure 3 · p. 7
• Stable self-citation levels contrast with citations to recent collaborators; distant citations can point to potentially independent subgroups connected through citation rather than co-authorship.The authors present these co-author citations as evidence of social proximity and distant citations as a complement to micro-level analysis of citation reasons.Source: B) Empirical evidence for… · p. 6
• Recent increases in citing socially proximate authors are partly associated with larger collaborations, as seen in astrophysics and astronomy.In some fields, co-authorship reflects instruments or recognition of a wider division of labor beyond writing the article.Source: B) Empirical evidence for… · pp. 6–7
• Reference proximity varies greatly by specialty; specialty size, co-authors per paper, and intra-specialty references do not alone explain the observed trends.Source: B) Empirical evidence for… · pp. 5, 7, Figure 3 · p. 7

C) Topology of the networks and citation practices

• Differences in co-author counts do not closely track social-reference rates, and level-2 co-author availability varies widely among papers within a network.Figure 4 compares co-author distributions for 1960–1969 and 2000–2006.Source: C) Topology of the networks and… · p. 8, Figure 4 · p. 8
• Reducing skew in author-per-paper distributions and diluting networks produced no major effect on citation proximity; k-cores and clique counts reflected the differences only weakly.Source papers were randomly removed, including up to around 15% of a network, while retaining its general shape.Source: C) Topology of the networks and… · p. 8
• For each paper, propensity compares empirical citation matches with a random-selection expectation, removing already matched references sequentially by proximity level.Source: C) Topology of the networks and… · p. 9
• Level-2 citation propensity is very low; self-citation propensity is an order of magnitude above level-1, and overall propensity rose from the 1950s, slowing after the 1970s.Table 1 reports propensity for self-citation and level-1 and level-2 co-author citations across three time periods.Source: C) Topology of the networks and… · p. 9, Table 1 · p. 9
• Large fields containing groups working on largely independent topics tend to have high propensities for self-, level-1, and level-2 citations.Source: C) Topology of the networks and… · p. 9
• Level-1 citations are far from random across specialties, while organic chemistry and, less strongly, biochemistry recently trend toward fewer socially close citations.Source: C) Topology of the networks and… · p. 10
• In 1985, astrophysics had many papers with low level-1 propensity, whereas economics and atmospheric sciences were dominated by a few high-propensity papers; level-2 had no non-zero local maximum.Source: C) Topology of the networks and… · p. 10, Figure 5 · p. 10

4. Discussion

• Co-author networks reveal only micro-fragmentation: disconnected groups may still cite one another when they work on similar topics.Astrophysics and atmospheric sciences show high co-authorship citations linked to large author groups and limited information-seeking outside the specialty; sociology differs, while economics remains very low.Source: §4 · p. 11
• Organic chemistry has low, stable co-authorship citations, and only a few researchers cite their collaboration network heavily; self-citation may also be declining.The authors suggest declining perfunctory referencing or decreasing value placed on the local network as possible interpretations.Source: §4 · p. 11
• High propensity to cite a local network may reflect social or cognitive reliance, while hyperauthorship may make some co-author citations less perfunctory.Source: §4 · p. 11
• The analysis points to closer research groups, increased co-author citations, and more fragmented research topics and groups, alongside a correlation between high co-author referencing and high intra-specialty citation.Source: §4 · p. 11

• Field and time differences reflect co-authorship levels, while subtler changes track reliance on local networks and their shape, likely linked to fragmentation and intra-specialty referencing.Source: §5 · p. 12
• The gap in co-author citations between social and natural sciences remains large, reflecting differences in co-authorship levels and citation practices.Source: §5 · p. 12
• Self-citation remains constant over time and across specialties, but is much lower in the social sciences.Source: §5 · p. 12
• Propensity to cite co-authors and their co-authors varies widely across fields and papers, especially in social sciences and less in astronomy and astrophysics.Source: §5 · p. 12
• Authors who tend to cite collaborators also tend to cite collaborators of collaborators.Source: §5 · p. 12
• Proximate referencing does not necessarily indicate citation cartels or nepotism, though compact fields may make arm’s-length reviewers harder to locate.Source: §5 · p. 12