The logic of co-citation
Small (1973) introduced co-citation as a measure of relatedness: two documents are co-cited when both appear in the reference list of a third. The more often that happens across a literature, the stronger the tie between them. Crucially, the link is created by citing authors rather than by the two documents themselves, so co-citation captures how the community has come to perceive those works, and that perception can evolve for decades after publication.
Mapped across a whole field, co-citation clusters reveal schools of thought, methodological traditions and the canonical texts each of them leans on. It is the technique of choice when your question is about foundations: what this literature builds on, and how those roots relate to one another. It is less suited to describing what is happening in the most recent couple of years, which is where other techniques come in.
Document, author and journal co-citation
Document co-citation
Nodes are individual cited works. Best for identifying seminal papers and the theoretical or methodological clusters they form.
Author co-citation
Nodes are cited authors. Useful for describing intellectual traditions associated with particular scholars.
Journal co-citation
Nodes are cited sources. Shows which disciplines and journals a field draws its knowledge from.
Document co-citation is the most common starting point because its clusters can be traced back to specific texts you can read. Author and journal co-citation are coarser but valuable for interdisciplinary topics, where the question is often which neighbouring fields are feeding ideas into the area. We usually recommend one primary level and, at most, one supporting level, so the paper stays readable.
Co-citation or bibliographic coupling?
The two are often confused. Bibliographic coupling, proposed by Kessler (1963), works in the opposite direction: two papers are coupled when they cite the same source. The contrast matters when choosing a design:
| Aspect | Co-citation | Bibliographic coupling |
|---|---|---|
| Nodes | Cited works | Papers in your dataset |
| Time orientation | Backward: intellectual foundations | Present: current research front |
| Does the tie change? | Yes, it grows as new citing papers appear | No, it is fixed at publication |
| Recent papers | Poorly represented because of citation lag | Well represented, including new publications |
For many projects we recommend running both, using co-citation to describe the foundations and bibliographic coupling to describe the streams that are building on them today.
How we run a co-citation study
Records with full references
Co-citation is impossible without complete cited reference fields, so we verify their presence and completeness before anything else.
Reference disambiguation
The same book can be cited with different years, author spellings, editions or page numbers. We merge those variants so one work does not appear as several weak nodes.
Unit and citation threshold
We choose a minimum number of citations that retains roughly several dozen to a few hundred works, detailed enough to be informative and small enough to read.
Network and clustering
The network is built in VOSviewer with association strength normalisation, and cluster stability is checked across neighbouring thresholds.
Cluster tables
Every cluster is listed with its works, local citation counts and link strength, giving you a reading list for each school of thought.
Caveats worth stating in your paper
- Citation lag: important recent work has not yet had time to be co-cited, so maps lean towards older literature.
- First-author bias: in some export formats the cited reference field records only the first author, so author co-citation mainly reflects first authors.
- Local versus global citations: citations within your dataset differ from total citations in the database; we always state which one is used.
- Works outside the database: books, reports and theses appear as nodes because they are cited, even though they have no record of their own in Scopus or Web of Science.
Stating these caveats openly tends to reassure reviewers rather than weaken the paper, because it shows the method has been applied with awareness of its limits. Where a caveat could materially change the picture, for instance a field in which key texts are books poorly indexed by citation databases, we test an alternative threshold or a second data source and tell you whether the main clusters survive the change.
Reporting the analysis
Your methods section should name the unit of analysis, the citation threshold, how many cited works met it out of the total, how references were disambiguated, and the normalisation and clustering settings. Our technical report provides every figure, and the reference concordance table can go straight into supplementary material so readers can see how variants were merged.
An illustrative, hypothetical example of the detail involved: 41,250 distinct cited references were reduced to 38,900 after variant merging, and 118 works cited at least 25 times were retained for the document co-citation map. We also suggest reporting whether citation counts are local or global, and how many clusters resulted at the chosen resolution, because reviewers frequently ask for exactly those two details.
The narrative that follows, explaining what each cluster represents, how the schools of thought differ and why their relationship matters for your research question, is where your expertise as an author is essential. We provide the evidence base; the argument is yours.
Scope and integrity
All reference merges are listed in the concordance table, so anyone can verify that no work was added, removed or strengthened without a documented reason. Read our guide to co-authorship and co-citation analysis, check pricing or place an order.