Alongside the keyword map, the two analyses reviewers ask about most are co-authorship and co-citation. They answer different questions: one describes the social structure of a field, the other its intellectual structure. This guide explains the concepts, the technical choices that matter and the interpretation traps that catch many authors.
Two analyses, two different questions
Co-authorship
Two units are linked when they appear together on a publication. It maps collaboration between authors, organizations or countries.
Co-citation
Two documents are linked when a later paper cites both (Small 1973). It maps the knowledge base the community treats as related.
Bibliographic coupling
Two papers are linked when they cite the same earlier work (Kessler 1963). It maps the current research front.
Choosing the unit for co-authorship
| Unit | Best for | Data caveats |
|---|---|---|
| Authors | Finding research groups and central researchers | Homonyms and initials cause false merges or splits; cross-check author identifiers. |
| Organizations | Comparing universities and institutes | One institution may appear under many names; standardize with a thesaurus. |
| Countries | Describing international collaboration | The most stable unit, though variants such as "Viet Nam" and "Vietnam" still need merging. |
For author-level maps, VOSviewer by default ignores documents with very many authors (a maximum authors per document setting). Large consortium papers can list hundreds of names and distort the network, so say clearly whether you kept or excluded them.
Cleaning names before you map
Author-level maps deserve extra care. Two different researchers named "Nguyen, T." can be merged into one node, while one researcher who sometimes publishes with a middle initial and sometimes without can be split into two. Scopus author identifiers help, but they are not perfect either. For organization maps, the same university may appear under its full name, an abbreviation, a former name or a faculty name. Build a thesaurus for the top 50 to 100 units, check it against the affiliation strings in the raw data, and report that names were standardized. Countries need the least work, but still merge variants and confirm that records with missing affiliations are a small share.
Full counting versus fractional counting
This choice is often overlooked, yet it can noticeably change the result.
- Full counting: every link between two co-authors gets a weight of one, however many authors the paper has. A 50-author paper creates 1,225 links, each at full weight.
- Fractional counting: weights are divided according to the number of authors, so every paper contributes the same total. This dampens the effect of papers with long author lists.
A small illustration makes the difference concrete. Suppose a dataset contains one paper by two authors from Vietnam and Japan, and one consortium paper with authors from 30 countries. Under full counting, the consortium paper adds a link of weight one between every pair of its 30 countries, so the map may suggest dense ties among countries that only ever met on that single paper. Under fractional counting, the consortium paper still counts, but its weight is spread across all those pairs, so the two-author paper is no longer drowned out.
Three flavours of co-citation
Co-citation analysis needs complete cited reference data. It comes in three forms:
- Cited references: each cited document is a node. This reveals schools of thought and seminal works and is the most detailed option.
- Cited sources: each cited journal or source is a node. This shows which knowledge domains a topic draws on.
- Cited authors: each cited author is a node. Useful but sensitive to homonyms, and usually based only on the first author in each reference string.
The same work is often cited in slightly different ways across papers (missing pages, wrong year, abbreviated journal names), so a classic reference can be split into two or three nodes. Review the top nodes and merge variants with a thesaurus before you interpret anything.
Thresholds for co-citation are usually higher than for keywords, because the number of distinct cited references in a dataset can easily run into the tens of thousands. Start with a minimum of around 20 citations for a few thousand records and adjust from there. For cited sources, a lower threshold often works, since citations concentrate on fewer journals. Whatever values you settle on, check that the most cited works in your field actually appear on the map; if they do not, the problem is usually split reference variants rather than the threshold itself.
Co-citation versus bibliographic coupling
| Criterion | Co-citation | Bibliographic coupling |
|---|---|---|
| Link based on | Being cited together by later papers | Citing the same earlier works |
| Time orientation | Looks back to the knowledge base | Looks at the present research front |
| Stability | Changes as new citations arrive | Fixed once a paper is published |
| Recent papers | Rarely appear; not yet cited enough | Appear immediately |
Many studies use both: co-citation to describe the foundations and coupling to describe the lines of work that are active today.
The same analyses are available in bibliometrix through biblioNetwork with analysis set to "collaboration", "co-citation" or "coupling". Running both tools on the same dataset is a useful robustness check: if the main clusters look similar in VOSviewer and bibliometrix, you can present them with more confidence. If they differ sharply, look first at the counting method and thresholds, which are the usual culprits.
Running the analysis in VOSviewer
Load complete records
Confirm the export actually contains the cited references field.
Pick analysis and unit
Co-authorship by countries, or co-citation by cited references; choose the counting method.
Set the threshold
For example, references cited at least 20 times; adjust to roughly 50 to 100 nodes.
Merge variants and rerun
Apply a thesaurus to organization names, countries or split references.
Reading a co-authorship map
Start with the size of the nodes, which reflects the number of documents, and the thickness of the links, which reflects the strength of collaboration. Then look at structure. A network dominated by a few large hubs suggests that collaboration in the field is concentrated around a small number of countries or institutions. Several separate clusters may indicate regional or linguistic communities that rarely publish together. Isolated nodes, which VOSviewer may hide if you show only the largest connected component, often represent actors who publish actively but mostly alone.
Total link strength is a useful complement to document counts. A country with modest output but high link strength is well connected internationally; a country with high output and low link strength mostly publishes domestically. Reporting both numbers in a table next to the map makes these contrasts visible without overstating them. The overlay view by average year can also show which partnerships are recent.
Reading a co-citation map
In a co-citation map of references, each cluster usually corresponds to a school of thought, a methodological tradition or a body of empirical work. To label a cluster, read the most cited references in it rather than guessing from titles. Methodological classics, such as statistical techniques or software papers, often form their own cluster or sit in the centre because many papers cite them regardless of topic; it is reasonable to note them and, if they add nothing to the question, exclude them in a rerun.
References that link two clusters are worth special attention, because they often mark the works that brought ideas from one tradition into another. Finally, remember that a co-citation map describes how the citing papers in your dataset use the literature. It says nothing about literature your dataset does not cite, so the scope of your search directly shapes the intellectual base you will see.
Interpretation pitfalls
- Co-authorship is not always deep collaboration: shared bylines may reflect data sharing, supervision or disciplinary norms.
- Highly cited does not mean endorsed: a paper may be cited to be criticized.
- Database bias: Scopus and Web of Science cover languages and regions unevenly, which affects how countries such as Vietnam appear.
- Citation lag: new work needs time to gather citations, so co-citation always favours older literature.
- No individual rankings: avoid using these maps to judge the merit of a specific researcher.
When you write up these analyses, state the unit, counting method, threshold, any limit on authors per document and how names or references were standardized. A sentence such as "Country co-authorship was analysed with fractional counting; documents with more than 25 authors were excluded; country names were standardized with a thesaurus" gives reviewers everything they need to judge and reproduce the map.
Technical support for network analysis
The Biblio VietSCI co-citation analysis service covers standardizing author, organization and country names, merging reference variants, running co-authorship, co-citation and bibliographic coupling, and delivering maps and data tables with all source files. Arguments and scientific conclusions remain the author's responsibility. See pricing or submit a request on the order page.
Frequently asked questions
What is the difference between co-citation and bibliographic coupling?
Co-citation links two documents when a later paper cites both, reflecting the knowledge base. Bibliographic coupling links two papers when they cite the same earlier work, reflecting the current research front.
Should I use full or fractional counting?
Fractional counting reduces the influence of papers with many authors and is often recommended for co-authorship networks. Either way, report your choice in the Methods section.
Why does one classic reference appear as several nodes?
Because citing papers record the reference inconsistently. Review the top nodes and merge variants with a thesaurus before interpreting the map.
Does a country co-authorship map need cleaning?
Yes, though less than an author map. Merge different spellings of country names and check for missing affiliation data.
Can co-authorship maps be used to evaluate researchers?
They should not be. The maps describe aggregate publication patterns and do not measure the ability or research quality of individuals.