Citing a dataset is not the same task as citing a journal article, even though both use the word “citation.” A dataset is a living, versionable, sometimes-partial research object rather than a fixed text, and the citation conventions that have grown up around it, led by DataCite and formalized in the FORCE11 Joint Declaration of Data Citation Principles, exist specifically to handle that difference. This guide covers what belongs in a data citation, how the major style guides format one, and the underlying principles that citation practice is trying to satisfy.
Why citing data is different from citing an article
A journal article citation points to a fixed, published text that does not change after publication. A dataset citation has to do more work, because the object it points to may be updated, may be one version among several, may only be partially used in a given study, and may not have a human-readable “publication” at all in the traditional sense. Data citation conventions were built to solve four problems a plain article citation does not have to:
- Versioning: the same dataset can exist in multiple released versions, and a citation needs to specify which one was actually used.
- Granularity: a citation may need to point to a subset of a larger dataset (a specific file, table, or query result) rather than the whole collection.
- Persistence: unlike a journal’s own archive, a dataset’s long-term availability depends on the repository holding it, which is why a persistent identifier matters more, not less, than it does for an article.
- Non-traditional authorship and credit: datasets are often produced by teams, instruments, or consortia rather than a byline of named authors, and citation has to accommodate that.
These are exactly the concerns a properly structured data management plan anticipates before a dataset is ever deposited: naming who the citable creator is, which repository will mint the identifier, and under what conditions the data can be reused.
The core elements of a data citation
Whatever style guide you follow, a complete data citation is built from the same underlying elements, because they map directly onto the properties a repository must supply to register a dataset. DataCite’s Metadata Schema requires six mandatory properties before it will mint a DOI for a dataset: Identifier, Creator, Title, Publisher, PublicationYear, and ResourceType. DataCite’s own recommended citation format is built directly from that mandatory set:
Creator (PublicationYear): Title. Publisher. ResourceType. Identifier.
For example, a minimal DataCite-style citation looks like:
Smith, J. (2024). Survey responses on research data sharing attitudes [Dataset]. University Data Repository. https://doi.org/10.xxxx/xxxxx
Two elements are worth calling out specifically because they are easy to omit and both are load-bearing for reuse:
- Version: DataCite’s schema includes an optional but strongly recommended Version property. If the dataset you used has been updated since, citing without a version number means a reader cannot be certain they are looking at the same data you analyzed.
- Persistent identifier (PID): a DOI is the most common PID for datasets, minted through DataCite (as distinct from Crossref, which mints DOIs for articles). Some disciplines and repositories use other persistent identifier schemes, but the identifier, whichever scheme it uses, is what makes the citation resolvable and machine-actionable rather than just descriptive text.
Data citation in APA, MLA, and Chicago style
The major style guides each publish their own formatting rules for datasets, but all of them ask for the same underlying elements described above, just arranged and punctuated differently. In broad terms:
- APA (7th edition) treats a dataset as a work with a bracketed content-type descriptor, structured roughly as: Author. (Year). Title of data set [Data set]. Publisher/Repository. URL or DOI.
- MLA (9th edition) applies its general “container” model to data much as it would to any other digital source: Author. “Title of Dataset.” Repository/Database Name, Version, Publisher, Date, URL or DOI.
- Chicago (author-date) follows the same author-date logic used for other sources, with the repository standing in for a publisher and the identifier taking the place of a page range.
Because these style guides periodically revise their exact punctuation and ordering, always check the current edition of the style manual you’re required to use rather than relying on a remembered template. Many repositories (including data repositories built on DataCite) now auto-generate a suggested citation string in several formats on the dataset’s landing page, which is a reliable shortcut precisely because it is generated from the same structured metadata the repository registered.
The Joint Declaration of Data Citation Principles
Style-guide formatting tells you how to punctuate a citation; it doesn’t tell you why the practice exists or what a “good” data citation is trying to achieve. That’s the role of the Joint Declaration of Data Citation Principles, published by FORCE11 and endorsed across the scholarly communication community. It sets out eight principles:
- Importance: data should be considered a legitimate, citable product of research, on the same standing as a journal article, not merely supplementary material.
- Credit and Attribution: citations should facilitate giving scholarly credit to all contributors responsible for producing the data.
- Evidence: any claim relying on data should cite that data, in the same way a claim relying on prior published work would cite that work.
- Unique Identification: a data citation should include a standard identifier that is machine-actionable, globally unique, and widely used by a community.
- Access: citations should facilitate access to the data itself, and to the metadata, documentation, code, and other materials needed for its proper use.
- Persistence: identifiers and metadata describing data should persist, even beyond the life of the dataset that describes them or the repository in which they were originally deposited.
- Specificity and Verifiability: citations should facilitate identification of, access to, and verification of the specific data that support a given claim, including precise identification of the version and subset of data used.
- Interoperability and Flexibility: citation methods should be sufficiently flexible to accommodate the wide variety of practices and requirements across different research communities, but should not differ so much that they compromise interoperability of citation practices across communities.
These principles are why the mandatory-metadata approach described above exists: unique identification and specificity/verifiability are exactly what the required Identifier, Version, and ResourceType properties in DataCite’s schema are built to satisfy.
Citing data you didn’t create yourself
If you reuse a dataset someone else deposited, the citation obligation is the same as it is for any other secondary source: cite the dataset itself, using the creator, year, title, publisher, and identifier the depositing repository registered, not the paper that happens to describe it (unless you are also specifically citing claims made in that paper). Many repository landing pages provide a “preferred citation” or “how to cite this dataset” block for exactly this reason. Where a dataset has been assigned a separate DOI from any accompanying article, both should typically be cited if you are drawing on both the data and the paper’s analysis of it, since they are distinct citable objects under the Importance principle above.
Frequently asked questions
Is a data citation the same as a reference to a repository?
No. Citing “Zenodo” or “the NCBI Gene Expression Omnibus” by name is a description of where data lives, not a citation of the specific dataset used. A proper citation identifies the specific creator, title, version, and persistent identifier of the dataset itself, the repository is the publisher element within that citation, not a substitute for it.
Do I need to cite a dataset if I only used it for supplementary analysis?
Yes, under the Evidence principle above. If a claim in your work relies on the data, however incidentally, it should be cited, on the same basis that any other source a claim depends on should be cited.
What if the dataset has no DOI?
Cite the most persistent identifier available (an ARK, a handle, or a stable repository-assigned accession number are common alternatives), and include the access date, since without a persistent identifier the resource is more likely to move or change. Where possible, prefer datasets deposited in a repository that mints a proper persistent identifier, this is one of the practical reasons DMPs specify a named repository in advance.
Does citing a dataset’s version matter if I didn’t modify the data?
Yes. Even unmodified reuse should specify the version, because the repository or dataset itself may be updated after you accessed it. Citing without a version number leaves a reader unable to verify they are looking at the same data you used, which is precisely the Specificity and Verifiability principle above.
