How to cite datasets in academic papers
Cite a dataset the way you would cite a paper: creator, year, title, publisher, and a persistent identifier such as a DOI, displayed as a resolvable link rather than a plain string. The details below cover where that identifier comes from, what belongs in the reference, and what your journal's data availability policy expects on top of the citation itself.
The short answer
A complete dataset citation includes the creator, publication year, dataset title, the repository or publisher, a version number if the dataset has one, and a persistent identifier — usually a DOI — shown as a full resolvable link such as https://doi.org/10.xxxx/xxxxx. If the dataset has no DOI, substitute a stable URL plus an access date. Most journals also expect a short data availability statement separate from the reference list.
Why datasets need citations at all
The Joint Declaration of Data Citation Principles, published by FORCE11, sets out the underlying reasoning: data should be considered a legitimate, citable product of research in its own right, not merely a supporting file mentioned in passing. Citing a dataset the same way you cite a paper gives its creators credit, gives readers a way to verify your results, and gives the wider research community a way to track how often and how a given dataset is reused. Treat "data available upon request" or an unlinked mention in the methods section as a citation gap, not a substitute for one.
Where the identifier comes from: DOIs and DataCite
DataCite is the international consortium, founded in 2009, that registers DOIs for datasets and other research outputs across a global network of repositories and data centers. When a repository deposits a dataset with DataCite, it also submits structured metadata — creator, title, publisher, publication year, resource type — following the DataCite Metadata Schema. That is exactly the metadata a proper citation needs, which is why the easiest way to build a correct dataset citation is to pull the fields directly from the repository's own citation box rather than guessing at a format. DataCite's own guidance recommends displaying the DOI as a linkable, permanent URL (for example, https://doi.org/10.6068/DP15E5374E97A17) rather than the bare "doi:" prefix form.
What belongs in the citation
Regardless of the exact style your target journal uses (APA, Chicago, or a field-specific variant), a dataset citation needs six things: who created it, when it was published, what it is called, who published or hosts it, which version you used if the dataset is versioned, and a persistent identifier. If the dataset is dynamic — updated on a rolling basis rather than published as a fixed snapshot — cite the specific version or the access date you actually used, since the content behind the same DOI may differ from what a later reader retrieves.
What journals expect beyond the citation: data availability statements
Nature Portfolio journals require a data availability statement in every published research article, stating where the data behind the results can be found and, where relevant, any conditions on accessing it. Nature is also a signatory of the FORCE11 declaration and explicitly encourages citing datasets with DOIs in the reference list, treated the same as any other cited source, rather than only mentioning them in a availability statement. Many other publishers now follow the same pattern; check your target journal's author guidelines for the specific wording it requires, since the statement and the citation serve different purposes and most journals want both.
Citation elements checklist
| Element | Where to find it | Required? |
|---|---|---|
| Creator / author | Repository's citation box or dataset metadata | Yes |
| Publication year | Repository metadata (publicationYear field) | Yes |
| Dataset title | Repository landing page | Yes |
| Publisher / repository | Repository name (e.g. the archive that hosts it) | Yes |
| Version | Repository version history, if the dataset is versioned | If applicable |
| Persistent identifier | DOI resolved through doi.org, registered via DataCite | Yes if one exists; else a stable URL and access date |
FAQ
What information does a proper dataset citation need?
At minimum: creator, publication year, dataset title, publisher or repository, version if applicable, and a persistent identifier such as a DOI displayed as a resolvable link. This mirrors the core fields in the DataCite Metadata Schema, which most repositories use to register dataset DOIs.
Does every dataset have a DOI I can cite?
No. Datasets deposited in a repository that mints DOIs through DataCite or a similar registration agency will have one, but many datasets, especially older ones or those hosted on general project pages, do not. When no DOI exists, cite the stable URL, the version or access date, and as much of the creator and publisher information as is available.
What is a data availability statement and do I need one?
A data availability statement is a short section, required by most Nature Portfolio journals and many other publishers, that tells readers where the data behind a paper's results can be found and under what conditions it can be obtained. If your journal requires one, it goes alongside your citations, not instead of them — you still cite the dataset in the reference list if you used one.
Tracking down the dataset behind a citation, or finding one to cite?
DeepSData can run a real availability search and verify sources for you — locating the original dataset, checking whether a DOI exists, and confirming provenance before it goes in your reference list.