Most researchers learned to cite articles and books in graduate school. Fewer were ever shown how to cite a dataset, so data end up named in a methods paragraph ("we used the 2019 wave of the survey") and nowhere in the reference list. That's a problem for the reader, who can't tell which version you used, and for the people who built the data, who get no credit for it.
This guide covers the parts every data citation needs, two formats you can copy, how to handle versions and subsets, and the mistakes that make a data citation useless. It's written for researchers and grad students, and for the librarians and research office staff who get asked about this every semester.
Why a data citation matters
A citation does two jobs. It tells a reader exactly which data you used, so they can find and check them. And it gives credit, in the one currency academia counts. When a dataset sits in the reference list with a DOI, indexing services can link it to your paper and count its reuse. When it's only mentioned in the text, that link never gets made.
This isn't a niche view. The Joint Declaration of Data Citation Principles, published by FORCE11 in 2014 and endorsed by hundreds of publishers, repositories, and societies, says data should be cited like any other research output: in the reference list, with a persistent identifier, precise enough to point to the exact version used. Journal author guidelines increasingly say the same thing, often in the same paragraph as the data availability statement requirement.
The parts every data citation needs
Formats differ in punctuation. The parts don't. A usable data citation has:
- Creator. The people or organization responsible for the data. For government data, that's usually the agency.
- Year. The year the version you used was published or released, not the year the data were collected.
- Title. The dataset's own title, as the repository shows it.
- Version. A version number, release name, or, if there's neither, the date you accessed it. This is the part people skip, and it's the one that matters most for reproducing your results.
- Publisher. The repository or organization that holds and distributes the data (ICPSR, Zenodo, Dryad, a Dataverse, a statistical agency).
- Resource type. A label saying it's a dataset, so readers and indexing tools don't mistake it for an article.
- Identifier. A DOI written as a full link (https://doi.org/...), or another persistent identifier. A plain URL only if nothing better exists.
Two formats you can copy
APA 7
APA treats a dataset like other references, with the label [Data set] in square brackets after the title and the version in parentheses:
Author, A. A. (Year). Title of dataset (Version) [Data set]. Publisher. https://doi.org/xxxx
Here's one of APA's own examples, for a dataset held at ICPSR:
O'Donohue, W. (2017). Content analysis of undergraduate psychology textbooks (ICPSR 36966; Version V1) [Data set]. ICPSR. https://doi.org/10.3886/ICPSR36966.v1
Notice that the version is in two places: in the parentheses and in the DOI itself. ICPSR issues a new DOI for each version, which is exactly what you want.
DataCite
DataCite, the organization that issues most dataset DOIs, suggests a format many repositories display by default. It isn't mandatory, but it's a good neutral choice when a journal doesn't specify a style:
Creator (PublicationYear). Title. Version. Publisher. ResourceType. Identifier
Chicago, MLA, Vancouver, and discipline styles all have their own versions. If you're writing for a journal, use its style. The parts above stay the same; only the order and punctuation move.
Where to find the citation
You rarely need to build one from scratch. Nearly every repository shows a suggested citation on the dataset's landing page, often with an export button for APA, BibTeX, or RIS so it drops into Zotero or EndNote. Start there, then check three things before you paste it in: that it names a version, that the DOI resolves, and that the year matches the version you actually downloaded.
Reference managers sometimes import datasets as "web page" or "journal article." Change the item type to Dataset, or the [Data set] label and the version field may drop out of the formatted reference.
Versions, subsets, and public data
Versions. Datasets get corrected and extended. If you cite "the survey" without a version, a reader who downloads it next year may get different numbers and assume your analysis is wrong. Cite the version you used. If the repository gives each version its own DOI, use that one, not the "latest version" DOI.
Subsets. If you used a few tables, one wave, or specific variables, cite the whole dataset in the reference list and name the part you used in the text or the data availability statement ("we used the 2019 wave, variables listed in the supplement"). Don't invent a separate citation for a slice unless the repository issued one.
Public data without a DOI. Many government releases don't carry one. The same parts still apply: the agency as creator, the release year, the exact release name (provisional or final, for example), the host, the URL, and the date you accessed it. The IPEDS guide explains why provisional and final releases can give different numbers, which is why the release name belongs in the citation.
Your own data. Deposit your data before the paper goes out, get the DOI, and cite it in your own reference list. It connects the paper and the data in both directions, and it means reuse of your dataset shows up where you can point to it later. The repository guide covers where to deposit.
Mistakes that make a data citation useless
- Only in the methods. Naming the data in the text but leaving it out of the reference list. Keep both: a mention where you use it, a full entry at the end.
- No version. "Downloaded from the repository" with no version or date means the reader can't get what you got.
- Citing the paper instead of the data. Many datasets come with a paper that describes them. If you used the data, cite the data. Cite the paper too if you relied on its methods or findings.
- A landing-page URL copied from the browser. Search-result URLs and session links break. Use the DOI.
- Ignoring the repository's request. Some datasets, especially from ICPSR and government agencies, come with a terms-of-use condition that you cite them in a set way. Check the terms before you publish.
If you support researchers
Data citation is a small topic that makes a good first workshop for a new research data service: it's concrete, it takes under an hour, and it leads naturally into deposit, versioning, and data availability statements. If your library or research office is building that kind of support, the guide to starting research data services describes where to begin, and the DMP checklist covers the planning side. I also help campuses set this up; get in touch if that's useful.
Common questions
Where do I find the citation for a dataset? On the dataset's landing page in the repository. Most show a suggested citation with an export button. Start from that and check it has a version and a DOI.
Should a dataset go in the reference list or just the methods? The reference list, plus a mention in the text. A dataset named only in the methods is invisible to the tools that count citations.
How do I cite a dataset that has no DOI? Use the same parts: creator, year, title, version or access date, publisher, and the most stable URL you can find. For government releases, the release name and access date carry the weight.
Do I cite my own dataset in my own paper? Yes. Deposit it, get the DOI, and list it like any other source.