Wonders

The Rabbit Hole · Citation guides

How to Cite a Dataset

Cite a dataset in APA 7, MLA 9, Chicago 18, Harvard and IEEE, with the version number and the DataCite DOI, using a real ICPSR study as the worked example.

Sep 22, 2026
How to Cite a Dataset

TL;DR

Cite the data, not the paper that used it. A dataset reference needs the investigator or agency as author, the year the version you used was released, the title, the version or edition, a bracketed [Data set] description, the repository, and the DOI. The version matters: a DOI without a version suffix usually resolves to whatever is current, which may not be the data you analysed.

A dataset is a citable work with an author, a date, a title and a persistent identifier, and the major styles all treat it as one. What makes data different from an article is the version: the file you downloaded in March may not be the file the same link serves in December. Every format below therefore has somewhere to put a version or edition number, and leaving it out is the single most common mistake in a data reference.

What to Collect

The worked example throughout is a real study: the Survey of Prison Inmates, United States, 2016, produced by the United States Bureau of Justice Statistics and distributed by the Inter-university Consortium for Political and Social Research as study 37692, version 1, released in 2020, DOI 10.3886/ICPSR37692.v1. Every element was checked against the DataCite record for that DOI on 22 September 2026.

Quick Reference by Major Style

APA (7th Edition)

Author, A. A. (Year). Title of the data set (Version 1) [Data set]. Publisher. https://doi.org/xxxx

United States Bureau of Justice Statistics. (2020). Survey of prison inmates, United States, 2016 (ICPSR 37692; Version V1) [Data set]. Inter-university Consortium for Political and Social Research. https://doi.org/10.3886/ICPSR37692.v1

In-text: (United States Bureau of Justice Statistics, 2020)

The bracketed [Data set] description sits after the title and any version, before the period. APA's own Common Reference Examples handout shows the same shape with a different ICPSR study; note that in the copy of that handout downloaded on 22 September 2026 the parenthetical study number (ICPSR 21600) does not match the DOI in the same reference (10.3886/ICPSR36966.v1, which is study 36966). That is a useful reminder rather than a rule: check the study number against the DOI yourself, because reference managers copy whichever one the record happens to carry.

For a dataset you produced and did not deposit anywhere:

Author, A. A. (Year). Title [Unpublished raw data]. Institution.

MLA (9th Edition)

MLA has no dataset template of its own. You build the entry from the template of core elements: the producer goes in the Author slot, the dataset title in Title of Source, the repository in Title of Container, and the version, publication date and DOI in the version, date and location slots.

Producer. Title of Dataset. Version, Repository, Year, DOI or URL.

United States, Bureau of Justice Statistics. Survey of Prison Inmates, United States, 2016. Version V1, Inter-university Consortium for Political and Social Research, 2020, doi.org/10.3886/ICPSR37692.v1.

In-text: (United States, Bureau of Justice Statistics)

MLA's rule for organisational authors, stated on the MLA Style Center and quoting the MLA Handbook, is that a work published by the organisation that wrote it begins with the title and lists the organisation only as publisher, but that a division or committee of a larger body is listed as the author with the parent body as publisher. (MLA flags that post as eighth-edition content, so check it against the ninth edition.) Here the producing agency and the distributing archive are different, so the agency stays in the author slot. Whatever begins the entry is what the in-text citation must key to.

Chicago (18th Edition)

The Chicago Manual of Style is subscription-only and its free Citation Quick Guide does not cover datasets, so the form below is the one published by Murdoch University Library in its Chicago 18th edition author-date guide, checked on 22 September 2026. Murdoch puts the words "Data from" before the title and the title in quotation marks.

Author, A. A., and B. B. Author. Year. Data from "Title of Dataset". Publisher or Source, Month Day. doi.

United States Bureau of Justice Statistics. 2020. Data from "Survey of Prison Inmates, United States, 2016". Inter-university Consortium for Political and Social Research. https://doi.org/10.3886/ICPSR37692.v1.

In-text: (United States Bureau of Justice Statistics 2020)

Murdoch's Chicago notes and bibliography guide gives the other system's form, with the version and the source repository spelled out:

Author, A. A., and B. B. Author. "Title of Dataset." Version. Publisher, Month Day, Year. Source. doi.

United States Bureau of Justice Statistics. "Survey of Prison Inmates, United States, 2016." Version V1. Inter-university Consortium for Political and Social Research, 2020. https://doi.org/10.3886/ICPSR37692.v1.

That guide also advises naming the type of source — "dataset" — whenever it is not already clear from the rest of the entry, and adding an access date when the repository is updated regularly and carries no revision date.

Harvard (Cite Them Right)

Cite Them Right is a paywalled manual, so the formats below follow two named library guides that reproduce it: the University of Sheffield Library's Harvard guide, which states that it is based on Cite Them Right (10th rev. and expanded edn), and the University of the West of England's Cite Them Right Harvard guidance, which became UWE's default style in September 2026. Both were checked on 22 September 2026. Sheffield puts a dataset title in single quotation marks and puts the version before the availability statement; one of its own examples is Office for National Statistics, Social Survey Division (2019) 'Annual Population Survey, April 2015-2016'. 6th edn. SN: 8003. doi: http://doi.org/10.5255/UKDA-SN-8003-6.

Author Surname, INITIAL(S). OR Organisation (Year) 'Title of dataset'. Edition (if necessary). Number or Version of dataset (if necessary). Available at: URL (Accessed: date) or doi:

United States Bureau of Justice Statistics (2020) 'Survey of prison inmates, United States, 2016'. Version V1. doi: https://doi.org/10.3886/ICPSR37692.v1

In-text: (United States Bureau of Justice Statistics, 2020)

IEEE

The IEEE Reference Guide gives datasets their own section and three variants, depending on whether you have a DOI, a DOI resolver link or only a web address. Note the unusual order: the date comes second, immediately after the author.

[1] Author, Date, "Title of Dataset," Source, doi: xxx.

[1] United States Bureau of Justice Statistics, 2020, "Survey of Prison Inmates, United States, 2016," Inter-university Consortium for Political and Social Research, doi: 10.3886/ICPSR37692.v1.

In-text: [1]

The guide's own worked example is a different ICPSR study and ends with the DOI in exactly this form. Where there is no DOI, it uses [Online]. Available: http://… instead.

Vancouver and AMA

Health-sciences data are cited often enough that the numbered styles have a settled shape for them. The format below is the one the University of Queensland Library publishes in its Vancouver/AMA guide, checked on 22 September 2026: the phrase "Data from:" before the title, the repository in place of a journal, and the deposit date.

  1. Author AA, Author BB. Data from: Title of dataset. Title of Website. Deposited Month DD, YYYY. DOI
  1. United States Bureau of Justice Statistics. Data from: Survey of Prison Inmates, United States, 2016. Inter-university Consortium for Political and Social Research. Deposited 2020. doi:10.3886/ICPSR37692.v1

Where there is no DOI, the same guide adds an access date before the URL. Journal house styles vary here more than in any other reference type, so check the instructions to authors before submission, and see the Vancouver guide for the numbering rules.

Missing Elements

No version number. Give the release or deposit date instead, and say in your methods when you downloaded the file.

No named investigator. Use the producing organisation as author. If the repository is also the producer, it appears once as author and the reference does not repeat it as publisher.

No DOI. Use the landing-page URL, not a direct download link to a file, and add an access date in Harvard and MLA. A direct file link breaks when the repository reorganises; the landing page usually survives.

No date at all. APA uses (n.d.); IEEE uses "(n.d.)" in the date position; Harvard uses (no date).

A dataset inside a supplement. If the data are a supplementary file attached to an article and have no separate identifier, cite the article and name the supplement in your text.

Common Mistakes

Sources for These Formats

Full style rules: APA, MLA, Chicago, Harvard, IEEE, Vancouver.

Wonders has no index of its own: it searches OpenAlex and Semantic Scholar, where the papers that analyse a dataset are indexed, and exports what you save as BibTeX, RIS or CSV for your reference manager. The dataset record itself belongs in the repository, and that is where a reference to it should point.

Frequently asked questions

Do I cite the dataset or the article that describes it?

If you analysed the data, cite the dataset. If you are reporting what the authors found, cite their article. Many projects publish both and ask you to cite both; the repository's landing page usually says which it prefers, and the polite thing to do is follow it.

Why does the version number matter so much?

Repositories revise data. ICPSR, Dryad, Zenodo and figshare mint a version-specific DOI as well as a concept DOI that always resolves to the newest release. If you cite only the concept DOI, a reader following it next year may see different numbers from the ones in your table. Cite the version you actually downloaded.

What if the dataset has no DOI?

Give the URL of the landing page and, in Harvard and MLA, an access date. Government statistics portals and many national archives issue stable URLs rather than DOIs; that is acceptable as long as the page identifies the data and its date.

How do I reference my own unpublished data?

APA treats it as an unpublished raw dataset: author, year, an italic title or a bracketed description if it is untitled, then [Unpublished raw data] and the institution. It stays in the reference list only if a reader could in principle request it; otherwise describe it in the text.

Who is the author of a government dataset?

The agency that produced the data, not the archive that distributes it. In the ICPSR example below the author is the Bureau of Justice Statistics and the publisher is ICPSR. Repository and producer are different roles and both belong in the reference.

More useful guides

All citation guides

Finding these guides useful?

Try these techniques in Wonders — an AI workspace for literature review. 14 days free. Students get 50% off.

Start free →