How to Cite a Dataset (APA 7, MLA 9 and the Rules That Change)

The APA 7 anatomy, element by element

APA 7 formats a dataset like a piece of software, and the template runs: Author. (Year). Title of data set (Version 1.2) [Data set]. Publisher. https://doi.org/xxxx. The bracketed description is the piece students miss - [Data set] tells the reader the source is data rather than prose, and it replaces the [Video] and [PowerPoint slides] brackets the genre-citation pages use. The version number goes in parentheses right after the title when the archive publishes one, because a 2019 dataset revised in 2023 is a different citation depending on which files you downloaded. The how-to-cite-a-pdf page uses the same bracket-description logic for government reports. The DOI belongs to the archive deposit, which is the object that stays stable while platforms change.

The author question resolves by production, not convenience: cite whoever produced the data - the principal investigator, the research team, or the agency - not the platform that hosts it. ICPSR deposits cite the depositing investigator; World Bank indicators cite the World Bank as both author and publisher. When no named author exists, the title moves to the author position, exactly as it does for articles. If the dataset carries both a paper and an archive deposit, the dataset citation points at the archive DOI, because that is the object you actually analyzed - the paper cites the findings, the dataset cites the files. Splitting the two is the reproducibility habit instructors are grading when they ask for data citations at all.

The numbers at a glance

Dataset citation anatomy by style — the range, and where each row actually shows up in coursework. Skim it first; the sections below explain why the range exists and which row applies to you.

The MLA 9 anatomy and the container logic

MLA 9 runs datasets through its container model: the dataset title sits in quotation marks, and the repository that publishes it is the italicized container. The template runs: Author. Title of Dataset. Repository Name, Publisher, Year, DOI or URL. A World Bank example: World Bank. Gross Domestic Product, Current Prices, 1960-2024. World Development Indicators, World Bank Group, 2025, databank.worldbank.org/source/world-development-indicators. The container principle is the same one the article-titles page applies - the inner work quotes, the outer work italics - and datasets just inherit it without a new rule. A dataset is simply an inner work whose container is a repository, and once that mapping clicks, every MLA habit a writer already has transfers to data without a second rulebook.

Two MLA details change for data. The access date appears only when the dataset has no dated version, because unversioned live databases change under the reader; a versioned download is a stable object and needs no Accessed line. And the URL or DOI ends without a period, because trailing periods corrupt live links - the same link-hygiene rule the PDF-citation page enforces. When the dataset sits inside a larger archive with its own publisher, list repository then publisher separately, exactly as a streamed film lists the platform and the studio. The cite-what-you-read rule binds here too: cite the repository page you downloaded from, not a citation copied from the paper's footnote. Copied citations drift out of date the moment the archive revises a deposit, which is why the downloaded page is the authority.

Versions, access dates and the living-data problem

Datasets are the one source class that legitimately changes after publication, which is why version discipline matters more than for any other genre. Three scenarios cover most cases: a versioned archive deposit (cite the version number and the deposit DOI - stable forever); an unversioned continuously-updated database (cite the retrieval year in APA, add an access date in MLA, because yesterday's download is not today's); and a dataset accompanying a published paper (cite the dataset separately with its archive DOI, and cite the paper separately - they are two sources). When an instructor asks for reproducibility, the dataset citation plus version plus download date is the minimal standard.

Authorship gets one extra check for data: some archives require a suggested citation printed on the dataset page, and when it exists, use it - the archive has already resolved producer, version and format questions. When it does not, construct the entry from the elements above and verify against the archive's own examples, which APA and MLA both maintain. The verification habit from the citation-generator page applies with one addition: after formatting, click your own DOI and confirm it resolves to the exact files you analyzed. A dataset citation that resolves elsewhere is a broken citation in a way prose citations never are. A two-minute read of the archive's own suggested citation resolves producer, version and format questions before the entry is even formatted.

In-text citation and the reference-list reality check

In-text forms stay ordinary: APA uses (World Bank, 2025) or (Klein & Okafor, 2023) exactly as with any author-date source, and MLA uses (World Bank) with the Works Cited entry doing the year work. What changes is where the citation sits in the sentence: data claims cite at the point of the number, not the paragraph end, because a figure without its source attached reads as invented. Tables and figures built from the data add a note under the table - Data from the World Bank (2025) - which the APA table-format rules treat as part of the table itself. The note keeps the table self-contained when it is screenshotted into a presentation, which happens more often than any instructor admits.

The reference-list audit for datasets has three checks. First, every dataset entry's DOI or URL resolves - click them, because link rot is the most common dataset-citation failure in submitted work. Second, versions match: the version in the entry is the version in your analysis folder, not the archive's newest release. Third, the alphabetizer holds: datasets file under their producer exactly like articles file under authors, and the reference-alphabetizer orders mixed lists of articles, books and datasets in one paste. Ten minutes on the three checks is the difference between a reproducible data citation and one a grader cannot trace. Reproducibility is the grading standard hiding under every dataset assignment, and the three checks are its minimum viable version.

Three minutes to a clean citation

Edge-case citations punish hesitation, not ignorance: the student who spends twenty minutes googling a no-author format loses more time to doubt than the format ever costs. Decide the pattern once — title moves forward, organization counts as author, n.d. when no date exists anywhere — and apply it mechanically. Consistency across your list matters more to graders than perfect recall of manual arcana.

Automate what's mechanical. The citation generator below formats journal articles, books, chapters and websites in all three major styles, including the awkward cases on this page. Paste the metadata, copy the output, done — the minutes you save belong to analysis, which is where rubric points actually live. Cross-check in-text forms against the table above and the reference list is submission-ready.