“Data publication practices” refers to the set of choices and steps involved in formally publishing a research dataset as a standalone, citable scholarly output — distinct from simply sharing a file with a collaborator, depositing it in a repository to satisfy a funder mandate, or governing who may access it. This guide covers what actually makes a dataset “published,” the two main routes for doing it (repository deposit with a DOI versus the data-journal / data-paper route), how to choose between them, and the metadata, licensing, versioning, and review practices that a properly published dataset needs.
Data Publication vs. Data Sharing vs. Data Deposit
These three terms get used loosely, but they describe different levels of formality:
- Sharing is the broadest term — sending a file to a named collaborator, posting it on a lab website, or attaching it to an email. It carries no persistent identifier and no guarantee the data will still be there, or the same, next year.
- Deposit means placing the dataset in a repository, often specifically to satisfy a funder or journal data-availability requirement. A deposit becomes a publication once the repository assigns it a persistent identifier and a public, citable metadata record.
- Publication is the formal end state: the dataset has a persistent identifier (almost always a DOI), a public metadata record sufficient to find and cite it independently of the paper it supports, and, in the case of a data paper, has been through editorial or peer review.
CASRAI’s dictionary defines the related concepts of published data and dataset as a research output in more detail. The practical test for whether something counts as “published” rather than merely “shared” is citability: can another researcher cite this dataset, on its own, in a reference list, the way they would cite a journal article?
The Two Main Routes to Publish a Dataset
There are two broadly distinct ways to move a dataset from “deposited” to “published”:
1. Repository deposit with a DOI (the data-only route)
The dataset is deposited in a general-purpose repository (e.g. Zenodo, Dryad, Figshare, an institutional repository) or a domain repository (e.g. GenBank, ICPSR, PANGAEA), which mints a DOI — typically through DataCite — and publishes a structured metadata record alongside the files. Most repositories perform curation checks (file integrity, completeness of required metadata fields) rather than substantive peer review of the data itself. This is the route most funder data-management-plan (DMP) commitments describe, and it is usually faster than the data-paper route.
2. The data journal / data paper route (the peer-reviewed route)
A data paper is a short, structured manuscript describing a dataset — how it was collected, processed, and structured, and its potential reuse value — submitted to a journal that specializes in this format, or to a “data paper” track of a subject journal. Established data journals include Scientific Data (Springer Nature), Data in Brief (Elsevier), Earth System Science Data (Copernicus/EGU), and GigaScience (Oxford University Press/BGI, which also runs the GigaDB repository). The manuscript goes through the journal’s editorial and peer-review process, but reviewers evaluate data quality, completeness, and documentation rather than the novelty of a scientific finding — there is no “result” to be original in the way a research article has one. The dataset itself is still deposited in a repository (often specified by the journal) and linked to the data paper by DOI in both directions.
The two routes are not mutually exclusive: a dataset can be deposited in a repository for its DOI and citability, and separately described in a data paper if the authors want the added discoverability, structured peer review, and journal-indexed citation record that a data paper provides. See CASRAI’s guide on how to cite data for how citation formats differ across the two.
Choosing a Publication Venue
The choice between a bare repository deposit and a data-paper submission (and, within that, which repository or journal) generally comes down to:
- Funder or journal mandate. Many funder DMP policies specify or recommend a particular repository, or require deposit in a domain-appropriate repository where one exists (e.g. genomic data to GenBank, not a generalist repository). Check the mandate before choosing.
- Domain fit. A domain repository (built around a specific discipline’s data types and metadata standards) is usually preferable to a generalist repository when one exists for the field, because it surfaces the dataset to the researchers most likely to reuse it and applies discipline-appropriate curation.
- Certification and longevity. Repositories certified against a recognized standard such as CoreTrustSeal, or listed in the re3data registry, offer stronger guarantees that the data will remain accessible over time — a material consideration since a published dataset’s DOI is a long-term promise, not a one-time upload.
- Whether a narrative description adds value. If the dataset needs substantial context to be reusable — unusual collection methods, a large multi-part structure, domain-specific caveats — a data paper is often worth the additional effort over a bare repository record with only structured metadata.
Metadata and Persistent Identifiers: What Publication Actually Requires
A DOI alone does not make a dataset discoverable or reusable; the metadata record attached to it does the real work. DataCite’s Metadata Schema defines the properties a repository must supply to mint a DataCite DOI: Identifier, Creator, Title, Publisher, PublicationYear, and ResourceType are mandatory, with Subject, Contributor, Date, RelatedIdentifier, and others recommended. CASRAI’s guide to descriptive metadata for datasets covers these fields in detail, and the FAIR checklist guide walks through the broader Findable/Accessible/Interoperable/Reusable requirements a published dataset should meet, per the FAIR principles (Wilkinson et al., 2016, Scientific Data).
Citation formatting for published data should follow the Force11 Joint Declaration of Data Citation Principles, which establishes that data citations should be treated with the same importance as citations of other research objects, include a persistent identifier, and credit all contributors. CASRAI’s guide to citing data covers the practical formats.
Licensing at Publication Time
Publishing a dataset means making an explicit, machine-readable statement of what others may do with it — this is usually done through a license (a Creative Commons license, or a public-domain dedication such as CC0, are the most common choices for research data) applied at the point of deposit. Getting this step right matters as much as the DOI itself: a dataset with a DOI but no clear license is technically accessible but not confidently reusable, since downstream users cannot be sure what’s permitted. See CASRAI’s guide on describing reuse rights and permissions for a shared dataset for how to choose and document a license.
Versioning a Published Dataset
Unlike a journal article, a published dataset is often not a one-time, final artifact — errors get corrected, new observations get added, or a superseding processing method gets applied. Good data publication practice treats each substantive change as a new version with its own identifier, while keeping the earlier version resolvable rather than silently overwriting it. Most major repositories (Zenodo and Dryad are common examples) support DOI versioning natively: a “concept DOI” resolves to whichever version is current, while each individual version keeps its own DOI so that a paper citing a specific version continues to resolve to exactly the data that was analyzed. When choosing a repository, checking whether it supports versioning this way is worth doing before, not after, the first deposit.
Peer Review of Datasets: What Reviewers Actually Check
Where a dataset goes through formal review — as part of a data-paper submission, or through a repository’s curation process — the review criteria differ from manuscript peer review. Reviewers and curators typically check: whether the files are complete, readable, and match their documentation; whether metadata is sufficient to interpret the data without the original authors’ help; whether the collection and processing methods are adequately described; and whether any necessary de-identification, consent, or licensing conditions have been met. They generally do not evaluate whether the underlying research question was novel or important — that judgment belongs to the article(s) that use the data, not the dataset itself. CASRAI’s guide on data dictionaries in research data management covers one of the documentation artifacts curators most often check for.
How Data Publication Fits into the Wider Research Data Lifecycle
Data publication is a specific, late-stage step within the broader research data lifecycle, not a substitute for the governance and planning work that precedes it. CASRAI’s guide to research data governance covers the roles and policies that determine who is authorized to publish which data and under what conditions, and a project’s data management plan should specify, before the project begins, whether and how its outputs will eventually be published. Where a dataset contains personal or sensitive information, data anonymisation or another form of de-identification is typically a prerequisite for publication, not a step that can be skipped by relying on access controls alone. Not every dataset needs a narrative data-availability statement in its home journal article once it is formally published with its own DOI, but many journals still require one; see CASRAI’s guide on writing a data availability statement for how the two relate.
Frequently Asked Questions
What’s the difference between data sharing and data publication?
Sharing is informal and ad hoc — sending a file to a named person, with no persistent identifier or guarantee of continued access. Publication is formal: the dataset has a persistent identifier (typically a DOI), a public and citable metadata record, and, in the case of a data paper, has been through editorial or peer review.
Do I need a DOI to “publish” a dataset?
In practice, yes. A DOI (or another persistent identifier from a recognized scheme) is what makes a dataset independently citable and is what most funders, journals, and repositories treat as the marker of formal publication rather than informal sharing.
What is a data paper, and do I need to write one?
A data paper is a short, peer-reviewed manuscript describing a dataset’s collection methods, structure, and reuse potential, published in a data journal such as Scientific Data or Data in Brief. It is optional in most cases — a repository deposit with a DOI already counts as publication — but a data paper adds structured peer review, journal indexing, and a citable narrative description, which can be worth the extra effort for complex or high-reuse-value datasets.
Are published datasets peer reviewed?
Only when submitted through a data-paper route to a journal that peer-reviews that format. A bare repository deposit typically goes through curation checks (completeness, file integrity, metadata adequacy) rather than substantive peer review.
Can a published dataset be updated after its DOI is minted?
Yes, through versioning. Most major repositories support assigning a new DOI to each substantive new version while keeping earlier versions resolvable, so that citations to a specific version remain accurate even after the dataset is updated.
