Data Licensing for Research Datasets

Choosing a licence for a research dataset is a different decision from choosing one for a manuscript or a report. Datasets are frequently harvested, aggregated, mined, and redistributed by machines as well as people, and — unlike most prose — raw facts and measurements often aren’t protected by copyright at all in the first place. That changes which licence actually does useful work, and it’s why repositories, funders, and standards bodies converge on a narrower set of recommendations for data than for text. Two toolkits cover almost every practical case: the Creative Commons suite (CC0, CC BY, and — less often recommended — CC BY-NC) and Open Data Commons (PDDL, the Attribution License, and ODbL), a smaller set of licences built specifically for databases. This guide covers both, plus the question those toolkits exist to answer in the first place: whether a given dataset needs an explicit licence or public domain dedication at all.

What each licence actually requires

A Creative Commons licence is a standard, pre-written set of permissions a rights holder attaches to a work, telling reusers in advance what they may do without asking permission first. There are seven current tools in the Creative Commons suite, but three cover almost every case relevant to research data:

CC0 (“CC Zero”) is not technically a licence — it’s a public domain dedication. Per Creative Commons’ own deed, the rights holder waives copyright and related/neighbouring rights “to the fullest extent permitted by law,” and reusers may copy, modify, distribute, and build on the work, including commercially, without asking permission or giving credit. It does not touch patent, trademark, or privacy/publicity rights.
CC BY (Attribution) permits any use — including commercial use and adaptation — on one condition: the reuser must credit the creator and indicate if changes were made. Per the licence deed, this is the most permissive of the actual (non-waiver) CC licences.
CC BY-NC (Attribution-NonCommercial) adds a restriction on top of attribution: reuse is permitted only for purposes “not primarily intended for or directed towards commercial advantage or monetary compensation,” per Creative Commons’ own definition of noncommercial.

The other four variants — CC BY-SA, CC BY-ND, CC BY-NC-SA, CC BY-NC-ND — add share-alike (adaptations must carry the same licence) or no-derivatives (only the unmodified work may be redistributed) terms. They’re common for articles, images, and educational material, but rare for data, since a share-alike or no-derivatives condition sits awkwardly with the way datasets actually get reused: merged into larger compilations, subsetted, reformatted, and combined with other sources under different terms.

Why “creative commons licensing” for data is a narrower question than for text

In practice, data licensing conversations mostly reduce to a single choice: CC0 or CC BY. CC BY-NC and the no-derivatives/share-alike variants get discussed, but are actively discouraged by most data-focused infrastructure for reasons covered below.

Open Data Commons (ODC): a second toolkit built for databases

Creative Commons isn’t the only licensing suite aimed at data. Open Data Commons, maintained by the Open Knowledge Foundation, publishes three licences drafted specifically for databases rather than adapted from a suite built for creative works:

PDDL (Public Domain Dedication and License) — per Open Data Commons’ own text, PDDL places a database as close to the public domain as the licensor can manage, waiving copyright and database rights and imposing no conditions on reuse. It plays the same role for databases that CC0 plays for content generally.
ODC-BY (Attribution License) — permits copying, distributing, using, and modifying a database and its contents, on condition the original source is credited, per Open Data Commons’ licence text. Functionally close to CC BY, but drafted around database rights specifically rather than copyright in creative works.
ODbL (Open Database License) — a share-alike, copyleft licence: derivative databases must be released under ODbL too. Per the current v1.0 text, this is the one ODC licence with no direct Creative Commons equivalent recommended for data — CC BY-SA exists, but isn’t the convention repositories reach for. OpenStreetMap is the best-known dataset published under ODbL.

In practice, most general-purpose research data repositories default to Creative Commons rather than Open Data Commons — Dryad, Zenodo, and most institutional repositories offer CC0/CC-BY at deposit and don’t list ODC licences as an option. ODC licences see the most use in the geospatial and structured-database communities (OpenStreetMap being the clearest example) and are worth reaching for specifically when a database’s compilation or structure — not just the individual facts inside it — is the thing you most need to licence.

Does a dataset need a licence at all?

Not always — which is exactly why CC0 and PDDL exist as dedications rather than licences. Two separate legal questions determine whether a dataset is protected in the first place, and the answer changes what “licensing” even means for it:

Copyright protects original creative expression. Raw facts and measurements — a temperature reading, a gene sequence, a survey response — are generally not copyrightable on their own in most jurisdictions, including the US; only a sufficiently original selection, arrangement, or presentation of those facts might be. A dataset that’s mostly raw observations may attract little or no copyright protection regardless of what licence is attached to it.
Database rights are a separate, narrower protection that exists in some jurisdictions independent of copyright. The EU’s sui generis database right, for example, can protect a database reflecting “substantial investment” in obtaining, verifying, or presenting its contents, even where the individual facts inside it aren’t copyrightable. The US has no equivalent right. This is precisely why Creative Commons 4.0 and every Open Data Commons licence explicitly address database rights alongside copyright — for data specifically, database rights are often the more relevant protection to waive or license, not copyright.

The practical consequence: a dataset composed entirely of uncopyrightable facts, deposited in a jurisdiction with no applicable database right, may genuinely need no licence at all to be freely reused. But a reuser has no reliable way to determine that from the outside — they would have to independently research the copyright and database-right status in every relevant jurisdiction before reusing anything, which defeats the purpose of sharing data openly. That’s why repositories, funders, and the decision path below converge on attaching an explicit CC0/PDDL dedication or licence regardless of the underlying legal status: it removes the ambiguity for the reuser even where, strictly, a court might find nothing needed waiving in the first place.

Version matters: why 4.0, not 3.0

A large share of search traffic on this topic still looks for “Creative Commons 3.0 licence” — worth addressing directly, because the version number isn’t a formality for data specifically. Creative Commons has released four generations of its licence suite; version 4.0, published in November 2013, is the current, recommended version, per Creative Commons’ own version history. Two 4.0 changes matter directly for research data:

Database rights are explicitly in scope. Earlier versions (including 3.0) required jurisdiction-specific “ports” to address rights like the EU’s sui generis database right, and coverage was inconsistent across ports. The 4.0 suite is unported and international by design, and explicitly licenses applicable database rights alongside copyright — directly relevant to a dataset, where the underlying facts may not be copyrightable but the database structure or a regional database right could still restrict reuse.
Moral rights are addressed uniformly. 4.0 explicitly waives moral rights where legally possible, removing ambiguity that existed under 3.0 and earlier.

3.0-licensed works already published remain validly licensed under those exact 3.0 terms — a licence doesn’t silently upgrade itself. But for anything newly published, including a new dataset deposit, there’s no reason to select 3.0 (or any pre-4.0 port) over the current 4.0 suite.

What repositories and funders actually do

This is where the CC0-vs-CC-BY choice gets resolved in practice, not just in theory:

Dryad requires CC0 for all deposited datasets — it does not accept files under licensing terms incompatible with the CC0 waiver. There’s no CC-BY option for the data itself (software and supplemental files uploaded through Dryad’s Zenodo integration follow different rules).
Zenodo lets depositors choose from a broad list of licences at upload, including CC0 and CC-BY, so the choice is left to the depositor rather than fixed by the platform.
DataCite, which underpins DOI registration for most research data repositories, doesn’t itself mandate a licence, but its own metadata schema and API terms are released under a CC0 waiver — a common pattern where the metadata describing a dataset (title, authors, identifiers) is CC0’d for frictionless harvesting even when the underlying files carry a different licence.
The Open Definition (maintained by the Open Knowledge Foundation) formally lists CC0, CC BY, CC BY-SA, PDDL, ODC-BY, and ODbL as conformant open licences for data and content. CC BY-NC and the ND variants are explicitly not conformant, because the Open Definition’s “no discrimination against fields of endeavour” principle rules out licences that carve out commercial use.

The pattern across all of this: infrastructure built specifically for data defaults to, or requires, CC0 or CC BY — not CC BY-NC, and only rarely ODC.

A real worked example: why CASRAI’s own content is CC-BY, not CC0

CASRAI’s Dictionary and the CRediT Contributor Roles Taxonomy — the standard behind ANSI/NISO Z39.104-2022 — are both released under CC-BY 4.0, as stated on CASRAI’s own licensing policy page: reusers “can copy, redistribute, adapt, translate, and build upon” the content for any purpose, including commercially, provided CASRAI (and, for CRediT specifically, NISO as joint steward) is credited as the source.

That’s a deliberate choice, and it illustrates the CC0-vs-CC-BY decision from the other side. The Dictionary and CRediT aren’t raw measurement data — they’re curated, authored, normative content: definitions, a controlled vocabulary, a named taxonomy with governance behind it. CASRAI has an ongoing interest in that authorship being traceable when a term definition or a CRediT role gets embedded in a publisher’s submission system, a CRIS, or a third-party tool — attribution is how the standard’s provenance stays visible as it propagates. A raw dataset generally has no equivalent need: the value is in frictionless reuse and aggregation, not in the record of who curated any single row, which is exactly why CC0 dominates for that case. Same underlying logic (open licensing), opposite conclusion, because the two categories of content are functionally different — worth keeping in mind rather than defaulting to “whatever CC0 does” for every kind of research output.

Choosing a licence for your own dataset: CC0, CC-BY, or ODC

A practical decision path, consistent with how repositories above actually behave:

Check whether your target repository or funder already decides this for you. If depositing to Dryad, the licence is CC0, full stop. Many funder Data Management Plan (DMP) templates and open-data policies specify a required or preferred licence directly — check the DMP commitment before assuming there’s a free choice.
If there’s a genuine choice, default to CC0 for the raw data itself — this is the position most repository infrastructure and the Open Definition converge on, precisely because facts and measurements are frequently uncopyrightable in the first place, and CC0 removes any residual database-right ambiguity along with the copyright question.
Use CC-BY where attribution genuinely matters to how the output will be reused — curated reference content, a documented methodology, code with authorship you want preserved, or (as above) a standard/taxonomy rather than a raw dataset.
Consider Open Data Commons only if a database-specific copyleft or attribution term is genuinely what you need — ODbL’s share-alike condition mirrors CC BY-SA’s role but is drafted around database rights specifically; reach for it when your repository or community convention expects it (geospatial and structured-database contexts being the clearest case), not by default.
Avoid CC-BY-NC for data for the reasons in the next section, unless a funder or institutional policy specifically requires it.

Whichever licence or dedication is selected, it belongs in the data availability statement and the repository’s own metadata record, not just the DMP — see CASRAI’s guide on writing a data availability statement and on choosing an open data repository for how licensing fits into that broader deposit workflow. For the wider question of choosing between Creative Commons and OSI-approved software licences across data, code, and text together, see CASRAI’s open licensing for research guide.

Why CC-BY-NC is a poor fit for most research data

CC-BY-NC looks like a reasonable middle ground — open, but not for commercial use — and it does get used, particularly by individual authors uncertain about commercial reuse of their work. For research data specifically, three practical problems recur:

“Noncommercial” is genuinely ambiguous in practice. Creative Commons defines it as use “not primarily intended for or directed towards commercial advantage or monetary compensation,” but that line is often unclear for downstream reuse — a university technology-transfer office, a contract research organisation, or a commercial diagnostics lab building on published research data may all have a legitimate claim to be doing noncommercial research, or not, depending on interpretation.
It fails the Open Definition’s field-of-use test, so datasets under CC-BY-NC don’t qualify as “open data” under the definition most funders and repositories actually cite when they say “open.”
It’s incompatible with major open-access and reuse mandates that specify unrestricted reuse, including commercial reuse, as a condition of compliance — the same reasoning that leads most generalist and discipline-specific repositories to either exclude NC licences outright (Dryad) or discourage them in guidance.

Frequently asked questions

What is a Creative Commons licence?

A Creative Commons licence is a standardised, publicly available legal tool that a copyright holder attaches to a work to grant reusers specific permissions in advance — such as the right to copy, redistribute, or adapt it — without requiring individual permission requests. CC0 is technically a public domain waiver rather than a licence, but is issued by the same organisation and used for the same purpose.

What is the Creative Commons licence definition, formally?

Creative Commons describes its licences as tools that work “in addition to copyright” — they don’t replace copyright, they grant permissions the rights holder chooses to give up in advance, using six standard combinations of three conditions (Attribution, ShareAlike, NonCommercial, NoDerivatives), plus CC0 as a separate public-domain dedication.

What is Open Data Commons and how does it differ from Creative Commons?

Open Data Commons is a separate licensing suite — PDDL, the Attribution License (ODC-BY), and the Open Database License (ODbL) — maintained by the Open Knowledge Foundation and drafted specifically around database rights rather than adapted from a suite built for creative works generally. It covers similar conceptual ground to CC0/CC-BY/CC-BY-SA for data, but most general-purpose research data repositories default to Creative Commons; Open Data Commons sees the most use in geospatial and structured-database communities, such as OpenStreetMap’s use of ODbL.

Does my dataset need a licence, or is it already public domain?

It depends on jurisdiction and content. Raw, uncopyrightable facts may already be free to reuse with no licence needed in a jurisdiction with no applicable database right — but there’s no reliable way for a reuser to verify that from outside, which is why attaching an explicit CC0 or PDDL dedication (or a licence such as CC BY) is still the recommended practice even when the underlying legal protection is uncertain or minimal.

Do I need to use a Creative Commons 3.0 licence for older deposits?

No new reason to. If a dataset or work was already published under a 3.0 (or earlier, or jurisdiction-ported) licence, that licence remains valid as published — but for anything newly licensed, use the current 4.0 suite, which is unported, international, and explicitly covers database rights.

Can I change a dataset’s licence after publishing it?

Not retroactively for copies already distributed under the original terms — a Creative Commons or Open Data Commons licence is irrevocable once granted for that specific release. A rights holder can release a later version under a different licence, but earlier recipients keep their rights under the licence they received the work under.

Does CC0 mean I lose credit entirely?

Legally, yes — CC0 imposes no attribution requirement, and neither does PDDL. In practice, scholarly citation norms operate independently of licence terms: researchers are still expected to cite a dataset they reuse under standard citation-ethics norms (see CASRAI’s reuse license entry), even though CC0 doesn’t make that a legal condition of reuse the way CC BY does.

Related CASRAI vocabulary

See also: Licence, Reuse license, Open data, Data Management Plan (DMP), and Data availability statement. For the broader artifact-by-artifact licence decision (data, code, and text together), see Open Licensing for Research.