Direct comparison
Fair Data Vs Open Data: Key Differences & Comparison | CASRAI
FAIR data is machine-actionable and reusable under defined conditions; open data is legally available to all. All open data can be FAIR, but FAIR data is not necessarily open.
Side-by-side comparison
| Dimension | FAIR data | Open data |
|---|---|---|
| Definition origin | Wilkinson et al. (2016), Scientific Data, doi:10.1038/sdata.2016.18; 15 guiding principles under 4 headings | Open Knowledge Foundation's Open Definition (opendefinition.org); also Open Data Institute (ODI) and related frameworks |
| Core requirement | Data and metadata must be Findable (with a PID and rich metadata), Accessible (conditions machine-actionable, data retrievable via standardised protocol), Interoperable (uses shared vocabularies and formats), and Reusable (clear licence and provenance) | Data must be legally and technically accessible and reusable by anyone without discrimination; maximum restriction is attribution (BY); NC and ND conditions are generally incompatible with the Open Definition |
| Does it require publicly free access? | No. FAIR's Accessible principle requires that access conditions are documented and machine-actionable — not that access is free or unrestricted. Authentication-gated or embargoed data can be fully FAIR. | Yes. Open data must be available for anyone to use, reuse, and redistribute without restriction beyond attribution. Paywalled or access-controlled data is not open data. |
| Applies to metadata | Yes — FAIR requires that metadata is persistent and accessible even if the data itself is not (e.g., "tombstone" records after embargo or withdrawal) | Open data frameworks focus on the data itself; open metadata is a separate concept (e.g., open bibliographic metadata) |
| Machine-actionability | Central requirement. FAIR emphasises that both data and metadata should be understandable and processable by machines without human intervention — using PIDs, community standards, and controlled vocabularies | Not a core requirement. Open data must be machine-readable (i.e., in a processable format rather than a PDF scan) but the Open Definition does not specify machine-actionability standards |
| Sensitive data (health, personal, indigenous) | FAIR is explicitly designed to work with sensitive data. The CARE principles (for indigenous data) are complementary to FAIR and address sovereignty and ethical dimensions that FAIR does not cover. | Personal and sensitive data cannot typically be open data without anonymisation, due to GDPR and ethical obligations. Fully open sensitive data is rarely appropriate. |
| Funder requirements | Horizon Europe, UKRI, Wellcome Trust, and NIH all require data to be as FAIR as possible. UKRI's Research Data Policy mandates FAIR data principles. | Some funders require open data for specific output types (e.g., genomic sequence data, clinical trial data). Many use "as open as possible, as closed as necessary" — not unconditional open data. |
Common questions
FAQ
Does making data FAIR mean making it open?+
No. FAIR's Accessible principle requires that the access conditions for data are machine-readable and clearly documented — but those conditions can include controlled access, authentication, embargo periods, or data-sharing agreements. A health dataset with patient records can be FAIR (discoverable, accessible under a documented data-sharing agreement, interoperable, and reusable under defined conditions) without being publicly open. The phrase "as open as possible, as closed as necessary" captures the current funder consensus: FAIR first, open where appropriate.
Can personal data be FAIR?+
Yes. Personal and sensitive data can be FAIR if: the data has a persistent identifier and rich metadata (Findable); the access conditions — including the data-sharing agreement or ethics approval required — are machine-actionable and documented (Accessible); the data uses standard formats and vocabularies (Interoperable); and it has a clear licence or conditions for reuse by authorised parties (Reusable). The metadata record is often publicly available even when the data itself is access-controlled. UK Biobank and the UKRI-funded health data research infrastructure are examples of access-controlled datasets that adhere to FAIR principles.
Do funders require open data or FAIR data?+
Most major funders require FAIR data and encourage open data where appropriate. Horizon Europe requires data to be "as open as possible, as closed as necessary" and explicitly endorses FAIR. UKRI's Research Data Policy (updated 2023) requires that UKRI-funded research data is FAIR-aligned and made openly available unless there are legitimate restrictions (legal, ethical, commercial, or practical). Wellcome Trust requires open access to research data supporting publications. Few funders require unconditional open data for all data types; most accept that sensitive, personal, or commercially sensitive data should be access-controlled.
Is open data always FAIR?+
No. Open data can be published without meeting FAIR requirements. For example, a CSV file deposited on a local institutional website with no PID, no metadata beyond a file name, and no licence statement is technically open (anyone can download it) but not FAIR — it lacks a persistent identifier, rich metadata, a community-standard format declaration, and a reuse licence. Making open data FAIR requires adding persistent identifiers, structured metadata using community standards, and clear reuse conditions.
Going deeper
