Digitally Anonymised Data

“Digitally anonymised data” is a term used loosely across IT and consumer-privacy writing, but in a research data management context it has a precise, testable meaning: an electronic dataset that has been processed with computational techniques to the point that no individual can reasonably be re-identified from it, with no key or additional information retained anywhere that could reverse that process. The word “digitally” describes the mechanism — software-driven masking, generalisation, aggregation, noise addition, or synthetic-data generation applied to an electronic source dataset — not a separate or weaker standard than anonymisation achieved any other way.

Why the word “digitally” matters here

The distinction researchers actually need is not between “digital” and “non-digital” anonymisation — almost all research data anonymisation today happens on electronic datasets using software. The distinction that matters is between a dataset that has genuinely cleared the anonymisation bar and one that has only had some identifiers removed or altered digitally without a rigorous disclosure-risk assessment. Simply deleting a name column or hashing an ID in a spreadsheet is a digital operation, but it does not by itself produce digitally anonymised data unless the remaining fields, taken together with what a realistic adversary could plausibly obtain, no longer make re-identification reasonably likely. That test comes directly from data protection law, not from the mechanics of the tooling used.

The legal test

Under the EU/UK General Data Protection Regulation, Recital 26 sets the operative standard: information is anonymous, and therefore falls outside GDPR’s scope entirely, only when the data subject is “not or no longer identifiable”, with identifiability assessed against “all the means reasonably likely to be used” by the controller or a third party, taking into account cost, time, and available technology at the time of processing. That is a substantive, risk-based test, not a checklist of which fields were removed. A dataset can have every direct identifier stripped digitally and still fail this test if indirect identifiers make re-identification realistically feasible (this is sometimes called the “mosaic effect” or linkage risk).

Two related but distinct concepts sit either side of digitally anonymised data:

• Pseudonymisation (GDPR Article 4(5)) replaces identifiers with a token or key, but the re-identification key is retained, so the output remains personal data under GDPR Recital 26 — even if the key is encrypted, access-controlled, or held by a separate party. Article 89(1) names pseudonymisation as an accepted research safeguard, and states a preference for full anonymisation over pseudonymisation where the research purpose can still be met that way.
• De-identification is a broader, less legally precise term (more common in US contexts, e.g. HIPAA’s Safe Harbor and Expert Determination methods) for removing or altering identifiers; not every de-identified dataset meets the GDPR anonymisation bar, and not every anonymised dataset was produced via a formal de-identification method.

What actually makes a dataset qualify

In practice, a dataset earns the label “digitally anonymised” only when three conditions hold together: (1) a recognised anonymisation technique was applied by software to the electronic dataset (aggregation, generalisation, suppression, k-anonymity, differential privacy, or synthetic-data generation being the common families); (2) no re-identification key, lookup table, or linkage path was retained by anyone involved, including the original data holder; and (3) a disclosure-risk assessment — not just an assumption — supports the conclusion that re-identification is not reasonably likely given realistically available auxiliary data. NIST Special Publication 800-188 (De-Identifying Government Datasets) is a widely cited framework for structuring that disclosure-risk assessment, covering direct versus indirect/quasi-identifiers and methodology for testing re-identification risk, and is a useful reference distinct from GDPR’s own pseudonymisation/anonymisation distinction.

How this differs from the CASRAI anonymisation guide

This page defines the term itself — what has to be true for a specific dataset to correctly be called digitally anonymised data. For the practical methodology — the actual anonymisation techniques, the GDPR/HIPAA/NIST standards that govern when anonymisation is required, and when anonymisation is the right control versus pseudonymisation or controlled access — see the full guide, Data Anonymisation in Research: Techniques, Standards, and When It’s Required.