Assess combinations, not names alone
Names, email addresses and telephone numbers are direct identifiers. Interviews also combine occupation, small organisation, unusual event, location, age and relationships. Each detail may seem innocuous on its own while the cluster makes somebody recognisable to colleagues, a local audience or anyone using public information.
Define likely recipients and the auxiliary knowledge available to them. An internal research team holds different context from readers of an open publication. A transcript may be suitably pseudonymised for controlled analysis yet need stronger transformation for public reuse. Removing a name does not by itself make the content anonymous.
Transform consistently while preserving meaning
Define a controlled placeholder vocabulary such as [PERSON_04], [SMALL LOCATION], [EMPLOYER] and [DATE GENERALISED]. Use one identifier consistently when analysis must follow the same participant or institution. Random replacement names can break relationships or accidentally suggest real people.
Generalisation often preserves more analytic value than deletion. Convert an exact age into a justified band, a town into a region, or a precise job into a functional group. Excessively broad categories can distort accounts of career path or institutional setting. Each rule should state which analytic property is retained.
Free paraphrase is a stronger intervention. Use it when a distinctive quotation remains identifiable after replacements and the research purpose permits a meaning-based account. Record that the passage was paraphrased. A publication must not present transformed prose as an exact quotation from the participant.
Test whole passages for residual identification
Read the output as someone who knows the setting. A role, project, unusual phrase and dated event may identify a person together. Searching distinctive wording may itself disclose sensitive text to an external service, so use only tools and environments permitted by the project’s security and privacy controls.
A second authorised reviewer can inspect selected high-risk passages without automatically receiving the identification key. Record the combination found, the chosen response and any loss of meaning. If a conflict cannot be safely resolved, omit the quotation or narrow its audience rather than claiming anonymity.
Inspect filenames, speaker labels, comments, tracked changes, document properties, captions and export metadata. A visually clean transcript may still reveal names outside the body text. Review the actual dissemination file, not merely the editor view.
Separate originals, keys and working copies
The link between participant and pseudonym is stored separately with strict access and a justified retention period. While that key exists, or identification remains reasonably possible by another route, the material is not automatically anonymous. Use secure transfer and access logging in line with the approved data-management arrangements.
The transformation legend need not contain real identities. Maintain a shareable methods key and a private lookup table as separate objects. Version the transformed corpus. If new contextual information changes risk, revise the affected output and do not assume that an earlier export remains suitable for release.
The Research methods hub links further workflows. The guide to writing a reproducible methods section helps report transformations. Exactly one foundation destination is linked: responsible research foundations.
Approve a specific output for a specific audience
An analysis file, team quotation sheet, thesis, journal article and public data archive reach different audiences. Conduct a separate release review for each purpose. Material suitable inside a small authorised team may be readily attributable when published. Avoid one blanket approval for every future use.
- Direct and combinable identifiers are inventoried.
- Transformations follow a versioned key.
- Loss of analytic meaning has been reviewed.
- Metadata and tracked changes have been inspected.
- The lookup key and working corpus are separate.
- The intended recipients are part of risk assessment.
- Unsafe passages are restricted or omitted.
Describe categories and quality controls in the methods report without reproducing sensitive original values. Do not claim full anonymity when the process only assigned pseudonyms or meaningful combinations remain. Accurate terminology is itself a safeguard.