An organization wants to share the results of a survey. Before publishing the table, it removes respondents' names. Their ages, locations, and answers remain. Can the organization now call the data anonymous?
Not necessarily. Someone who knows a person of an unusual age in a small community might recognize their record. A combination of details can reveal an identity even when no column contains a name.
Data anonymization aims to transform information so that the people it concerns are no longer identifiable. Understanding it requires looking at both the data retained and the possibility of linking them to other information.
What does it mean to anonymize data?
Anonymization aims to make information usable without allowing people to be identified from it. The aim might be to study trends, publish statistics, or conduct research while preserving the privacy of the people who supplied the data.
The outcome matters more than the name of the operation. Deleting a column or using a tool does not, by itself, prove that anonymity has been achieved. The Spanish Data Protection Agency explains that anonymized information should present no reasonable likelihood of identifying the person.
This avoids two misunderstandings: assuming that any table without names is anonymous, and demanding a promise of safety against every imaginable scenario. The question is whether there is a reasonable possibility of identification in the context being considered.
From an individual liberty perspective, that caution makes sense: taking part in a survey should not automatically expose someone to companies, authorities, or other parties. Personal data protection helps explain the concern for individual autonomy.
Key idea: Removing a name eliminates an obvious identifier. Anonymization also requires assessing what the remaining information can reveal.
Anonymization and pseudonymization: the role of codes
Return to the survey. Instead of simply deleting names, the organization replaces them with codes: each participant appears as a numbered record. It keeps the correspondence between codes and names in a separate file with restricted access.
This illustrates pseudonymization. Attributing the data to a person requires additional information, which must be kept separately and protected. It allows people to work with records while reducing direct exposure of identities, but the organization retains a way to make the link.
The European Data Protection Board distinguishes the two concepts: pseudonymized data do not thereby become anonymous. In the example, they remain personal data for the organization that holds the correspondence.
Nor does deleting the correspondence file settle the question. Age, location, or other attributes might allow someone to be identified by another route. The resulting dataset must be examined as a whole.
Encryption protects access to information, but it does not by itself establish that the information is anonymous. Each measure serves a different purpose.
How an identity can reappear
A full name or an identification number is a direct identifier. Other details act as indirect identifiers: they can help identify a person when combined with one another or with outside information.
In the hypothetical survey, an exact age may reveal little on its own. Combined with a location and an uncommon occupation, it could point to a particular person. If that person has published their occupation and place of residence on an accessible page, a possible link between the two sources emerges.
Re-identification means linking information that appears disconnected from an identity back to a person. It does not always require recovering a secret key: it can result from combining attributes.
The basic anonymization guide from Singapore's data protection authority, published in Spanish by the Spanish Data Protection Agency, explores the difference between removing identifiers and assessing the risk that remains. It is useful here as a technical and educational resource; its legal references should not automatically be applied to other countries.
Three transformations and what they leave out
The organization can modify the survey in different ways. These three operations help explain the process, although none guarantees anonymity on its own:
- Suppression: removing information, such as a name or a particularly revealing attribute. This reduces exposure but also removes possibilities for analysis.
- Generalization: replacing a precise value with a broader one. An exact age becomes an age range, or a locality is replaced by a larger area.
- Aggregation: replacing individual records with summaries, such as the total number of responses in each age group.
Generalization and aggregation are different operations. With generalization, there may still be one row per participant, albeit with less detail. Aggregation presents combined results.
The guide describes these techniques and their limits. A summary of a very small group can still reveal information about its members. Publishing an average or a total therefore does not automatically mean the result is safe.
In the example, publishing results only for large areas might be enough to identify a general trend. If the research question requires comparisons between localities, that transformation could conceal important differences. Usefulness depends on what the researcher wants to find out.
Key idea: A transformation should be judged by the risk that remains and the analysis it still permits. Reducing detail can lower risk, but it can also limit the usefulness of the result.
Sharing a table is different from allowing queries
The way people access data changes the assessment. Publishing a file for open download and allowing analysis inside a protected environment give users different opportunities to work with the information.
The technical publication NIST SP 800-188 recommends that US public agencies assess their objectives, disclosure risks, and access model. Among the options it considers are publishing transformed data and making them available for use in protected environments.
For the survey, this raises a concrete decision: is it necessary to share every record, or are the results of certain queries enough? Restricting access can be part of the protection, but it does not by itself make data anonymous.
As a practical editorial rule, the assessment should be revisited when the information available or the conditions of disclosure change. A label applied when a file is created should not replace that examination.
What anonymity means under the European framework
Recital 26 of the European Union's General Data Protection Regulation states that its principles do not apply to anonymous information. In assessing identifiability, the means reasonably likely to be used by the controller or another person must be considered, including costs, time, and available technology.
Pseudonymization, defined in Article 4(5), retains a relationship with additional information and does not automatically mean the data fall outside the regulation. Moreover, transforming personal data to produce anonymous information does not by itself exempt the prior processing from the rules that apply to it.
This is the European framework. Other jurisdictions may apply different criteria. A technical assessment provides evidence about risk; a legal conclusion requires attention to the applicable law and the specific case.
The useful question for the survey at the beginning is what someone could discover using the remaining data and other sources available to them. Answering it helps determine how much detail to retain, how to make analysis possible, and what protection the people who supplied the information need.
Share