How can you determine whether a dataset with removed names is truly anonymous or merely pseudonymous?
If the remaining data points (quasi-identifiers) still allow re-identification through cross-referencing with external data, the dataset is pseudonymous, not anonymous — because the mapping can be reversed.
Example:
A dataset with names struck through but retaining gender, birth date, and ZIP code:
| Gender | Birth Date | ZIP | ||
|---|---|---|---|---|
| F | 11.05.1983 | 10115 | ||
| M | 03.12.1965 | 10115 | ||
| M | 27.08.1977 | 10178 |
Why this is pseudonymous, NOT anonymous:
- Only Frau Richter and Herr Schröder live in the ZIP area starting with "1011"
- Combined with the gender field, each person can be uniquely identified
- An attacker with access to the public voter registry could easily match these records
The test: Ask yourself — "Can these remaining fields be linked to any other available dataset to identify the person?" If yes, the data is pseudonymous. If truly no one can restore the personal reference, it's anonymous.
Tip: Removing names is the most obvious but least effective anonymization step. The quasi-identifiers (birth date, ZIP, gender) are what actually enable re-identification.
Go deeper:
Data re-identification (Wikipedia) — the linkability test that distinguishes the two.