Linkability
Linkability is a fundamental criterion in evaluating the anonymity of synthetic data. It refers to the risk that two or more records within a dataset—or across datasets—can be linked to the same individual based on shared attributes.
Consider two datasets: one containing medical records and another containing shopping habits. An attacker who knows that an individual purchased a specific health-related product may link that information with medical conditions listed in the other dataset if overlapping attributes, such as age, gender, or location, are shared between the datasets. This connection can lead to the identification of sensitive personal information.
List of Metrics Measuring Linkability Criteria
Factors Influencing the Risk of Linkability
The risk of linkability increases when datasets share a significant number of overlapping attributes, especially if these attributes are specific or unique. Attributes like exact birth dates, rare job titles, or detailed geographic information can create strong links between records.
Additionally, the use of highly structured datasets with limited variability across records can amplify the risk. For example, a dataset containing many individuals with identical demographic traits can create patterns that are easier to connect between different datasets.
This criterion is particularly relevant for individuals whose records appear in multiple datasets. Linking multiple data sources could reveal a more complete profile of the individual, increasing the risk of re-identification.
Importance of Mitigation
To limit the risks of linkability, the metrics we have developed systematically assess the probability that records can be connected across datasets based on shared attributes. These analyses help identify overlaps and reduce the likelihood of linkages by introducing controlled variability or obfuscation in synthetic data. By addressing these vulnerabilities proactively, the Avatar method ensures greater protection of individual privacy.