Course contentsShow
← Security
Lesson 2904 of 3,05254. Privacy EngineeringFree lesson

Quasi-Identifiers and Re-identification Risk

Understand how combinations of non-unique attributes can re-identify individuals in anonymous data.

Quasi-Identifiers and Re-identification Risk

You'll learn how seemingly innocent combinations of non-unique attributes can unmask individuals in "anonymized" datasets.

The Danger in Combinations

After removing direct identifiers like names and Social Security numbers, organizations often believe their data is safe to share. But here's the problem: quasi-identifiers are attributes that, while not uniquely identifying someone individually, can re-identify people when combined.

Think of it like this: Knowing someone lives in California isn't revealing. Knowing they're 43 years old also isn't. But knowing someone is a 43-year-old female cardiologist living in a small Montana town? That combination might point to exactly one person.

Common Quasi-Identifiers

Typical quasi-identifiers include:

  • Demographics: age, gender, ethnicity, marital status
  • Geography: ZIP code, city, neighborhood
  • Temporal data: dates of events (birth year, admission dates)
  • Professional info: occupation, employer size, industry
  • Behavioral patterns: purchasing habits, travel patterns

The Re-identification Attack

Attackers perform re-identification by:

  1. Taking an "anonymized" dataset with quasi-identifiers
  2. Obtaining an auxiliary dataset with overlapping attributes (voter registrations, social media, public records)
  3. Matching records based on quasi-identifier combinations
  4. Successfully linking anonymous records back to real identities

The famous example: researchers re-identified Massachusetts Governor William Weld's medical records by matching just his ZIP code, birth date, and gender against voter registration lists—only six people shared those three attributes.

Measuring Risk

Re-identification risk depends on:

  • Number of quasi-identifiers present
  • Uniqueness of combinations in the population
  • Availability of auxiliary data sources
  • Population size and diversity

Key Takeaway: Removing direct identifiers isn't enough—quasi-identifiers create combinatorial risk that can unmask individuals even in large "anonymized" datasets, requiring careful analysis and additional privacy techniques like generalization or suppression.