Multivariate Top-Coding for Statistical Disclosure Limitation

Area of Science:

• Statistics
• Data Privacy
• Computer Science

Background:

• National statistical agencies face challenges in releasing microdata with many attributes while controlling disclosure risk.
• Altering microdata for disclosure limitation requires considering variable relationships to maintain data quality.
• Univariate Statistical Disclosure Limitation (SDL) methods may inadequately protect certain subpopulations.

Purpose of the Study:

• To propose a multivariate top-coding method for enhanced statistical disclosure limitation.
• To address the limitations of univariate top-coding in protecting subpopulations.
• To improve the quality and privacy of public microdata sets.

Main Methods:

• Developed a multivariate top-coding approach by clustering variables based on a closeness metric.
• Utilized Association Rule Mining techniques within variable clusters to formulate top-coding rules.
• Extended the methodology for a similar multivariate bottom-coding procedure.
• Illustrated the method using a realistic, large-scale multivariate data set.

Main Results:

• The proposed multivariate top-coding method offers improved disclosure control compared to univariate methods.
• Clustering variables and applying association rule mining within clusters effectively identifies and manages extreme values.
• The approach is demonstrated to be applicable to genuine, complex datasets.

Conclusions:

• Multivariate top-coding, by considering inter-variable relationships, provides superior privacy protection for microdata.
• This method enhances the utility of public-use microdata by better balancing data utility and privacy.
• The proposed approach offers a robust framework for advanced Statistical Disclosure Limitation.