Abstract

Recent international trends in privacy-protecting techniques applied to official statistics include the use of perturbative methods. The U.S. Census Bureau has investigated the applicability of perturbative methods based on the methodology of differential privacy, which was originally developed in the field of computer science in order to prevent “database reconstruction attacks”, in which attackers attempt to identify personal information by combining multiple published statistical tables. On this basis, the U.S. Census Bureau has created and published statistical tables that use differential privacy for the 2020 United States Census.
Exploring the applicability of differential privacy techniques to Japanese official statistics is worthwhile both for expanding the future scope of creation and publication of official statistical tables, and shaping the future direction of secondary use of official statistics. Several empirical studies have examined the potential of differential privacy as an anonymization method for detailed geographical data from the Japanese Population Census.
This paper investigates the possibility of adapting differential privacy to cross-tabulated data created using individual data from the 2020 Japanese Population Census, and conducts a comparative analysis of data usability at different geographical levels for perturbed statistical tables created based on differential privacy.