Differential Privacy and Its Applicability for Official Statistics in Japan – A Comparative Study Using Small Area Data from the Japanese Population Census

Abstract

As part of its preparations for the 2020 U.S. Population Census, the U.S. Census Bureau uses the methodology of differential privacy to create privacy-preserved official microdata. It is expected that the use of differential privacy for official statistical data will become a topic in Japan in the future. In this paper, we survey the current discussion on the use of differential privacy for creating official statistical data. We describe the differential privacy method used by the U.S. Census Bureau, develop a method to apply differential privacy to Japanese population grid data, and conduct a comparison between different differential privacy methods by applying them to Japanese small area data. Results demonstrate that for population grid data, both the non-negative Wavelet method (and its public-n variant) and the US census top-down method preserve the non-negativity and sparsity of the original data, but each results in a different degree of error. This provides an important criterion for choosing the most appropriate method among the three options.