Rationing Out Privacy-Loss: Proportional Budget Expenditure in the 2020 Decennial Census Disclosure Avoidance System

• Activity Number: 458 - Differential Privacy Research and Applications at the U.S. Census Bureau
• Type: Topic Contributed
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Government Statistics Section
• Author(s): William Sexton*
• Companies: U.S. Census Bureau
• Keywords: Differential privacy; privacy-loss budget; statistical accuracy; production possibilities frontier

Abstract

Differentially private data publication systems must weigh the cost of increased statistical accuracy against foregone privacy. However illustrations of the tradeoff between a single global privacy-loss budget and a single accuracy measure fail to capture the naunce of complex privacy algorithms where a global budget may be allocated among multiple query workloads of varying importance. Given a weighted set of workloads, the matrix mechanism will, in theory, derive optimally accurate noisy answers. Ideally, subject matter experts will define the relative importance of the workloads. We apply the matrix mechanism as a subroutine of the 2020 Decennial Census Disclosure Avoidance System on publicly available 1940 Census data. We illustrate the tradeoff on accuracy among multiple workloads across various budget distributions.