PHSafe: Disclosure Avoidance for the 2020 Census Supplemental Demographic and Housing Characteristics File (S-DHC)
William Sexton, Skye Berghel, Bayard Carlson, Sam Haney, Luke Hartman, Michael Hay, Ashwin Machanavajjhala, Gerome Miklau, Amritha Pai, Simran Rajpal, David Pujol, Ruchit Shrestha, Daniel Simmons-Marengo
TL;DR
PHSafe addresses the challenge of privately releasing the U.S. 2020 Census Supplemental Demographic and Housing Characteristics File (S-DHC) by introducing a discrete-Gaussian-based mechanism that achieves zero-concentrated differential privacy ($z$CDP). The algorithm operates on basis-table cells through a filter–join–transform–measure workflow, producing independent noisy measurements for eight PH tables and employing postprocessing to ensure nonnegativity and credible intervals while preserving privacy. A formal privacy preliminaries and analysis establish stability and composition properties, with conversions to bounded $z$CDP to meet practical accuracy targets measured by 90% margins of error (MOE). The implementation leverages Tumult Analytics for private joins and stability accounting, and parameter tuning is guided by the PHExplore tool, balancing privacy budgets, MOEs, and data availability. The work provides a complete framework from algorithm design to deployment, offering rigorous privacy guarantees and practical pathways for producing interpretable, privacy-protected census statistics.
Abstract
This article describes the disclosure avoidance algorithm that the U.S. Census Bureau used to protect the 2020 Census Supplemental Demographic and Housing Characteristics File (S-DHC). The tabulations contain statistics of counts of U.S. persons living in certain types of households, including averages. The article describes the PHSafe algorithm, which is based on adding noise drawn from a discrete Gaussian distribution to the statistics of interest. We prove that the algorithm satisfies a well-studied variant of differential privacy, called zero-concentrated differential privacy. We then describe how the algorithm was implemented on Tumult Analytics and briefly outline the parameterization and tuning of the algorithm.