Abstract: Abstract: We propose Bayesian methods to assess statistical disclosure risks for count data released under zero-concentrated differential privacy, focusing on settings with a hierarchical structure. We discuss applications of these risk assessment methods to differentially private data releases from the 2020 U.S. decennial census and perform empirical studies using public individual-level data from the 1940 U.S. decennial census. Here, we examine how the data holder's choice of privacy parameters affects disclosure risks and quantify the increases in risk when an adversary incorporates substantial amounts of hierarchical information.

Key words and phrases: Confidentiality, privacy, re-identification.