Samuel Haney, Tumult Labs, and Rachel Marks, U.S. Census Bureau
This talk describes proposed differentially private (DP) algorithms that the U.S. Census Bureau is considering to release the Detailed Demographic and Housing Characteristics (DDHC) Race & Ethnicity tabulations from the 2020 Census. The tabulations contain hundreds of millions of statistics (counts) of demographic and housing characteristics for the entire U.S. population crossed by detailed races and tribes at varying levels of geography. We will describe a differentially private algorithm that adaptively chooses what statistics to release and what noise to add to the statistics based on the size of the population group. We will highlight our methodology of engaging with key stakeholders to iteratively elicit requirements for privacy and fitness for use, as well as design and tune a differentially private algorithm that meets these requirements.
Samuel Haney, Tumult Labs
Sam is a Scientist at Tumult labs where he works on creating systems for provably private data release. Before Tumult, he completed his PhD at Duke University, where he worked on differential privacy and graph algorithms.
Rachel Marks, U.S. Census Bureau
Rachel is chief of the Racial Statistics Branch in the Census Bureau’s Population Division where she leads a research team that analyzes data on race and ethnicity from the 2020 Census, 2020 Island Areas Census, American Community Survey, and the Current Population Survey. Rachel has conducted extensive outreach, presentations, and workshops with various stakeholder groups throughout her career and was a lead researcher for the 2015 National Content Test, which examined alternative ways to collect data on race and ethnicity.
author = {Samuel Haney and Rachel Marks},
title = {Differentially Private Algorithms for 2020 Decennial Census Detailed {DHC} Race \& Ethnicity},
year = {2022},
address = {Santa Clara, CA},
publisher = {USENIX Association},
month = jun
}
