We are seeking panelists to help assess the potential impacts of new privacy protections on U.S. census data. We ask only that each panelist selects a use case of interest and compares some results using differentially private versions of 1940 or 2010 census data with results using original census data. To facilitate comparisons, the organizers can provide panelists with subsets of the data suited to their needs.
To express interest, please email Jonathan Schroeder (email@example.com), David Van Riper (firstname.lastname@example.org), and Tracy Kugler (email@example.com) by November 19, 2019.
Protecting the privacy of census respondents while publishing quality data are dual mandates for the U.S. Census Bureau. In August 2018, the Bureau announced a major change in their approach to privacy protection. According to the Bureau, increases in computing power and access to large individual-level databases mean that their traditional disclosure avoidance techniques no longer provide strong enough protection. In response, the Bureau plans to adopt a framework termed “differential privacy” for its 2020 Census disclosure avoidance system, which will entail injecting random noise into nearly all published data in order to guarantee a minimal risk of privacy loss.
While the planned approach would achieve state-of-the-art privacy protections, we believe census data users should be deeply concerned. A strict application of differential privacy would have pervasive and potentially severe impacts on the utility and accuracy of the country’s benchmark population data. In particular, for any given count statistic (e.g., total populations of urban areas, Black populations of census tracts, children in American Indian reservations, etc.), the amount of injected noise is roughly scale-independent, which tends to produce proportionally large errors for many small places and subgroups.
To help data users assess potential impacts and report their concerns, the Census Bureau has released demonstration data products, which supply differentially private versions of 1940 and 2010 data. In this session, the organizers will provide an overview of the Bureau’s reported plans, and panelists will present and discuss findings about the accuracy and utility of the demonstration data for various use cases.
|Introduction||David Van Riper Minnesota Population Center||15|
|Panelist||Tracy Kugler IPUMS||7|
|Panelist||Elizabeth Delmelle University of North Carolina at Charlotte||7|
|Panelist||Jason Jurjevich Portland State University||7|
|Panelist||John Cromartie USDA||7|
|Panelist||Nicholas Nagle University of Tennessee||7|
|Discussant||Jonathan Schroeder University of Minnesota||7|
To access contact information login