Kaitlyn Webb - Beyond Cell Suppression: A Formal Privacy Framework for Establishment Data

Title: Beyond Cell Suppression: A Formal Privacy Framework for Establishment Data
 

Abstract: The Quarterly Census of Employment and Wages (QCEW), a dataset on U.S. establishments used for economic measurement and policy decisions, is released as aggregate tables of wages and employment grouped by public attributes like state, county, or industry using cell suppression as a disclosure avoidance method. This limits how much information can be released and may be vulnerable to reidentification attacks. To increase confidentiality protection guarantees and data availability, we introduce a formal private framework, Establishment Gaussian Differential Privacy (EGDP), designed for skewed non-negative data like the QCEW. Adapted from Gaussian Differential Privacy (GDP), our framework controls the relative errors between privacy-preserving output and the original data to be small for large establishments and aggregates. We propose two mechanisms satisfying EGDP that add calibrated noise at the establishment and group-aggregate levels, scaling noise either by a group's sum or by its privacy-preserving maximum. These noisy values feed into a weighted least squares post-processing step that produces establishment-level privacy-preserving microdata (PPM), from which aggregate tables can be released that respect relationships across public attributes (e.g., county totals summing to state totals). We evaluate the privacy-accuracy trade-offs of both mechanisms on a synthetic dataset mimicking QCEW's structure to identify each mechanism's strengths and best use cases.

Short bio: Kaitlyn Webb is a Visiting Assistant Professor of Statistics at Amherst College, whose research develops formal privacy and differential privacy methods for economic data and official statistics, using sources from the Bureau of Labor Statistics, U.S. Census Bureau, and economic replication packages. Her work expands data access while providing rigorous confidentiality protections, advancing transparent methodology and reproducible science through privacy-protected data products and practical tools for researchers. She earned a dual-title Ph.D. in Statistics and Social Data Analytics from Penn State in 2026 and collaborates across computer science, political science, and economics. Her paper proposing Establishment Gaussian Differential Privacy won the student paper award at the 2024 International Conference for Establishment Statistics and earned her the 2023 Wray-Jackson Smith Scholarship from the ASA's Government Statistics Section.