Introduction to Special Section

Abstract

In “Differential Privacy Applications to Bayesian and Linear Mixed Model Estimation,” John Abowd, Matthew Schneider, and Lars Vilhuber investigate questions about the quality of differentially private statistical models. Using two frameworks for constructing differentially private algorithms (sample-and-aggregate and objective perturbation), they fit linear mixed models and Bayesian linear mixed models using data from the U.S. Census Bureau’s Quarterly Workforce Indicators. While pointing out practical model-building issues for which differentially private solutions are still needed, they evaluate the quality that one can expect from privacy-preserving versions of these models.

Locations

• Journal of Privacy and Confidentiality - View - PDF
• DOAJ (DOAJ: Directory of Open Access Journals) - View