Obstacles to Adoption of DP at National Statistical Agencies
In this talk, I outline a set of problems that face National Statistical Agencies hoping to adopt differentially private (DP) statistical disclosure limitation (SDL) methods, drawing on my own experiences with the 2020 Decennial Census. I argue that a number of these obstacles are unlikely to be resolved in the near term, but that we can ease adoption of DP SDL algorithms by building semantics (interpretations of the privacy guarantee) that acknowledge long-term challenges in the SDL implementation environment. I close by briefly discussing a few obstacles to adoption of DP that are concerned with design of mechanisms more than interpretation of the privacy guarantee.