Abstract
Many microdata-masking methods either have their parameters distorted by outliers or destroy the very outliers that legitimate analyses need to preserve. This paper robustifies noise-based masking so that the perturbation is estimated reliably even in the presence of extreme values, keeping the masked data both safe and analytically faithful. It compares the robustified methods against established masking techniques to show what is gained. Presented at Privacy in Statistical Databases, the work is part of a broader programme bringing robust statistics into disclosure control for continuous data.
statistical disclosure control microdata masking robust statistics noise addition