Multi-Dimensional Randomized Response

In our data world, a host of not necessarily trusted controllers gather on individual subjects. To preserve her privacy and, more generally, informational self-determination, the has to be empowered by giving agency own data. Maximum is afforded local anonymization, that allows each anonymize before handing them controller. Randomized response (RR) anonymization approach able yield multi-dimensional full sets anonymized microdata are valid for exploratory analysis and machine learning. This so because an unbiased estimate distribution true individuals can obtained from their pooled randomized Furthermore, RR offers rigorous guarantees. The main weakness curse dimensionality when applied several attributes: as number attributes grows, accuracy estimated quickly degrades. We propose complementary approaches mitigate problem. First, we present two basic protocols, separate attribute joint all attributes, discuss limitations. Then introduce algorithm form clusters in different viewed independent performed within cluster. After that, adjustment set repairs some loss due assuming independence between using separately or cluster-wise RR. also empirical work illustrate proposed methods.