Computational issues in the design of transition probabilities and disclosure risk estimation for additive noise

Abstract

The Australian Bureau of Statistics has developed an additive noise method for automatically and consistently confidentialising tables of counts ‘on the fly’. Statistical properties of the perturbation are defined by a matrix of transition probabilities. The present paper looks at mathematical and computational aspects of an approach mentioned in the literature for how to design those probabilities. In the second part, the paper proposes computation of feasibility intervals as a technique to compare the effects of different variants for post-tabular perturbative protection methods on disclosure risk and provides experimental results.