The talk considers some of the techniques highlighted in the Federal Committee on Statistical Methodology's Working Paper 22, which is the working "bible" of official statistical disclosure limitation methods in the US, from the cryptographic privacy perspective. An essential feature of cryptographic models of confidentiality protection is the randomized sanitizer, which generates the conditional probability distribution of the release data, given the confidential data. Properties of the randomized sanitizer determine the extent of provable confidentiality protection provided by a disclosure limitation technique. Analysis of some standard methods in WP-22 using their implied sanitizers reveals
the points of failure that might be addressed by using methods that borrow from the cryptographic privacy literature. |
