Extensions of the Re-identification Risk Measures Based on Log-Linear Models

摘要

A global measure of the re-identification risk in microdata files is analyzed. Two extensions of the log-linear models are presented. The first methodology considers the weights in the analysis of contingency tables. The results of several tests performed on real data are presented. In the framework of statistical disclosure control, the second methodology proposes a maximum penalized likelihood approach to the computation of smooth estimates.