Computational Experiments with Minimum-Distance Controlled Perturbation Methods

摘要

Minimum-distance controlled perturbation is a recent family of methods for the protection of statistical tabular data. These methods are both efficient and versatile, since can deal with large tables of any structure and dimension, and in practice only need the solution of a linear or quadratic optimization problem. The purpose of this paper is to give insight into the behaviour of such methods through some computational experiments. In particular, the paper (1) illustrates the theoretical results about the low disclosure risk of the method; (2) analyzes the solutions provided by the method on a standard set of seven difficult and complex instances; and (3) shows the behaviour of a new approach obtained by the combination of two existing ones.