An investigation of the effectiveness of statistical distributions for additive fixed data perturbation

Statistical databases provide security of confidential data by preventing access to individual values. However, under certain situations, the security of individual data can be compromised by statistical functions alone. A number of approaches have been suggested to counter this problem. This paper addresses one such approach, namely, fixed data perturbation. The purpose of the research is to evaluate the effectiveness of different statistical distributions in perturbing different forms of database populations when employing an additive form of fixed data perturbation. Specifically, this paper evaluates the effectiveness of the Normal, Log-normal, Gamma and Uniform distributions in perturbing data sets with a defined distribution. The results of extensive Monte-Carlo experiments conducted on different database populations reveal that the Uniform distribution provides the best performance (high security and low bias) for all database populations except those described by the Log-normal distribution.