Inference control in statistical databases :

Abstract

An efficient and effective data distortion technique is introduced. This "Probability Data Distortion, " which is not easily compromisable and has asymptotically the same statistical properties as the original data, is significantly different from the conventional Point Data Distortion technique which adds random errors to the original values. This mechanism, the data distortion by probability distribution, is resistant to compromise and provides better exposure for statistical analysis than do the existing data distortion techniques.

The legal code on "The Protection of Human Subjects in Research Activities" requires that sensitive information about an individual should be protected from unauthorized release and at the same time, those data should be available for statistical analysis. To meet these conflicting goals, recent research efforts focus on creating distorted data which is not easily compromisable and yet preserves the statistical properties of the original data.