Another Look at the Use of Noise Addition for Database Security
Norman Matloff · 1986
A number of mechanisms have been proposed for protecting the privacy of individual records in statistical databases. One such method is to perturb the data by adding random noise. It has been previously thought that this method, although increasing estimation variance, produces unbiased estimates. This paper demonstrates that in fact the addition of noise may introduce quite serious biases in the responses given to user queries.