Application of the Truncated Distributions and Copulas in Masking Data
Rahul A. Parsa, Jay J. Kim, Myron J. Katzoff · 2009
In masking microdata, two approaches - adding independent noise and multiplying by independent noise have been used. The truncated distribution has been used for masking microdata. The random variable which follows the truncated distribution serves as the noise factor. For multiplicative noise method, the natural candidate distribution is the one which is centered at 1 and for additive noise method, it would be one centered at 0. Kim (2007) investigated triangular distribution truncated around 1 as noise distribution for multiplicative noise. In this paper we generalize his idea using copulas and correlated noise. We show that by using correlated noise, we can protect the moments, that is, the moments of the perturbed variable will have the same values as the original variable. We present two examples, one using correlated noise from a triangular distribution and second, from a truncated uniform distribution.