Chances and Obstacles for Multiply Imputed Synthetic Datasets

Abstract

The main focus of the first statistical disclosure limitation (SDL) techniques proposed in the literature was on providing sufficient disclosure protection. At that time, agencies paid only little attention to the negative impacts of these approaches on data utility. Over the years, more and more sophisticated methods evolved. However, these methods also became more complicated to implement and often required correction methods difficult to apply for nonstandard analysis. For these reasons, most agencies still tend to rely on standard, easy-to-implement SDL techniques such as data swapping or noise addition, although it has been shown repeatedly that these methods can have severe negative consequences on data utility and may even fail to fulfill their primary goal – to protect the data sufficiently (see, for example,Winkler (2007b)).