Lack-of-Balance and R-indicators As Measures of Utility in Statistical Disclosure Control

Within the total survey error paradigm (TSE) one no rmally tries to identify the principal sources of e rror in a survey, e.g. errors due to coverage, sampling, nonresponse, measurement, processing and imputation, a nd of course issues pertaining to validity or relevanc e. The impacts of these errors are mitigated when p ossible, or at least we should try to characterize them and their sources (Groves, 2004). Statistical disclosur e control (SDC) methods, i.e. measures taken to protect confi de tial data can be viewed as an additional error s ou ce, and in some cases this is exactly how risk reductio n is achieved, e.g. noise is purposely added to mic ro or tabular data. Similarly, when protection is achieve d by suppressing data, uncertainty is introduced. Consequently, Karr (2012) includes disclosure limit ation error as a component of TSE at the conceptual level. This uncertainty may consist of an increase in variances but may also introduce bias in estimat es stemming from a protected data set. The main differ ence compared to other error sources is that the producers of official statistics are in a position t assess the increase of uncertainty and decide ho w much to add and attempt to do so in a controlled manner. Op timally, uncertainty is added while at the same tim e the utility of the data is preserved, thus there is an intention from the producer to find a balance betwe n risk and utility.