Measurement Error and Statistical Disclosure Control

Statistical agencies release microdata to researchers after applying statistical disclosure control (SDC) methods. Noise addition is a perturbative SDC method which is carried out by adding independent random noise to a continuous variable or by misclassifying values of a categorical variable according to a probability mechanism. Because these errors are purposely introduced into the data by the statistical agency, the perturbation parameters are known and can be used by researchers to adjust statistical inference through measurement error models. However, statistical agencies rarely release perturbation parameters and therefore modifications to SDC methods are proposed that a priori ensure valid inferences on perturbed datasets. Measurement Error and Statistical Disclosure Control Natalie Shlomo1 1 Southampton Statistical Sciences Research Institute, University of Southampton, Highfield, Southampton, SO17 1BJ, United Kingdom [email protected] Abstract. Statistical agencies release microdata to researchers after applying statistical disclosure control (SDC) methods. Noise addition is a perturbative SDC method which is carried out by adding independent random noise to a continuous variable or by misclassifying values of a categorical variable according to a probability mechanism. Because these errors are purposely introduced into the data by the statistical agency, the perturbation parameters are known and can be used by researchers to adjust statistical inference through measurement error models. However, statistical agencies rarely release perturbation parameters and therefore modifications to SDC methods are proposed that a priori ensure valid inferences on perturbed datasets. Statistical agencies release microdata to researchers after applying statistical disclosure control (SDC) methods. Noise addition is a perturbative SDC method which is carried out by adding independent random noise to a continuous variable or by misclassifying values of a categorical variable according to a probability mechanism. Because these errors are purposely introduced into the data by the statistical agency, the perturbation parameters are known and can be used by researchers to adjust statistical inference through measurement error models. However, statistical agencies rarely release perturbation parameters and therefore modifications to SDC methods are proposed that a priori ensure valid inferences on perturbed datasets.