Noise Multiplication and Multiple Imputation as Alternatives to Top Coding for Statistical Disclosure Control: An Overview and Comparison

• Activity Number: 89
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Government Statistics Section
• Author(s): Martin Klein*+
• Keywords: EM Algorithm ; Partially synthetic data ; Synthetic data ; Tuning parameter

Abstract

When statistical agencies release microdata to the public, a major concern is the control of disclosure risk, while ensuring utility in the released data. When releasing a variable such as income, top coding (TC) is a commonly used method for protecting large values against disclosure. In this paper, multiple imputation (MI) and noise multiplication (NM) are considered as two alternatives to TC for statistical disclosure control. We present methodology for both the MI and NM methods, following work of An and Little (2007), and Klein, Mathew and Sinha (2012), respectively. For NM, data analysis methods are presented under two types of data releases: (i) each released value includes an indicator of whether or not it has been noise perturbed, and (ii) no such indicator is provided. An empirical comparison of the three methods, MI, NM, and TC, is presented. The empirical comparison considers some parametric models, and evaluates the methods in terms of accuracy of resulting inferences for the unknown parameters. We show that by appropriately adjusting the variance of the noise generating distribution, NM can provide either more or less accurate inference than MI.