Statistical Disclosure Techniques Based on Multiple Imputation

• Activity Number: 172
• Type: Invited
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Trivellore E. Raghunathan*+ and Roderick J. Little and Fang Liu
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: multiple imputation ; combining rules ; partial synthesis ; full synthesis ; inferentially valid ; Bayesian method

Abstract

Statistical disclosure control (SDC) is concerned with the modification of statistical data that contain confidential information on individual entities (persons, households, businesses, etc.), to prevent third parties from revealing sensitive information about identifiable individuals. In recent years, SDC has received increasing attention. Today's sophisticated computer technology and the increased access to data via the internet and electronic media allow data intruders to access information and identify individuals more easily, increasing concerns about respondent privacy. Confidentiality is vital for the future cooperation of respondents to provide high-quality data. On the other hand, policymakers need maximum information to make efficient and timely decisions. This paper summarizes a cluster of SDC techniques that protect against disclosure by deleting values of variables in the dataset and replacing them by values drawn from their predictive distribution. The imputation uncertainty is reflected by the method of multiple imputation (MI). The repeated sampling properties of inferences from the altered data are evaluated.