Imputation as a Practical Alternative to Data Swapping

• Activity Number: 311 - SPEED: Environment and Health, Governmental Policies and Population Surveys, Part 2
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 9:25 AM to 10:10 AM
• Sponsor: Government Statistics Section
• Author(s): Saki Kinney* and David Wilson and Alan Karr and Kelly Kang
• Companies: RTI International and RTI International and RTI International and NSF
• Keywords: Data confidentiality; Disclosure; Surveys

Abstract

Data swapping is typically used to protect against identity disclosure by adding uncertainty to any potential record linkage. It is commonly used by several agencies but can be difficult to implement properly for complex surveys and there are few automated routines publicly available to facilitate the process. The perturbation rate is typically constrained to maintain satisfactory analytic validity between swapped and unswapped variables. Model-based imputation provides a more intuitive, transparent, and flexible approach while providing a better balance between disclosure risk and data utility. It can also be easier to implement with automated routines available in R. We will describe this approach and provide a comparison to data swapping using an example from the Survey of Earned Doctorates. We find that the methods provide comparable analytic utility at low levels of perturbation and when analysis variables have been accounted for in both algorithms. Imputation, however, is able to preserve a larger number of relationships in the data and at higher levels of perturbation.