Bayesian Multiscale Multiple Imputation with Implications to Data Confidentiality

• Activity Number: 14
• Type: Topic Contributed
• Date/Time: Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Scott Holan and Daniell Toth and Marco A.R. Ferreira*+ and Alan F. Karr
• Companies: University of Missouri-Columbia and Bureau of Labor Statistics and University of Missouri-Columbia and National Institute of Statistical Sciences
• Keywords: Confidentiality ; Disclosure ; Dynamic linear models ; Multiple imputation ; Multiscale modeling ; QCEW

Abstract

Many applications present data that are collected on multiple scales of resolution, as for example when data are aggregated across different scales both longitudinally and by economic sector. Frequently, such data sets experience missing observations in a manner that they can be accurately imputed using our Bayesian multiscale multiple imputation method. This method borrows information both longitudinally and across different levels of aggregation to produce accurate imputations of missing observations. One important implication of such methodology is its potential effect on confidential databases protected by means of cell suppression. In a large scale empirical study using the U.S. Bureau of Labor Statistics Quarterly Census of Employment and Wages, we find that several of the predicted cells are within 1% accuracy, thus causing potential concerns for data confidentiality.