Adjusting for Errors in Blocking Variables in Record Linkage

• Activity Number: 112
• Type: Invited
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Government Statistics Section
• Author(s): Nicole Dalzell and Jerry Reiter and Jody Heck Wortman*
• Companies: Duke University and Duke University and Duke University
• Keywords: Record Linkage ; Bayesian ; social science

Abstract

Data on a single individual or entity are often available from many different sources. Researchers seeking information from such diverse sources must identify which records in one file correspond to the same individual in another. The difficultly of this task is compounded when confidentiality concerns mandate that records are stored without a unique identifier. Record linkage is the process of comparing information in such de-identified records in order to determine groups of records which correspond to the same individual and create a final linked data set to be used for inference. The linking process is often hindered by the possibility of errors in the record fields used for comparison. This work presents a Bayesian method for record linkage which accounts for potential errors in the linking fields while incorporating the uncertainty of the matching process into inference conducted the final linked data set. An example with social science data is presented to illustrate the method.