Multiple Record Linkage: Generalizing the Fellegi-Sunter Theory

• Activity Number: 499
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Mauricio Sadinle*+
• Companies: Carnegie Mellon University
• Keywords: EM algorithm ; Fellegi-Sunter theory ; Mixture models ; Undirected graphs

Abstract

A probabilistic method to link multiple data files is presented. This task is not trivial when there is no unique identifier available for the individuals recorded. In order to perform a record linkage for more than two data files, some authors have used record linkages for each pair of data files. However, this approach leads to discrepancies because the transitivity of the linkage decisions is not guaranteed. The method presented here is a generalization of the Fellegi-Sunter theory for record linkage for two data files. This method incorporates the transitivity of agreement since the computation of the data used to model matching probabilities. We firstly present the K-ary product of the data files to be linked as the union of all the possible subsets that contain the possible patters of agreement of the record K-tuples. Comparison data are used to model matching probabilities, incorporating the transitivity of the comparison. Also a mixture model is used to fit matching probabilities and the adjustment of this model is made through the EM algorithm. We present a method to decide the record K-tuples membership to the subsets of patterns of agreement.