Rank-Based Record Linkage for Re-Identification Risk Assessment

摘要

There is a tradition of data administrators using record linkage to assess the re-identification risk before releasing anonymized microdata sets. In this paper we describe a record linkage procedure based on ranks, and we compare the performance of this rank-based record linkage against the more usual distance-based record linkage to re-identify records masked using several different masking methods. We try to elicit the reasons why RBRL performs better than DBRL for certain methods and worse than DBRL for other methods.