Discriminating Powers of Partial Agreements of Names for Linking Personal Records

Abstract

Abstract:The preceeding paper examined the logical basis of an exact way of calculating the discriminating powers of people’s names when they only partially agree. The method has application to automated file searching and record linkage. The present account describes an empirical test of the approach. Use is made of some 2000 comparison pairs of male given names, obtained as a byproduct from an earlier linkage study. The test shows that exact value-specific ODDS can indeed be calculated for common names when compared with their accepted synonyms (e.g. JOSEPH versus JOE). Moreover, the use can be extended to include rare variants, by arranging these into groups defined in value-specific terms (e.g. as selected blocks in an alphabetically sequenced listing, or combinations of such blocks). A majority of all name comparisons may be handled in this manner.The added precision serves to reduce the numbers of records that are ambiguously linked and require labour intensive clerical resolution.

Results from simulated data sets: probabilistic record linkage outperforms deterministic record linkage

“…Characteristics of string variables are that they have a large number of possible values (high discriminating power), but they are also prone to error (high error rate). Using Soundex [22] or another phonetic algorithm [23,24] can reduce errors and make a string variable more suitable for record linkage. In anonymous data sets, however, names are usually absent.…”