String Comparator Metrics and Enhanced Decision Rules in the Fellegi-Sunter Model of Record Linkage.

Abstract

To locate matches across pairs of lists without unique identifiers it is sometimes necessary to compare strings of letters. String comparators are used in production computer matching software during the Post Enumeration Survey for the 1990 U.S. census. A string comparator metric is described that partial]y accounts for: (1) typographical variation in strings such as first name or surname; (2) decision rules that use the string comparator; and (3) improvements in empirical matching results. The string comparator metric for comparing partially agreeing strings extends the Jaro string comparator. How general methods of accounting for partial agreement fit with the Fellegi-Sunter (I. P. Fellegi and A. B. Sunter, 1969) model of record linkage is described. A formal method of modeling how to adjust matching weights between pure agreement and pure disagreement is presented. The procedure is illustrated for files for which the truth of matches is known. It is demonstrated that the theoretical rules of Fellegi and Sunter are still valid when general weighting adjustments accounting for partial agreement are performed. Eight tables contain illustrative data. (SLD) Reproductions supplied by EDRS are the best that can be mPde from the original document. U.S. DEPARTMENT OF SOUCATION Office al Educational ssssrcli and Improvement EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC) is document hes been reproduced ea received horn th penal O orEaruciroOn cognate% it 0 Minor changes Rove been made to improve reproduction quality Points of view or opinions stated in th4 <Metament do not neasseenly represent official OERI position or policy PERMISSION TO REPRODUCE THIS MATERIAL HAS BEEN GRANTED BY

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 787 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.