At Statistics Canada, matching without unique identifiers is done using both deterministic hierarchical linkage and the probabilistic record linkage method developed by Fellegi-Sunter [1]. We present a review of the challenges involved in probabilistic linkage and tools developed to determine optimal linkage parameters. A SAS-based generalized system called G-Link includes a full implementation of the Fellegi-Sunter theory within a graphical interface. Analysis with linked data is also a crucial issue. Naively treating linked data as if it were linked without errors can lead to biased inference. We focused on a framework to exploit linked data while taking into account both false positive and false negative matches.

[1] I.P. Fellegi and A.B. Sunter (1969). A Theory of Record Linkage, JASA, vol. 64, 1183-1210.