The development of matching criteria for epidemiological studies using record linkage techniques.

The hypothesis that pairs of records with identical surnames, given names and birth dates represent the same person was tested by compiling a frequency distribution of the number of birth date digits in common when the names contained in two registers were matched. This distribution was compared with a computer simulation of the distribution which would be expected if the paired records represented different people. The divergence of the two distributions in the region of five and six birth date digits in common confirmed the hypothesis. Where surname, two given names and at least four date of birth digits matched, only 0.012% of the matching records represented different people. Where surname, two given initials and six date of birth digits matched, 0.1% of the matched records represented different people.