Consider the level of information content

Matching is more than just a comparison of available data. Both the information content and the representation of the data being compared are considered.

One or more columns in one record must have equivalent columns in the other record to compare them. For example, in order to match on family name and age, both records must have columns containing family name and age information. Although for a two-source match, the metadata for the comparable columns does not need to be identical.

For a record linkage project to be feasible, it is possible for a human to examine the record pairs and declare with reasonable certainty which of the pairs are a match or a nonmatch. For example, if the only column in common between two sources is gender, you do not then conclude that, because the gender agrees, the pair represents the same individual.

Information content measures the significance of one column over another (discriminating value). For example, a gender code contributes less information than a tax identification number.

Information content also measures the significance of one value in a column over another. In the United States, John contributes less information than Dwezel in a GivenName column. The given name, John, in the United States is much more common than the given name, Dwezel. Significance is determined by the reliability of the value and the ability of the value to discriminate a match from a nonmatch. And any comparison of records needs enough information to reach a reliable conclusion. For example, two identical customer records that contain only the family name, Smith, and that miss values in every other name and address column, do not have enough information to determine that the records represent the same person.