NameSearch Probabilistic and Deterministic Matching

Probabilistic matching is achieved through the derivation of a weighting scheme that considers the frequency of identification information to formulate a score and/or ranking. In principal the goal is to apply lighter weights to common elements and use the uncommon information to differentiate records. As an example the record John Smith at 15 Vorris Dr., has a common name element, “John Smith” and an uncommon street address. Proponents of this theory would state that by applying greater weight to the uncommon street address a more precise match is realized.

NameSearch® comes with a function called predict. Predict is used to calculate the number of records that would be returned from a query without accessing the underlying database. To use the predict function you first must have a representative sample of names and/or addresses. The sample will be used as input into the predict frequency analysis tool that is part of the NameSearch® graphical user interface. The predict frequency routine creates a special table that is used by the predict function. By utilizing predict with the advanced comparison routines NameSearch® provides its clients with the capability to perform probabilistic matching.

The opposition to this methodology espouses extremely poignant criticisms. The lack of a unifying scoring scheme produces seemingly unpredictable results. Even though a specific algorithm is used for determining scores users get frustrated when they observe the application of varied match criteria. On-line applications utilizing this methodology are often abandoned due to the seemingly unpredictable nature of the results. Through empirical study it has been shown that the differentiation of records by placing greater weight on common elements is flawed. Uncommon elements have a greater degree of variation. By utilizing the field that is most prone to errors many good matches are missed.

Deterministic matching is significantly easer to implement and maintain. The probabilistic approach requires extensive frequency analysis to be performed in order to understand the frequency characteristic of the data population. To insure accuracy, volatile and fluid data populations require constant maintenance. Deterministic matching yields highly uniform results and has been shown to be more tolerant of variations and have fewer false positives. False positive are those situations where two records are incorrectly matched.

NameSearch® supports the probabilistic model however deterministic matching is known to yield significantly better and more reliable results.

NameSearch® General Information