Jaro-Winkler compares matching characters, character order, and a shared prefix. Its similarity score and the corresponding Jaro-Winkler distance run in opposite directions: similarity is 1 for identical strings; a corresponding distance of 1 - similarity is 0 for identical strings.

Datablist uses the similarity score. See distance matching for the difference between raw distance, similarity thresholds, and field scores. Check the convention before copying a threshold from another tool. The RapidFuzz documentation, for example, exposes separate distance and similarity functions.

The method extends Jaro similarity by increasing the score when two values start with the same characters. This makes it useful for short names and values where a typo appears near the end.

Jaro-Winkler example: Martha and Marhta

Compare Martha with Marhta. The letters are the same, but t and h appear in a different order. With the values lowercased as Datablist does:

  1. All six characters match within Jaro's matching window.
  2. The out-of-order characters contribute one transposition after dividing the mismatch count by two.
  3. Jaro similarity is about 0.9444.
  4. The shared prefix mar increases Jaro-Winkler similarity to about 0.9611.

Datablist applies a prefix weight of 0.1 to at most four initial characters. In this example, the calculation is 0.9444 + 3 × 0.1 × (1 - 0.9444).

A 95% similarity threshold accepts this pair; 98% rejects it. A 96.11% score is a string comparison, not a 96.11% probability that the rows identify the same person. Review another identifier before merging.

When Jaro-Winkler helps

Jaro-Winkler can find variations such as transposed or missing characters in:

  • Person names
  • Company names
  • Short labels
  • Addresses

It does not understand whether two values refer to the same real-world entity. A high name similarity can still join different people.

Jaro-Winkler compared with Levenshtein

Levenshtein distance counts the edits needed across the entire string. Jaro-Winkler considers matching character order and favors a shared prefix.

Test both on a labeled sample when matching quality matters. The best threshold depends on the length, language, and cleanliness of your values.

⚠️ Review approximate matches

Use another identifier such as email, phone, domain, address, or external ID before merging records based on a fuzzy name score.

Use Jaro-Winkler in Datablist

In the Duplicates Finder, select a text property, choose Distance (Levenshtein or Jaro Winkler), select Jaro Winkler distance, and set the similarity threshold. Distance matching is available on paid plans.

Read the data matching guide for algorithm selection and the Duplicate Finder reference for the current settings.