DataTerminology.comNEW
A statistical model for linking records using match and non-match probabilities.
DESCRIPTION
The Fellegi–Sunter model formalizes probabilistic record linkage by comparing fields and computing match likelihoods. It defines decision thresholds for match, non-match, and possible match. Weighting schemes reflect agreement and disagreement across fields. Blocking strategies improve scalability by limiting comparisons. The model underpins many entity resolution systems and tooling. It supports rigorous evaluation with precision, recall, and ROC analysis. Extensions incorporate machine learning and adaptive weights.
ALTERNATIVE DEFINITIONS
Wikipedia
The Fellegi–Sunter model is a probabilistic model of record linkage that classifies pairs of records by comparing their fields.
Captures probabilistic classification based on field-level comparisons.
U.S. Census Bureau
The Fellegi-Sunter model uses likelihood ratios based on agreements and disagreements to decide whether record pairs are matches, nonmatches, or possible matches.
Emphasizes likelihood ratio decision rules and thresholding.
RELATED TERMS
Entity Resolution Entity Matching Blocking Weighted Approximate Matching Expectation-Maximization Precision and Recall ROC Curve Duplicate Detection Bayesian Matching
COMPETING TERMS
Deterministic Matching
RELATIONS
- Broader: Model
KNOWLEDGE AREAS
Data Management Data Quality Reference & Master Data Data Governance Data Science
TARGET AUDIENCE
Data Steward Data Engineer Data Scientist Data Governance Manager Data Modeler
EXTERNAL SOURCES
- Wikipedia - Fellegi–Sunter model
- U.S. Census Bureau - Overview of probabilistic linkage
- Springer - Record linkage methodologies
- Wikipedia - Record linkage concept overview
TOP QUESTIONS
What is the Fellegi–Sunter Model in record linkage?
It’s a probabilistic framework that classifies record pairs using likelihood ratios based on agreements and disagreements across fields.
Category: Overview
How does the model decide matches?
By computing a weight or likelihood ratio and comparing against thresholds, classifying pairs as matches, nonmatches, or possible matches.
Category: Requirements
What is blocking in the Fellegi–Sunter Model?
Blocking partitions records into candidate sets to reduce comparisons and improve performance without compromising matching quality significantly.
Category: Example
What is a common mistake in applying Fellegi–Sunter?
Ignoring field-specific error rates and using uniform weights; calibration matters to reflect real data quality and field reliability.
Category: Mistakes
How do you evaluate linkage quality?
Use precision, recall, F1, and ROC; human review of possible matches helps calibrate thresholds and validate assumptions.
Category: Management
Does the model handle missing data?
Yes, but weights should account for missingness; advanced approaches use adaptive or learned weights to mitigate bias.
Category: Opinion
What is a competing approach?
Deterministic matching uses exact rules instead of probabilities; it’s simpler but often less robust to noise and variations.
Category: Alternatives
Is the Fellegi–Sunter Model still relevant?
Absolutely. It remains foundational, with modern systems blending its principles with machine learning and active learning.
Category: Relevancy
How to start implementing Fellegi–Sunter?
Profile data, choose comparison functions, estimate m/u probabilities, define thresholds, and iterate with validation feedback.
Category: Requirements