A Theory for Record Linkage.¶
Fellegi, I. P., & Sunter, A. B. (1969). A Theory for Record Linkage. Journal of the American Statistical Association, 64(328), 1183-1210.
Cited by¶
14 citations across 14 artifacts.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Binding Problem
- Data engineering: joining records across sources where the same entity carries different identifiers — the entity-resolution problem — is the binding problem for tabular data, with foreign keys, surrogate keys, and probabilistic record linkage as binding mechanisms.
This sourceFoundational theory of probabilistic record linkage / entity resolution — binding records that refer to the same entity across sources lacking a shared key.
Supported in partVerified against the publisher's abstract
The abstract supplies the probabilistic record-linkage principle — deciding whether records in two files denote the same entity — but says nothing about foreign keys, surrogate keys, or tabular-data binding.
“A mathematical model is developed to provide a theoretical framework for a computer-oriented solution to the problem of recognizing those records in two files which represent identical persons, objects or events (said to be matched).”
- Data engineering: joining records across sources where the same entity carries different identifiers — the entity-resolution problem — is the binding problem for tabular data, with foreign keys, surrogate keys, and probabilistic record linkage as binding mechanisms.
- Complexity
- Rather than treating it as straightforward technical merger, the team analyzes through multiple complexity lenses: (a) computational complexity—data-reconciliation algorithms (matching patient identities across systems with typos, missing fields, duplicates) are formally hard (record-linkage variant); they choose probabilistic matching with human review rather than exact algorithms
This sourceProbabilistic decision-theoretic model for matching records representing the same entity across files with noisy/incomplete data — supports the probabilistic-matching claim at D26-085. Verified existing and supporting. (See flag: prose asserts record linkage is 'NP-hard', which this statistical model does not establish.)
Supported in partVerified against the publisher's abstract
The abstract states a probabilistic decision-theoretic record-linkage model with link/non-link/possible-link decisions, backing the probabilistic-matching half of the claim but not the assertion that reconciliation is formally hard.
“A mathematical model is developed to provide a theoretical framework for a computer-oriented solution to the problem of recognizing those records in two files which represent identical persons, objects or events (said to be matched).”
- Rather than treating it as straightforward technical merger, the team analyzes through multiple complexity lenses: (a) computational complexity—data-reconciliation algorithms (matching patient identities across systems with typos, missing fields, duplicates) are formally hard (record-linkage variant); they choose probabilistic matching with human review rather than exact algorithms
- Criteria of Individuation
- The co-reference criterion is the master-patient-index matching rule — deterministic on a national identifier where present, probabilistic on name, date of birth, and address where not — resolving which presentations are the same patient, with transitive closure as its equivalence regime (so one over-permissive pairwise match can cascade unrelated records into a phantom patient).
This sourceFoundational probabilistic record-linkage theory underlying entity resolution / master-data matching with its false-merge and false-split error trade-offs.
Supported in partVerified against the publisher's abstract
Fellegi-Sunter 1969 supplies the probabilistic record-linkage frame and its false-link/false-non-link error trade-off, but not the deterministic national-identifier rule, the transitive-closure regime, or the phantom-patient cascade.
“A mathematical model is developed to provide a theoretical framework for a computer-oriented solution to the problem of recognizing those records in two files which represent identical persons, objects or events (said to be matched).”
- The co-reference criterion is the master-patient-index matching rule — deterministic on a national identifier where present, probabilistic on name, date of birth, and address where not — resolving which presentations are the same patient, with transitive closure as its equivalence regime (so one over-permissive pairwise match can cascade unrelated records into a phantom patient).
- Equivalence Relation
- Data engineering develops the most operationally significant industrial application of equivalence-relation reasoning in the form of record linkage and master data management, drawing on the probabilistic linkage framework of Fellegi and Sunter (1969).
This sourceFounds probabilistic record linkage: assigns match/non-match probabilities to record pairs from similarity-feature vectors; the match decisions yield (after transitive closure) an equivalence relation whose classes are master entities.
Supported in partVerified against the publisher's abstract
Supplies the probabilistic record-linkage decision framework (link/non-link/possible-link, optimal linkage rule) but says nothing about equivalence-relation reasoning, master data management, or the claim's superlative ranking.
“A mathematical model is developed to provide a theoretical framework for a computer-oriented solution to the problem of recognizing those records in two files which represent identical persons, objects or events (said to be matched).”
- Data engineering develops the most operationally significant industrial application of equivalence-relation reasoning in the form of record linkage and master data management, drawing on the probabilistic linkage framework of Fellegi and Sunter (1969).
- Identity Test
- A quantitative test (do these records match on enough fields?) and a qualitative test (do these records refer to the same person regardless of data?) give different answers on the same dataset and pursue different goals — a tax authority and a marketing department deduplicate the same customer database with different tests and produce different counts, each correct for its purpose.
This sourceFoundational probabilistic record-linkage framework (developed for census deduplication) with explicit false-match/false-nonmatch rates — the tunable identity test later ported into epidemiology and master-patient-index design.
Supported in partVerified against the publisher's abstract
Fellegi & Sunter's abstract fixes the probabilistic frame (link/possible-link/non-link decisions with false-match and false-nonmatch rates stipulated in advance) but not the qualitative-test contrast or the tax-authority/marketing example.
“A comparison is to be made between the recorded characteristics and values in two records (one from each file) and a decision made as to whether or not the members of the comparison-pair represent the same person or event, or whether there is insufficient evidence to justify either of these decisions at stipulated levels of error.”
- A quantitative test (do these records match on enough fields?) and a qualitative test (do these records refer to the same person regardless of data?) give different answers on the same dataset and pursue different goals — a tax authority and a marketing department deduplicate the same customer database with different tests and produce different counts, each correct for its purpose.
Mechanisms¶
- Administrative Record Linkage
- A false match silently fuses two units into one, and a missed match splits one unit into two — the probabilistic-matching literature frames this as the trade-off between false links and false non-links, and both corrupt the count in opposite directions.
This sourceFormalizes record linkage as a decision problem balancing false links against false non-links, errors that respectively merge distinct records or fail to join records belonging together.
- A false match silently fuses two units into one, and a missed match splits one unit into two — the probabilistic-matching literature frames this as the trade-off between false links and false non-links, and both corrupt the count in opposite directions.
- Canonical Identity Resolution Pass
- The formal backbone here is the Fellegi–Sunter model of probabilistic record linkage, which frames matching as a likelihood decision with explicit error rates rather than a yes/no guess.
This sourceFormulates the Fellegi-Sunter probabilistic record-linkage model as likelihood-ratio classification at stipulated error levels.
- The formal backbone here is the Fellegi–Sunter model of probabilistic record linkage, which frames matching as a likelihood decision with explicit error rates rather than a yes/no guess.
- Collision Detection Review
- This is the standard practice of record linkage, the discipline of deciding when two records describe one entity.
This sourceDefines record linkage as deciding whether a pair of records represents the same underlying person, object, or event.
- This is the standard practice of record linkage, the discipline of deciding when two records describe one entity.
- Deduplication Pass
- The problem it solves is a well-studied one; probabilistic record linkage gives it a principled basis
This sourceDevelops a probabilistic decision framework for the established problem of linking records that refer to the same entity.
- The problem it solves is a well-studied one; probabilistic record linkage gives it a principled basis
- Duplicate Resolution Queue
- This is the false-match/false-non-match trade-off, and pretending a single threshold optimizes both is the classic error.
This sourceDocuments the false-match/false-non-match trade-off; pretending a single threshold optimizes both is the classic error.
- This is the false-match/false-non-match trade-off, and pretending a single threshold optimizes both is the classic error.
- Entity Resolution Policy
- Statistical record linkage from the Fellegi–Sunter model onward gives principled match probabilities, but a probability is an input to the rule, not a license to skip it.
This sourceDevelops a probabilistic record-linkage decision rule in which comparison-pattern probabilities are inputs to link, non-link, and possible-link classifications.
- Statistical record linkage from the Fellegi–Sunter model onward gives principled match probabilities, but a probability is an input to the rule, not a license to skip it.
- Identity-Key Normalization
- It is applied record linkage: the discipline of deciding when records from different sources refer to the same real-world entity.
This sourceRecord linkage decides whether records from different files represent the same real-world person, object, or event.
- It is applied record linkage: the discipline of deciding when records from different sources refer to the same real-world entity.
- Identity Resolution Model
- It weighs each field by how discriminating it is — agreeing on a rare surname counts far more than agreeing on a common one.
This sourceWeights record fields by match-versus-nonmatch discrimination, so agreement on an uncommon surname receives more weight than agreement on a common surname.
- It weighs each field by how discriminating it is — agreeing on a rare surname counts far more than agreeing on a common one.
- Lineage or Panel Correspondence Matrix
- The craft is entirely in the matching: it is a record-linkage problem, resolving which records across two snapshots refer to the same underlying entity.
This sourceDefines record linkage as deciding which records in different files refer to the same underlying entity.
- The craft is entirely in the matching: it is a record-linkage problem, resolving which records across two snapshots refer to the same underlying entity.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Read against the text for 5 of 14 citations: 5 supported in part. Each verdict is shown under its citation below, with what in the work backs the sentence.
Was it audited? Yes. A second, independent pass read the citation against the article text and recorded a verdict.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Registry ID ref:56e8b18fa210 · see in the full table