[Paper Review] Cluster-based name embeddings reduce ethnic disparities in record linkage quality under realistic name corruption: evidence from the North Carolina Voter Registry

Joseph Lam, Mario Cortina-Borja+3 more|arXiv (Cornell University)|Jan 12, 2026

Data Quality and Management0 citations

TL;DR

The paper shows that cluster-based forename embeddings combined with TF-adjusted surname comparison reduces under-linkage disparities by ethnicity in record linkage under realistic name corruptions, using NC Voter Registry data.

ABSTRACT

Differential ethnic-based record linkage errors can bias epidemiologic estimates. Prior evidence often conflates heterogeneity in error mechanisms with unequal exposure to error. Using snapshots of the North Carolina Voter Registry (Oct 2011-Oct 2022), we derived empirical name-discrepancy profiles to parameterise realistic corruptions. From an Oct 2022 extract (n=848,566), we generated five replicate corrupted datasets under three settings that separately varied mechanism heterogeneity and exposure inequality, and linked records back to originals using unadjusted Jaro-Winkler, Term Frequency (TF)-adjusted Jaro-Winkler, and a cluster-based forename-embedding comparator combined with TF-adjusted surname comparison. We evaluated false match rate (FMR), missed match rate (MMR) and white-centric disparities. At a fixed MMR near 0.20, overall error rates and ethnic disparities diverged substantially by model under disproportionate exposure to corruption. Term-frequency (TF)-adjusted Jaro-Winkler achieved very low overall FMR (0.55% (95% CI 0.54-0.57)) at overall MMR 20.34% (20.30-20.39), but large white-centric under-linkage disparities persisted: Hispanic voters had 36.3% (36.1-36.6) and Non-Hispanic Black voters 8.6% (8.6-8.7) higher FMRs compared to Non-Hispanic White groups. Relative to unadjusted string similarity, TF adjustment reduced these disparities (Hispanic: +60.4% (60.1-60.7) to +36.3%; Black: +13.1% (13.0-13.2) to +8.6%). The cluster-based forename-embedding model reduced missed-match disparities further (Hispanic: +10.2% (9.8-10.3); Black: +0.6% (0.4-0.7)), but at a cost of increasing overall FMR (4.28% (4.22-4.35)) at the same threshold. Unequal exposure to identifier error drove substantially larger disparities than mechanism heterogeneity alone; cluster-based embeddings markedly narrowed under-linkage disparities beyond TF adjustment.

Motivation & Objective

• Motivate and quantify differential ethnic-based record linkage errors in epidemiologic estimates.
• Parameterize realistic name corruptions using empirical discrepancy profiles from NC Voter Registry data.
• Compare unadjusted, TF-adjusted, and cluster-based embedding linkage methods across corruption scenarios.
• Assess how unequal exposure to errors contributes to disparities in linkage quality.

Proposed method

• Construct empirical name-discrepancy profiles from NC Voter Registry snapshots (Oct 2011-Oct 2022).
• Generate five replicate corrupted datasets from an Oct 2022 extract (n=848,566) under three settings of mechanism heterogeneity and exposure inequality.
• Link records to originals using three comparators: unadjusted Jaro-Winkler, TF-adjusted Jaro-Winkler, and cluster-based forename-embedding with TF-adjusted surname comparison.
• Evaluate false match rate (FMR), missed match rate (MMR), and white-centric disparities across methods.
• Analyze performance at fixed MMR near 0.20 to compare overall error rates and disparities.

Experimental results

Research questions

• RQ1 How do different name-linkage algorithms perform under realistic name corruption with varying exposure to errors?
• RQ2 Do cluster-based forename embeddings reduce ethnic disparities in false and missed matches compared to string similarity alone?
• RQ3 What is the impact of unequal exposure to identifier error on disparities versus mechanism heterogeneity?
• RQ4 Can TF-adjusted similarity measures and cluster-based embeddings narrow white-centric under-linkage disparities while considering overall error trade-offs?

Key findings

• TF-adjusted Jaro-Winkler achieves very low overall FMR (~0.55%) at MMR ~20.34%, but white-centric disparities remain substantial.
• Hispanic and Non-Hispanic Black groups exhibit higher FMRs than Non-Hispanic White groups under unadjusted comparisons; disparities are reduced by TF adjustment.
• Cluster-based forename-embedding with TF-adjusted surname comparison further reduces missed-match disparities (Hispanic +10.2% vs +36.3%; Black +0.6% vs +8.6%), but increases overall FMR to 4.28%.
• At the same threshold, unequal exposure to corruption drives larger disparities than mechanism heterogeneity alone, and cluster-based embeddings markedly narrow under-linkage disparities beyond TF adjustment.