Year: 2026
Type: article
Abstract: Linking education and health records can offer critical insights into how policies and interventions shape children’s outcomes, yet these data linkage efforts are often constrained by inconsistent identifiers, incomplete population overlap, and variable data quality. We develop a supervised machine learning framework to link New York City Department of Education (DOE) and Medicaid records, one of the first large-scale linkages of its kind in the United States. Our two-step approach first deduplicates records within each dataset and then applies a random forest model to identify cross-dataset matches, integrating blocking, probabilistic scoring, and clustering to improve upon deterministic and off-the-shelf probabilistic linkage techniques while maintaining computational efficiency. Applying this framework, we achieve a high quality linkage, identifying matches for approximately 71% of records across the two systems. Match rates are consistent across birth cohorts and most demographic groups, though lower for some populations because the two systems capture partially overlapping subsets of children. Rather than optimising for a 100% match rate, we propose a framework for evaluating linkage quality under uncertainty, emphasising precision, transparency, and population representativeness. This approach provides scalable guidance for linking complex administrative systems and enables new opportunities to study how educational and health policies interact.
Language: English
Topic: Data Quality and Management
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,747
OpenAlex ID: W7221583360
