An empirical comparison of record linkage procedures

Area of Science:

• Data science
• Biostatistics
• Database management

Background:

• Record linkage often relies on non-unique identifiers like names and demographics.
• Probabilistic record linkage methods are well-documented, but deterministic strategies are less explored.
• Comparisons between deterministic and probabilistic linkage results are scarce in literature.

Purpose of the Study:

• To compare a stepwise deterministic record linkage strategy with a probabilistic strategy (AUTOMATCH).
• To evaluate performance using known true matches derived from Social Security Numbers.
• To analyze match rates and error rates for both linkage strategies.

Main Methods:

• Implemented a stepwise deterministic linkage strategy.
• Utilized the AUTOMATCH software for probabilistic record linkage.
• Linked medical records (Regional Perinatal Intensive Care Centers) with educational records (Florida Department of Education).
• Used Social Security Numbers to establish ground truth for record pairs.

Main Results:

• The stepwise deterministic strategy achieved performance comparable to the probabilistic AUTOMATCH strategy.
• Both methods were evaluated on match rates and error rates.
• The study provides a detailed comparison of strengths and weaknesses for each approach.

Conclusions:

• Deterministic record linkage can be a viable alternative to probabilistic methods when dealing with non-unique identifiers.
• The choice between deterministic and probabilistic strategies depends on specific data characteristics and linkage goals.
• Further research into deterministic linkage strategies is warranted.