Finding John Smith: Using Extra Information for Historical Record Linkage

We introduce a new rule-based linking method for historical Census records. We augment earlier algorithms based on name, age and place of birth (Abramitzky, Boustan, Eriksson, 2012, or “basic ABE”), with five matching characteristics – middle initial, county of residence, and spouse and parents’ names. Relative to basic ABE, ABE-Extra Information (“ABE-EI”) greatly increases match rates, improves accuracy and is similarly representative of the population on most attributes, with geographic mobility being one important exception. Relative to machine learning algorithms, ABE-EI has somewhat lower match rates, greater representativeness, and offers full replicability. We also create the first ABE-based links for women.

Duke Scholars

Author Hannah M Postel Sanford School of Public Policy