A New Strategy for Linking Census Data: A Case Study Linking the 1900 and 1910 Full-Count US Censuses

Jonas Helgertz, University of Minnesota/Lund University
Jacob Wellington, University of Minnesota
Kelly Thompson, University of Minnesota
Joseph P. Price, Brigham Young University

This paper aims to further advance the field of probabilistic methods of record linkage, focusing on how to achieve better results when linking historical censuses. We will show how additional information, frequently readily available in the very sources that one is trying to link, could – and should – be used in order to 1) increase the share of the underlying population that the linking algorithm is able to successfully link (confirmed matches), and 2) increase the quality of confirmed matches. We aim to elevate the discussion about the quality of training data that can be attained relying only on individual level information compared to the approach that we propose, as well as the usefulness of typically used metrics of the performance of any linking algorithm when based on sub-optimal training data, again comparing this to our approach.

No extended abstract or paper available

 Presented in Session 131. Evaluating Record Linkage Methods