How Well Do Automated Linking Methods Perform? Lessons from U.S. Historical Data

This paper reviews the literature in historical record linkage in the U.S. and examines the performance of widely-used automated record linking algorithms in two high-quality historical datasets and one synthetic ground truth. Focusing on algorithms in current practice, our findings highlight the important effects of linking methods on data quality. We find that (1) no method (including hand-linking) consistently produces representative samples; (2) 15 to 37 percent of links chosen by prominent machine linking algorithms are identified as false links by human reviewers; and (3) these false links are systematically related to baseline sample characteristics, suggesting that machine algorithms may introduce complicated forms of bias into analyses. We find that prominent linking algorithms attenuate estimates of the intergenerational income elasticity by up to 20 percent and common variations in algorithm choices result in greater attenuation. These results recommend that current practice could be improved by placing more emphasis on reducing false links and less emphasis on increasing match rates. We conclude with constructive suggestions for reducing linking errors and directions for future research.