Identifying Spatial Bias in Genealogical and Census Record Linkage

Caglar Koylu, University of Iowa
Jonas Helgertz, University of Minnesota/Lund University
Alice Bee Kasakoff, University of South Carolina
Evan Roberts, University of Minnesota
Maryam Torkashvand, University of Iowa
Loretta Nwajiaku, University of Iowa

Record linkage has enabled historical demographic research on migration, kinship networks, and intergenerational processes by connecting individuals across censuses and other sources such as crowdsourced family trees. Yet linked datasets rarely represent the historical population uniformly, especially geographically. This paper focuses on spatial bias in the linkage of crowdsourced family trees to the 1880 full count U.S. census, where the geographic structure of linkage success remains poorly understood. Prior work in record linkage generally shows that migrants, urban residents, and populations in more mobile regions are more difficult to link, while individuals in stable households and rural areas are more likely to appear in linked samples. We utilize a set of spatial statistics to evaluate geographic bias in linkage outcomes. We measure county level linkage bias using representation ratios (location quotients) that compare each area’s share of linked records with its share of the underlying census population. We map these ratios to identify overrepresented and underrepresented areas and test their spatial structure using Global Moran’s I to detect overall spatial clustering and Local Moran’s I to identify local clusters and spatial outliers. In addition, we evaluate several sources of geographic bias identified in prior research, including the underrepresentation of migrants, differences between rural and urban populations, regional variation in linkage rates, and biases associated with residential stability and household continuity. To demonstrate the broader applicability of this approach, we apply the same workflow to the IPUMS Multigenerational Longitudinal Panel linkage between the 1870 and 1880 censuses. Comparing the spatial structure of linkage outcomes across these two projects shows how geographic bias can emerge in different linkage settings. The results highlight the importance of diagnosing spatial bias when using linked historical data to study migration and long run demographic processes.

No extended abstract or paper available

 Presented in Session 144. Beyond the Census II