Lost in TIGER Translation: Accounting for Accuracy Improvements When Modeling Boundary Changes Across Censuses

Keywords: census, population, spatial mismatch, spatiotemporal, modeling, interpolation, neighborhood change
Abstract Type: Paper Abstract

Abstract

To help analysts determine geographic correspondences in census data across time, the U.S. Census Bureau provides relationship files that summarize associations between the geographic units used in different census years. The Bureau’s TIGER/Line files also supply boundary definitions for both current and previous units, enabling precise identification of the areas involved in changes. A major limitation of these resources is that they each span only two consecutive censuses at most. To determine correspondences across a longer period (e.g., 1990 to 2020), the simplest approaches are to join multiple relationship files sequentially or to overlay boundaries from different TIGER/Line vintages. Unfortunately, both approaches yield many false associations. Overlaying TIGER/Line boundaries is more exact in principle, but the differences between TIGER/Line vintages represent both real boundary changes and improvements in accuracy, and it can be impossible to determine which are which. The IPUMS National Historical Geographic Information System (NHGIS) provides crosswalks and time series that support comparisons across censuses for areas with fixed boundaries. I present how we account for TIGER/Line accuracy improvements when building NHGIS correspondence models, eliminating all determinably false associations and gleaning what we can from the remaining associations through a process of “balanced overlay.”