Reconsidering Network-Distance Effects in Spatial Segregation Measurement

Abstract

Recent advances in spatial data and routing algorithms allow researchers to define the areas surrounding individuals using network distance rather than Euclidean distance. These differently constructed local environments then serve as inputs to spatial segregation measures, with network-based environments often producing higher segregation values than comparable Euclidean-based environments. These differences are frequently interpreted as evidence that mobility barriers embedded in the built environment shape patterns of racial isolation. However, distance-based local environments constructed over road networks typically contain fewer people than those constructed using Euclidean distance, raising the possibility that observed differences may reflect variation in the population composition of local environments rather than mobility constraints alone. This methodological study evaluates that possibility by constructing spatial segregation measures from both distance-based and population-based (k-nearest-neighbor) local environments defined using network and Euclidean distance. Whereas distance-based local environments built using network distance systematically include fewer people than those constructed using Euclidean distance, population-based local environments hold the size of local populations constant across measurement approaches. Holding constant the population of local environments reveals that differences between segregation measures based on network and Euclidean distance are substantially reduced across most U.S. places, although meaningful differences remain in some places. These findings suggest that previously documented differences between network- and Euclidean-based spatial segregation measures may partly reflect differences in the number of people included in local environments due to network-constrained reach.