American cities are defined by isolated rings and pockets characterized by limited socio-economic mixing

TL;DR

The study analyzes experienced segregation in US cities using GPS-based mobility data to capture how people from different socio-economic and racial groups intersect in daily life. It introduces two metrics, $S$ (place segregation) and $I$ (neighborhood isolation), derived from origin-destination flows and income imputation, with both metrics scaled to $[0,1]$ to quantify mixing versus segregation. The authors uncover mesoscopic urban structure, revealing rings of isolation surrounding city cores and pockets of segregation within cores, with patterns largely stable over time though affected by the COVID-19 pandemic; larger cities show linear growth in isolated populations while segregation pockets scale superlinearly near centers. They further find that race and income interact with urban form to predict these zones, and demonstrate that understanding these mesoscopic patterns can inform targeted interventions to promote opportunity and urban dynamism.

Abstract

Cities generate gains from interaction, but citizens often experience segregation as they move around the urban environment. Using GPS location data, we identify four distinct patterns of experienced segregation across US cities. Most common are affluent or poor neighborhoods where visitors lack diversity and residents have limited exposure to diversity elsewhere. Less frequent are majority-minority areas where residents must travel for diverse encounters, and wealthy urban zones with diverse visitors but where locals sort into homogeneous amenities. By clustering areas with similar mobility signatures, we uncover rings around cities and internal pockets where intergroup interaction is limited. Using a decision tree, we show that demography and location interact to create these zones. Our findings, persistent across time and prevalent across US cities, highlight the importance of considering both who is mixing and where in urban environments. Understanding the mesoscopic patterns that define experienced segregation in America illuminates neighborhood advantage and disadvantage, enabling interventions to foster economic opportunity and urban dynamism.

Figures (6)

• Figure 1: Illustration of how we measure segregation and isolation. We begin with visits to places, with visitor income imputed from Census data, and then use the diversity profiles of those places to compute the exposure of neighborhoods that sent visitors to them.
• Figure 2: Measures of place segregation (S) and neighborhood isolation (I).A Place segregation, where each point represents a POI and each city is centered on downtown, shows that downtown businesses tend to see a diverse collection of visitors (and thus have low place segregation) but that businesses in surrounding neighborhoods often do not (and have high place segregation). Many of the wealthiest parts of the city, shown in black, also have fewer points of interest, which limits visitation and thus the diversity. B Neighborhood isolation is strong in those same wealthy areas with fewer POIs and also in areas with segregated POIs.
• Figure 3: The relationship between segregation and isolation.A Diversity and exposure nationally we see that integrated urban areas are often surrounded by rings of isolated suburban areas. B Locally, we see urban pockets of segregation with a range of low to high isolation, often near or surrounded by integrated urban areas. C Distributions of select variables by class show that segregated areas are often close the center, poorer than average, and nonwhite.
• Figure 4: Defining rings of isolation and pockets of segregation.A Isolation autocorrelation manifests at the national scale, delineating rings around cities. Centering and layering the cities, in B we count the number of times isolated zones occur in the same relative area: there is a clear prevalence of these zones in a ring surrounding each urban core. C Segregation autocorrelation manifests locally, with pockets appearing in large cities—less so in smaller cities like Atlanta or Boston. We also see tight scaling relationships between city size and isolation/segregated population in D, with a linear relationship between city population and the population in isolated zones along with a superlinear relationship for city population and its segregated zones.
• Figure 5: Factors associated with segregation and isolation.A Decision tree showing the defining characteristics of different classes, pruned for ease of viewing. Note on the right side that top segregated and isolated classes are the wealthiest cut, but also urban nonwhite—as indicated by density. The areas that have high segregation but low isolation tend to be urban, nonwhite and moderately dense. Partial dependence plots for B segregation and C isolation showing the joint relationship between key variables, with rings of isolation conditional on white/wealthy and pockets of segregation conditional on nonwhite/poor.