Residential Segregation by Race and Ethnicity and the Changing Geography of Neighborhood Poverty

Abstract

Racial and ethnic segregation in U.S. metropolitan areas contributes to the existence of neighborhood poverty, with segregation typically conceptualized as occurring between central city and suburban neighborhoods due to the racially exclusive nature of suburbanization through much of the twentieth century. However, increasing suburbanization across race, ethnicity and socioeconomic status since around 1970 has complicated the spatial structure of residential inequalities among metropolitan areas. In this study, I assess how patterns of racial and ethnic inequality in exposure to neighborhood poverty changed across urban and suburban locations since 1980, and I investigate how different dimensions of segregation by race and ethnicity correspond to worsened racial and ethnic inequality in exposure to disadvantaged neighborhoods in urban as well as suburban areas. To study differences between suburbs, I contribute a novel approach for measuring suburban neighborhoods based on their density, housing stock age and overall degree of development. Results demonstrate how the conventional city-suburb dichotomy masks substantial differences between suburbs in (a) Black, Latino and White exposure to neighborhood disadvantage and (b) the degree to which patterns of segregation by race and ethnicity exacerbate Black–White and Latino–White inequalities in exposure to suburban neighborhood disadvantage. Black–White segregation exacerbates Black exposure to neighborhood poverty across space, especially in cities and older suburbs, while Latino–White segregation worsens Latino exposure to poor neighborhoods in cities as well as farther-flung rural and exurban areas.

Appendices

Appendix 1: Additional Information About Hierarchical Clustering with Spatial Constraints

Using a spatial weights matrix that indicates which tracts are contiguous to other tracts, hierarchical clustering with spatial constraints seeks to maximize the homogeneity within clusters (on the sociodemographic measures) while limiting the solutions to those that maximize spatial homogeneity (i.e. contiguity of units into regions). The net result of this approach is that the spatial constraint reduces bumpiness or “checkerboarding” across space while preserving the derived clustering’s average composition on the sociodemographic measures of interest.

After matrices of distances in terms of sociodemographic measures and actual space are computed, the next step is to use an algorithm that uses a specific decision rule for determining how to cluster each case from the “bottom up”. The method of hierarchical clustering with spatial constraints introduced by Chavent et al. (2018) uses Ward’s method for this process. With each unit starting as its own cluster at first, Ward’s method then iteratively combines two existing clusters based on merging the pair that results in the smallest increase in the resulting clusters’ sum of squared variance. This method is useful for the present case because these clusters are accordingly compact (low variance) across the sociodemographic covariates of interest, in addition to being compact in terms of forming contiguous regions.

Ward’s method requires a single matrix of distances, so hierarchical clustering with spatial constraints accordingly uses a mixing parameter $\lambda$ to determine the relative contribution of sociodemographic and spatial distances used with Ward’s method. Estimating an optimal $\lambda$ value requires searching for the value with the greatest improvement in spatial homogeneity relative to lost sociodemographic homogeneity. I extract four clusters from the resulting dendrogram that Ward’s method produces corresponding to city, older suburb, newer suburb and rural/exurb locations.

Appendix 2: Housing Costs by Location

Distinguishing different types of locations in metropolitan area using characteristics of housing should have implications for the cost of renting or owning a housing unit in each area. For example, newer suburbs, by virtue of having a less dense, more recently constructed housing stock are expected to cost more to live in than city or older suburb neighborhoods, where units will be smaller and older, on average. To corroborate this expectation, I estimate models of logged median contract rent and value for neighborhoods according to the 2014–2018 American Community Survey for each of the four locations, with city neighborhoods set as the reference.

Table 4 presents coefficients from these log-linear models, with the coefficients for suburbs showing relative difference between the three suburban location types compared to city neighborhoods. These models indicate that median rents and values in newer suburbs ($exp(\hat{y_{rent}}) = \$1155$, $exp(\hat{y_{value}}) = \$262{,}268$) are about 28% greater than the average city neighborhood ($exp(\hat{y_{rent}}) = \$897$, $exp(\hat{y_{value}}) = \$203{,}975$), while older suburbs fall in between these locations in terms of housing cost ($exp(\hat{y_{rent}}) = \$1024$, $exp(\hat{y_{value}}) = \$240{,}247$). Rural/exurb neighborhoods have the lowest typical cost among the four types of locations ($exp(\hat{y_{rent}}) = \$797$, $exp(\hat{y_{value}}) = \$195{,}048$), with households trading proximity to valued amenities and job centers for more space per dollar.

Table 4 Log-linear models of 2014–2018 ACS median contract rent and value by location type

Appendix 3: Linear Regressions of Black, Latino and White Neighborhood Poverty Exposure Across Locations, 1980 to the 2014–2018 ACS

See Tables 5.

Table 5 Linear models of Black, Latino and White neighborhood poverty exposure rates, 1980 to the 2014–2018 ACS

Appendix 4: Fixed Effect Models of Black, Latino and White Neighborhood Poverty Exposure Across Locations and Metropolitan Segregation, 1980 to the 2014–2018 ACS

See Tables 6, 7, 8, 9, 10 and 11.

Table 6 Fixed effect models of Black and White neighborhood poverty exposure rates, measuring segregation with dissimilarity index

Table 7 Fixed effect models of Black and White neighborhood poverty exposure rates, measuring segregation with isolation index

Table 8 Fixed effect models of Black and White neighborhood poverty exposure rates, measuring segregation with spatial proximity index

Table 9 Fixed effect models of Latino and White neighborhood poverty exposure rates, measuring segregation with dissimilarity index

Table 10 Fixed effect models of Latino and White neighborhood poverty exposure rates, measuring segregation with isolation index

Table 11 Fixed effect models of Latino and White neighborhood poverty exposure rates, measuring segregation with spatial proximity index