Mapping the evolution of racially mixed and segregated neighborhoods in Chicago

Abstract

The Chicago metropolitan region consists of a spatially complex mosaic of neighborhoods, in which measures of racial and ethnic composition vary dramatically. Understanding these patterns and their evolution has been hindered by ambiguities in the use of terms like ‘diverse’ or ‘segregated’, which are often posited as opposite ends of a one-dimensional scale. Using a new taxonomy of neighborhood composition, we have mapped the evolving patterns of Chicago's neighborhoods in 1990, 2000, and 2010, and tabulated census tracts that have undergone transitions or remained stable. Looking beyond the Chicago metropolitan area, we have developed an interactive atlas of similar maps for states and metropolitan areas across the United States.

Keywords:

• Chicago
• race
• segregation
• diversity
• census

1. Introduction

If the United States is a nation of migrants and immigrants, then Chicago is a quintessentially American city – its neighborhoods have been shaped and reshaped by successive waves of new arrivals. Chicago's urban landscape contains both diverse and segregated neighborhoods, sometimes side-by-side (Howenstine, 1996), a seeming paradox that leads us to consider the shortcomings of the traditional framework in which these terms have come to represent opposite ends of a one-dimensional scale. Not all low-diversity neighborhoods are alike (compare, for example, a neighborhood where whites are the majority with one where Latinos are the majority) just as not all diverse neighborhoods are the same. Collapsing the variety of real neighborhoods onto one axis obscures important differences among them. Recognizing these shortcomings, scholars have begun to describe the racial-ethnic mosaic of cities in more nuanced terms, highlighting, for instance, the evolution of multi-ethnic (e.g., Farrell & Lee, 2011) or ‘global’ neighborhoods (Logan & Zhang, 2010).

We have developed a new taxonomy of neighborhood racial composition (Holloway, Wright, & Ellis, 2012; Wright, Holloway, & Ellis, 2011) that incorporates both the degree of diversity within a neighborhood, and, for low- or moderate-diversity neighborhoods, the identity of the numerically dominant racial group. Thus, we could speak of communities as ‘black dominant and moderately diverse’ or ‘Asian dominant and not diverse’. The advantage of this approach is that it considers neighborhood segregation and diversity simultaneously.

The set of maps accompanying this paper illustrates our taxonomy of neighborhood composition, using demographic data for the Chicago metropolitan region to examine changes in patterns of segregation and diversity that have occurred over the past two decades. To encourage the adoption of this approach, we have also created an interactive atlas of segregation and diversity for the 53 largest US metropolitan areas, as well as all 50 US states (http://mixedmetro.com).

2. Methods

Our neighborhood classification method uses scaled entropy to measure racial diversity (Holloway et al., 2012). The scaled entropy E for census tract j is calculated as:

where k is an index of the racial groups. The maximum value of E _j occurs when tract j's population is evenly divided among the k racial groups. This maximum value for E _j is a function of the number of groups present; we include a scaling constant s so that E _j ranges from 0 to 1.

We then categorize neighborhoods as low, moderate, or high diversity, using the following system:

•	‘Low diversity’ tracts have scaled entropy values less than or equal to 0.3707. (Typically, one racial group constitutes over 80% of the tract's population.)
•	‘High diversity’ tracts have a scaled entropy value of at least 0.7414 and do not contain a group that constitutes more than 45% of the tract's population.
•	‘Moderate diversity’ tracts are those not falling in the other two categories.

The rationale for the use of these thresholds is discussed in detail in Holloway et al. (2012). Low and moderate levels of diversity are further identified by the numerically dominant racial group. It should be noted that the descriptor ‘dominant’ in this context does not imply social, political, or economic dominance of the neighborhood by a given racial group; it simply identifies the numerically largest group. Also, the use of these discrete thresholds means that some tracts with relatively similar racial and ethnic compositions may be assigned to different categories, if their scaled entropy values happen to fall close to but on opposite sides of one of the thresholds.

This two-dimensional classification scheme lends itself naturally to a cartographic symbolization method whereby low-diversity neighborhoods with different racial groups are represented with highly saturated colors, and moderately diverse neighborhoods are mapped in less saturated colors. This system is generally well aligned with current practices and preferences for color selection in the cartographic and data visualization communities (e.g., Allen & Turner, 2002; Brewer, 2006; Tyner, 2010). Previous research into ‘tipping points’ in neighborhood composition informed the entropy thresholds used to differentiate among levels of diversity.

Demographic data from the 1990, 2000, and 2010 decadal censuses were used as the basis for this classification system. In each census year the tract boundaries change, and the racial categories in the 1990 census differed from subsequent censuses in several ways. To facilitate comparisons across years, we resolved the 1990 and 2010 census tracts to match 2000 tract boundaries, adjusting populations based on the proportional area of overlap for partially overlapping tracts. We matched the racial groupings to correspond with the categories used in the 1990 Census and so collapsed the Hawaiian/Pacific Islander into the Asian category, and American Indian and Alaska Native into a Native American category. Because the 2000 and 2010 Censuses allowed respondents to claim more than one race (unlike the 1990 Census), we also collapsed mixed-race individuals into 1990 single-race categories using a method of proportional assignment to nonwhite categories. Thus our analysis settles on six groups: white, black, Latino, Asian, Native American, and other.

The individual map panels representing each year can be used to explore the changing spatial patterns of neighborhood composition, while transition matrices tabulate the numbers of tracts that changed from one category to another, or remained unchanged, between any two census years. Table 1 illustrates this with the transition matrix for the Chicago region from 1990 to 2010. In the transition matrix, the rows represent classes in the earlier year (1990, for the matrix in Table 1) while the columns represent classes in the later year (2010). The number at the intersection of a given row and column represents the number of tracts that changed from a given category in the earlier year to another category in the later year. (The major diagonal of the matrix represents tracts whose category did not change during the interval in question.) For example, the first row of this table shows that of 1065 tracts that were classified as low-diversity white dominant in 1990, fewer than half (485 tracts) remained in that category in 2010. In contrast, 317 out of 360 low-diversity black-dominant tracts from 1990 remained in 2010. Similar transition matrices for the 1990–2000 and 2000–2010 intervals are available on the project's website.

Table 1. Transition matrix for the Chicago metropolitan area: 1990–2010.

The interactive web-based atlas uses a color scheme similar to the one employed for the set of Chicago maps accompanying this paper. For the online maps, the user can select from several alternate base map layers to provide spatial context, and then adjust the opacity of the neighborhood racial composition map layers as desired. For this paper and its maps, spatial data layers representing the transportation network, and rivers, lakes, and other water bodies, were obtained from the US Geological Survey and included in the maps to provide context.

3. Conclusions

The dominant trend in the Chicago region has been its transformation from a heterogeneous urban core surrounded by low-diversity white neighborhoods, into a network of more diverse sub-regions, some still white-dominated but others being reshaped by newcomers and their descendants. Few types of neighborhoods did not undergo substantial change over this time period; the exception being low-diversity, black-dominated tracts, whose numbers and locations barely changed during the past 20 years. The proliferation of Latino-dominated tracts outside the urban core is obvious and striking, but moderately diverse tracts with predominantly white or black populations also roughly doubled in number. Urban geographers recognize the importance of these increasingly diverse tracts, often neighborhoods where the racial and ethnic composition is in flux, as places where residents' circumstances and views are shaped by social diversity. Chicago presents a particularly interesting laboratory for exploration of this process (e.g., Berrey, 2005; Maly, 2000; Talen, 2010; Sandoval, 2011), due to its complex neighborhood mosaic. We hope that the maps, transition matrices, and discussions resulting from our analysis of neighborhood composition and change will promote interest in, and understanding of, the spatial manifestation of diversity in Chicago's metropolitan area, and in other metropolitan regions across the United States.

Software

The classification of census tracts into categories of segregation and diversity was accomplished using Stata, while the transition matrices were produced using Microsoft Excel. Spatial analysis, preparation of spatial data layers, and map design were performed using ESRI ArcGIS v.10. To create the web-based maps, GMapCreator (from the University College of London's Centre for Advanced Spatial Analysis [UCL-CASA]) was employed to rasterize and tile the vector data layers. Finally, custom software was developed at Dartmouth College to provide a visual user interface for the online maps using the Google Maps Javascript application programming interface (API).

Acknowledgements

Sandy Wong, Minal Caron, and Kevin Mwenda at Dartmouth College contributed to the development of the maps and transition matrices, along with Michael Wellman at the University of Georgia. Grant support from the National Science Foundation, the Russell Sage Foundation, and the Neukom Institute at Dartmouth College made these analyses, and the development of the online interactive atlas, possible.