Professional Documents
Culture Documents
3176131
3176131
Copyright 2005 by
Yang, Rui
Rui Yang
by
DISSERTATION
in Partial Fulfillment
of the Requirements
August, 2005
ACKNOWLEDGEMENTS
Jargowsky, for his guidance, support, and encouragement throughout the entire phase of this
study. I am especially grateful for his constructive comments and suggestions on early drafts
and for the numerous hours that he dedicated to this project. I also would like to express my
dissertation: Dr. Brian J. L. Berry, Dr. Ronald Briggs, and Dr. Roxanne Ezzet-Lofstrom.
Besides, I would like to thank Dr. Richard Scotch and Jeongdai Kim for their insightful
comments on early drafts of this study. I am also sincerely grateful for Dr. Irving Hoch for his
inspiring encouragements.
Additionally, I would like to thank Dr. Euel Elliott and Judy Robertson for their
professor who taught me; I am sincerely thankful for the knowledge and skills I acquired in all
my graduate courses. I also would like to thank Cristen Casey, Christina Spicher, and every
staff in the International Student Office who make my life as an international student in this
Finally, I wish to express my sincere appreciation my parents Jiangping Zhu and Shaofu
Yang, and my fiancé Hong Meng Lu for their unfailing love and support. Without them, this
July 2005
v
DOES SUBURBAN DEVELOPMENT HELP PRODUCE ECONOMIC SEGREGATION
ABSTRACT
Rapid suburbanization is changing the face of urban America. Despite voluminous literature on
the relationship between suburbanization and economic segregation, whether or not the current
suburban development patterns are significant contributors to the changes in economic
segregation remains an open empirical question. Using U.S. Census and PUMS data, this
research directly examines this empirical question by conducting a comprehensive analysis for
all metropolitan areas for the last decade. Empirical findings demonstrate economic
segregation, as measured by the Neighborhood Sorting Index, decreased significantly for all
racial and ethnic groups during the 1990s, reversing the earlier increasing trend from 1970 to
1990. Instead of constructing a comprehensive definition or a composite index, this research
identifies six specific indicators to characterize the contemporary suburbanization patterns and
tests their effects on economic segregation by multiple series of cross-sectional regressions and
first-difference, fixed-effects estimations. These indicators of suburbanization are: the urban
density gradient, the population density, the relatively rapid growth of the periphery, the
homogeneity of new growth, the exclusivity of local zoning, and the inaccessibility of jobs. The
empirical results essentially suggested that as a group, the contemporary suburbanization
patterns made a difference on the changes in economic segregation, although the individual
effect of each indicator was not robust.
vi
TABLE OF CONTENTS
Number Page
Acknowledgements..............................................................................................................v
Abstract...............................................................................................................................vi
List of Figures.....................................................................................................................ix
CHAPTER 9 CONCLUSION........................................................................................131
vii
APPENDIX D THE 2ND SET OF CROSS-SECTIONAL MODELS (overall pooled) .142
Vita
viii
LIST OF FIGURES
Number Page
Figure 6-1. Neighborhood Sorting Index (NSI) by racial/ethnic groups, 1970-2000 .......73
ix
LIST OF TABLES
Number Page
Table 5-5. Mean household income and its percentage change ........................................65
Table 6-5. Descriptive statistics of pooled NSI and pooled changes in NSI.....................79
Table 7-5. The gross population density by census division and metropolitan size..........88
x
Table 7-6. Mean values of the estimated central city density............................................89
Table 7-13. The homogeneity of new development by 2000 mean household income ....96
Table 7-17. The relatively rapid growth rate in suburbs compared to central cities .........98
Table 7-19. The relatively rapid growth rate by census division and metropolitan size .100
Table B-1. Correlations between the suburban indicators and log of population ...........140
xi
Table G-1. Descriptive statistics for fixed-effects estimations........................................146
xii
CHAPTER 1
INTRODUCTION
This research is part of a broader concern with understanding the influences of rapid
suburbanization process in the United States. Although the suburbanization process is quite
common in all wealthy western nations, metropolitan areas in the United States are remarkably
spread out (Musterd and Ostendorf 1998). The contemporary suburban development in this
country needs to be understood beyond just a land use aesthetic, an environmental problem, or
a fixed set of characteristics of a region’s housing stock. Instead, the discussion of this
phenomenon should devote more attention to the links to urban uneven development, racial and
economic inequality, and power and politics. Specifically, this research examines whether the
current suburban development patterns have played a significant role in changes in economic
Despite deceases in segregation by race and ethnicity that have been observed in
American cities, segregation by income increased from the 1970s to 1990s (Abramson, Tobin,
and VanderGoot 1995; Dreier, Mollenkopf, and Swanstrom 2001; Jargowsky 1996; Lucy and
Phillips 2000; Massey 1996). Over the same period, the United States experienced a significant
degree of suburbanization (Burchell et al. 1998; Jackson 1985; Kunstler 1993; Popenoe 1979;
Heinlich and Andersen 2001). These two phenomena are plausibly connected. Numerous
studies describe the contemporary suburbanization pattern as the affluent and middle-class
Americans living in suburban areas that are far from their work place and leaving the inner-city
1
2
poor residents behind (Dreier, Mollenkopf, and Swanstrom 2001; Jackson 1985; Squires 2002).
As suburban development occurs in the United States, it is linked very closely to the income
distribution (Jargowsky 2002). Furthermore, many policy makers and city planners suspect that
However, whether or not the current suburban development patterns are significant
the one hand, there may be no necessary connection between suburbanization and economic
segregation. If the new suburbs developed with a complete menu of housing types and the
benefits of these new developments were equally accessible across the economic spectrum,
suburbanization would not perpetuate economic segregation between neighborhoods nor reduce
access to opportunity; even, the rapid peripheral growth could be a source of jobs that helps
reduce poverty and inequality and a force for economic integration. On the other hand, if the
Based upon 1990 and 2000 U.S. Census data, this research directly examines this
empirical question by conducting a comprehensive analysis for all metropolitan areas across
US. Measuring economic segregation by the Neighborhood Sorting Index, this research
complements the existing literature on economic segregation by investigating the new trends of
this social problem in U.S. metropolitan areas during the last decade.
“sprawl”, has become an intriguing research topic and sparked a national debate (Staley 1999).
and constructing a comprehensive definition or a composite index, this research identifies six
uneven development perspective and relevant literature, and tests their effects on economic
segregation. By doing so, first of all, it makes the interpretation easier. Also, it allows specific
policy implications to be made. The indicators of suburbanization are the urban density
gradient, the population density, the relatively rapid growth of the periphery, the homogeneity
of new growth, the exclusivity of local zoning, and the inaccessibility of jobs. Some indicators
than others. By indicating which aspects of suburbanization (if any) are important to
understanding the changes in economic segregation, this study also fits into the debate over the
Certainly suburban development is driven by the dynamics of demand and supply in the
housing market; nevertheless, “sprawl” occurs, in part, because local governments in the
United States encourage this form of development via zoning and subdivision ordinances
which, in turn, reflect the desires of a large share of their citizenry (Burchell et al. 1998). Thus
the findings of this research have important implications for public policies regarding urban
politician’s lips – local, state, and even national (Hartman 2002), this research may help
policymakers to extend their understanding of this phenomenon and encourage them to rethink
from numerous scholars and policymakers. It has many disadvantageous consequences for the
4
society and complicates the background for making various social policies, such as welfare
policy and education policy. For instance, segregation between lower-income, central-city
concentration of poverty in the central cities. William Julius Wilson (1987) argues that people
live in neighborhoods with high levels of concentrated poverty experience a social isolation
that excludes them from job information networks. These neighborhoods are plagued not only
by unemployment, but also by crime and other self-destructive behaviors such as drug and
alcohol abuse, making the area even less attractive to outsiders. Furthermore, this community
isolation prevents many residents from developing a work ethnic. These “concentration effects”
cause neighborhoods to become almost completely cut off form the mainstream of society and
By numerous dimensions, the spatial isolation of the lower-income households not only
affects poverty and inequality in the short run, but also threatens the equality of opportunity in
the long run (Jargowsky 2002). First of all, the individual’s success in labor market is affected
the pattern of economic segregation. As most new office buildings and business centers are
being produced in the suburbs (Kasarda 1988) and new suburbs are being constructed at greater
and greater distances from the urban core, many jobs connected to this prosperous residential
base such as grocery stores, restaurants, and yard work move too. At the same time, similar job
opportunities in central cities shrink since the central-city neighborhoods concentrate more and
more lower-income households with less spending power. The divergence between the
centralized pattern of residence for the poor and the increasingly distant location of the new
jobs leads to a spatial mismatch (Dreier, Mollenkopf, and Swanstrom 2001). In an ideal world,
this type of spatial mismatch would not last long since households would simply relocate
5
the current suburbanization in this country has strengthened the barriers to movement. A well-
known example is that local jurisdictions use exclusionary zoning regulations to prevent low-
and moderate-income households from moving to the areas with the fastest growth in
employment (Helling 2002). Given constraints on housing mobility, the spatial mismatch may
make it difficult for the inner-city poor households to get suburban jobs because of the lack of
information about jobs may not reach into central-city neighborhoods that have limited social,
political, and economic connections with the suburbs (Ihlanfeldt and Sjoquist 2000).
disadvantageous neighborhoods diminish the life chances of their residents. The so-called
neighborhood’s residents and other closely related features such as family structure or
unemployment on the outcomes for specific individuals within the neighborhood, holding other
things constant (Jargowsky 2002). For instance, the “contagion” model argues that a
concentration of poor neighbors may lead people, especially teens, to emulate negative
behaviors, resulting in worse outcomes even when the influences of a person’s immediate
family and personal aspects are controlled (Crane 1991). Simultaneously, disadvantaged
neighborhoods may lack role models who could act as “social buffers” against the effects of
public institutions that would be available to assist poor residents living in better-off
neighborhoods (Jencks and Mayer 1990). In general, neighborhoods “can make a big difference
in residents’ quality of life, their ability to compete effectively in the labor market, and their
6
success in raising children” (Jargowsky 2002: 64). Especially, the isolated and impoverished
neighborhoods may have particularly strong negative impacts on adolescents who are searching
for their personal identities, forming their value systems, and shaping their life styles. More
exposure to crime, violence, drugs, child birth out of wedlock, and many other problem
behaviors increase the probability that youth living in poor neighborhoods will fall into a
geographically and school attendance zones are also imposed geographically, poor children will
also be clustered in school. Given that much school finance is local in this country, schools in
poor neighborhoods have greater needs but a lower tax base; it is harder for them to hire and
retain high-quality teachers. Farkas (1996) also notes that, in some cases, teachers who
continually deal with children from troubled families develop low expectations for students.
Given the importance of human capital in the modern economy, lower-quality education for
children from lower-income households may perpetuate poverty and will likely to set the stage
While the spatial separation of income groups compounds poverty and contributes
towards the growth of inequality, it also reduces public support for policies that may address
this cycle (Jargowsky 2002). The governmental fragmentation in suburban areas has continued
since the 1960s (Jargowsky 2002). Highly differentiated local governments can provide a wide
variety of choices. The Tiebout Hypothesis (1956) indicates that individual families will choose
the local jurisdiction that provides the optimal package of public services given the housing
cost and tax burden. Actually, during the decision-making process, better-off families often
7
tend to place a geographical distance between themselves and lower-income households (Park
1926). As wealthier segments of society no longer share a common local government with
other income groups, the ability to carry out large-scale changes in public policy may diminish
(Jargowsky 2002).
The organization of the study is as follows: Chapter 2 reviews major studies on racial
and economic segregation. The methods and measures used to study segregation were
substantive way to segregation by income. Thus this chapter first reviews the works focusing
on racial segregation, then discusses the literature on economic segregation, and finally
suburbanization patterns in the literature. Then, based upon the uneven development
perspective, this chapter identifies the urban density gradient, the population density, the
relatively rapid growth of the periphery, the homogeneity of new growth, the exclusivity of
local zoning, and the inaccessibility of jobs as six key indicators of suburbanization, and
discusses the theoretical relationship between each of them and economic segregation.
characteristics, and social distance and inequality. It discusses the theoretical relationships
between economic segregation and the variables characterizing metropolitan aspects other than
Chapter 5 focuses on the actual data sources: the 1990 and 2000 US Census summary
files and 5 percent Public Use Microdata Samples (PUMS) data, the unit of analysis, and the
Sorting Index), suburbanization indicators, and the variables controlling for metropolitan
context, structural economic characteristics, and social distance and inequality. It ends by
The empirical patterns of economic segregation in the 1990s are documented in Chapter
6. Using 1990 and 2000 Census and PUMS data, this chapter finds economic segregation, as
measured by the Neighborhood Sorting Index (NSI), decreased significantly over the last
decade, reversing the historical trend of increasing income segregation from 1970 to 1990. This
chapter also explores the potential causes of this trend, and illustrates the basic descriptive
statistics and histograms of the pooled NSI and the pooled changes in NSI, which serve as
respectively.
Regarding the suburbanization patterns in the 1990s, Chapter 7 discusses the basic
descriptive statistics and regional patterns of each suburbanization indicator, and illustrates how
all of them vary with the size of metropolitan areas. Of particular concern are the different
trends of gross population density and estimated central city density over the ten years, and the
comparison between the two alternative proxies of local zoning regimes: the total number of
local governments and the number of local governments per 100,000 households.
Results from three series of regression analyses are presented in Chapter 8. They are
presented in two sections: the first section focuses on two sets of cross-sectional models and the
9
second one discusses the fixed-effects estimations. Each section elaborates the estimation
procedures, variable descriptions, estimation equations, and empirical results from these
suburbanization indicators from the fixed-effects models are more consistent and robust,
of policy implications. Existing limitations in this research and potential improvements for
From a certain perspective, the rich studies of socioeconomic differentiation within the
economic inequality, and economic segregation. Although the economic inequality problem is
well documented, this study pays little attention to this literature since it is not a focal point
here compared with segregation. Worth noting, however, two studies by the US Bureau of
Census (1992; 1993) found that the economic inequality in US had been increasing since 1979.
The following Sections address three related topics. Although the primary concern in this study
is economic segregation, it begins with a discussion of racial segregation for two reasons. First,
the methods and measures used to study segregation were developed in this context. Second,
racial segregation contributes in a substantive way to segregation by income. Thus, Section 2.1
first summarizes the conceptual and empirical work focusing on racial segregation, followed by
a discussion about conceptual and empirical literature on economic segregation in Section 2.2.
Finally, Section 2.3 reviews the theoretical and empirical studies regarding the interrelationship
based upon physical separation. The physical barriers and spatial distance that set individuals
apart from the mass and group from group are sociologically significant only to the extent that
10
11
they became obstacles to communication and inhibit social contact (Wirth 1954). The major
use of this term in the social sciences for some years was to denote the ecological process by
which people settle or locate in those areas of a community occupied by people of similar
social characteristics or activities (Bain 1969). Race, culture, religion, moral, status, language,
caste, class, wealth, occupation, physique and mentality, talent, taste, attitude, interest, in fact
any set of characteristics in which the members of a group may be similar to one another and
different from outsiders may become the basis of segregation (Wirth 1954). Among all these
possible bases, the residential segregation by race (or so-called racial segregation), which can
be denoted as the spatial segregation of households by race/ethnic status, is very prevalent and
Massey and Denton (1987) pointed out that conceptually there were at least five
concentration, and clustering. Mainly, they discussed the first two dimensions. Evenness is the
most important conceptual aspect of segregation. White (1986) argued that segregation in a
population was indicated by the unevenness of the distribution of its members across places or
categories. By Massey and Denton (1987), evenness is the differential distribution of minority
and majority members across census tracts within an urban area. A minority group is said to be
segregated if it is unevenly distributed over tracts. Residential evenness has typically been
measured by the well-know index of dissimilarity (D). However, evenness is not the only
dimension concerning racial segregation; the second dimension is exposure, which refers to the
degree of potential contact between minority and majority members within census tracts of
urban areas. Currently, the most widely used measure of exposure is the so-called exposure
index, which is also referred as the asymmetric intergroup contact index or P*.
12
segregation by race (Massey and Denton 1987). White (1986) summarized various measures of
exposure index (also called exposure index), the Gini index, the Entropy index, Atkinson’s
Measure, Contingency table measures, analysis of variance (ANOVA) measures, and spatial
measures. The index of dissimilarity and the exposure index are the most frequently used two
in current research, and scholars presented diverse empirical results based upon them. The
Dxy = .5 ∗ ( xi X ) − ( yi Y )
where xi and y i are the numbers of X and Y members in tract i, and X and Y are their city-wide
totals (Massey and Denton 1987). This index assesses how evenly two racial/ethnic groups (e.g.
Blacks and Whites) are distributed across the urban landscape. Its numerical value indicates the
percentage of either group who would have to move from one census tract to another to
eliminate segregation, that is, to produce an index of zero, while keeping the population size of
each tract fixed (Farley 1991). It scales from 0 to 100. In a case of Black-White segregation
study, D will equal to 100 if all Blacks lived in exclusively Black neighborhoods and all Whites
in all-White neighborhoods; and it will equal to 0 if people were perfectly evenly assigned to
The exposure index estimates the probability of residential contact between groups X
and Y as:
xP * y = ( xi X ) * ( yi t i )
13
where t i is the total population of tract i and the other variables are denoted as before. Take
experienced by the average Black. Unlike the index of dissimilarity, the exposure index is
because its value depends partially on the relative number of X and Y members in the urban
area (Massey and Denton 1987). Therefore, for any two groups there are four possible exposure
indices: xP * x and yP * y are called isolation indices and xP * y and yP * x are called
interaction indices.
White (1983) argued that although regarded as the primary empirical indicator of racial
segregation, the index of dissimilarity, and all parcel-based measures, are subject to what has
been termed the “checkerboard problem”. The parcels can be shuffled into any spatial
configuration, and the index will not change. Also, these indices are sensitive to the size of the
parcels about which the tabulations are made. In fact, D increases as the size of the parcel
decreases. Moreover, the P* statistic can measure both the standard form of segregation and an
alternative form in which members of a given group live closer to members of the other group
Many empirical studies on racial segregation that employed the index of dissimilarity
and the exposure index had been done. Taeuber and Taeuber (1965) studied Black and White
segregation in 1940, 1950 and 1960 using the index of dissimilarity. They found that the levels
of racial segregation were very high in cities throughout the country. Using block data for 207
cities in 1960, the index of dissimilarity ranged from 60.4 to 98.1, only 8 cities had scores
Massey and Denton (1987) examined trends in residential segregation for Blacks,
Hispanics, and Asians in 60 SMSAs between 1970 and 1980. Segregation was measured using
both dissimilarity and exposure indices. They concluded that in spite of declines over the
1980s, Blacks remained the most spatially isolated of the three minority groups. They also
found that Black segregation from Anglos declined in some smaller SMSAs in the south and
west (Austin, Denver, Oklahoma City), but in large urban areas in the northeastern and north
central states (Baltimore, Chicago) there was little change. In Detroit, Newark, New York, and
Philadelphia, the likelihood of Black contact with Anglos actually even decreased. By the index
of dissimilarity, the results largely replicate those by exposure indices. In addition, Hispanic
segregation was markedly below that of Blacks, but increased substantially in some urban areas
that experienced Hispanic immigration and population growth over the decade. Asian
Farley and Frey (1994) analyzed changes in segregation for 232 metropolitan areas
during 1980 to 1990 and claimed much more optimistic results than many other scholars about
the prospects for racial integration. They used the more usual convention of comparing all
Whites to all Blacks and used block group data instead of tracts. They found that the Black-
White index of dissimilarity fell in 194 areas between 1980 and 1990. Furthermore, Jargowsky
(1997) calculated the tract-level Black-White dissimilarity index for the quartile of
metropolitan areas with the largest percentage of Blacks (80 out of 318) between 1980 and
1990, and found that the index declined by 0.03 points, while it decreased by 0.06 points for the
remaining three.
15
The study of economic segregation, which can be defined as “the spatial segregation of
households by income or social class” (Jargowsky 1996: 984), is rooted in the discussion of
class stratification and social mobility. In general, stratification refers to the systemic form of
inequality that divides societies and broadly impact occupational and social status. It may take
many forms based on wealth, gender, age, ethnicity, or some other forms of distinction
(Calhoun 2002). In social sciences, stratification as a term has come to mean the process by
which or the resulting structure in which families become differentiated from one another and
arranged in graded strata with varying degrees of prestige, and/or property, and/or power
(Tumin 1969). Therefore, the term economic segregation in current literature mainly refers to
the stratification by economic status (income or other standards) or social class and the
resulting patterns of spatial separation between families or groups. As a matter of fact, the
study of economic segregation in the US emerges from and is influenced by the investigation of
racial segregation. However, residential segregation by income and other measures of socio-
economic status has received much less attention by social scientists than segregation by race
(Kain 2000).
Worth noting, much of the research on economic segregation has dealt with the extent
of and trends in concentrated poverty. For instance, Wilson’s (1987) research on changes in the
geographic concentration of poverty is a quite famous and influential one. Wilson (1987)
showed that in the nation’s fifty largest cities, the poverty population rose by 12 percent and the
number of persons living in poverty areas (i.e., census tracts with a poverty rate of at least 20
presented diverse results. Applying index of dissimilarity (D) to variables that serve as proxies
for social class is a popular way to measure economic segregation and are often used by most
pre-1987 researchers, such as Ducan and Ducan (1955b), Erbe (1975), and Farley (1977). For
computation purpose this measure requires the variables to be broken down into discrete
categories. However, changes in the social meaning of these categories over time complicate
For example, being a high-school graduate means less in social and economic terms than it
used to.
On the contrary, household income offers greater comparability over time once inflation
is taken into account. Massey and Eggers (1990) applied the index of dissimilarity to the
income distribution. To compare interclass segregation for Whites, Blacks, Hispanics, and
Asians, they use census tract data from 1970 and 1980 for the 50 largest standard metropolitan
statistical areas (SMSAs) plus 10 others with large numbers of Hispanics. They defined four
social classes based upon specific income thresholds: poverty income, lower-middle class,
upper-middle class, and affluent; then for each racial and ethnic group, they computed the six
pair-wise indices of dissimilarity among the four social classes and averaged these indices to
come up with a final aggregate measure. Based on this analysis, they found that interclass
segregation among Blacks increased over the 1970s in an often quite sharp manner, while
declines in interclass segregation for Whites, Hispanics, and Asians. These findings generally
confirmed Wilson’s hypothesis that Black segregation by income increased between 1970 and
1980 (Wilson 1987). Nevertheless, Jargowsky (1996) pointed out several flaws of this
measurement. First, the cutoff points between income classes are unavoidably arbitrary.
17
Second, income is a continuous variable, and collapsing it into four categories discards
information. Third, and the most important, this measure confounds changes in the income
distribution with changes in spatial organization because this measure is not independent of the
Some other more advanced measures, such as the entropy index, could easily handle
multiple categories, but is still affected by the shifts in the underlying distribution that change
the meaning of the categories (Jargowsky 1996). Another possible measure is the Gini
coefficient of segregation. However, as White (1987) pointed out, although Gini coefficient is a
good measure of economic segregation, the computation involved is complicated and difficult.
The correlation ratio, which is also called as the eta-squared or the segregation statistic,
has been utilized to measure segregation as well (Bell 1954; Zoloth 1976; Farley 1977; Schnare
1980). In the case of dichotomous variables, the correlation ratio is equivalent to the
asymmetric intergroup contact index (or, exposure index, P*) after standardizing the latter for
the underlying population proportions of the groups being compared (Duncan and Duncan
1955a; White 1986). Yet compared with P*, the correlation ratio could easily extend to
Neighborhood Sorting Index (NSI) as a measure of economic segregation. The NSI can be
defined as:
18
( )
N
2
hn y n − y
n =1
σ H
NSI = N =
σH
(y )
H
2
i −y
i =1
of households in neighborhood n, and H and N are the total number of households and
tracts as proxies of neighborhoods. As the formula above illustrated, the square of NSI is just
The Neighborhood Sorting Index is the correlation ratio with respect to the distribution
measuring economic segregation, NSI treats income as truly continuous and not a set of
categories in its computation. NSI also has an intuitive interpretation in terms of the income
income, and the distribution of household incomes has a mean and a standard deviation. Also
each household is located in a neighborhood. Each neighborhood has a mean income, and the
distribution of households by the mean income of their neighborhood has a mean and a
standard deviation too. NSI is simply the ratio of these two different standard deviations. In a
hypothetical case that all neighborhoods have exactly the same mean income, there would be
no economic segregation; then the standard deviation of the neighborhood distribution is zero,
and NSI would be zero as well. At the other extreme, suppose all households live in
neighborhoods that have mean incomes identical to their own incomes, then the standard
deviation of the neighborhood distribution would be identical to the standard deviation of the
19
household distribution, and NSI would be one. Thus, the value of NSI varies between 0 and 1.
Values close to 0 represent low levels of economic segregation, and values close to 1 indicate
More important, the standard deviation ratio base of NSI discussed above allows it
implicitly control for the shifts in the underlying income distribution of household income,
which is the most desirable advantage of this measurement compared to other economic
segregation proxies. Additionally, NSI also controls for income inequality because it is
expressed as a percentage of total income variance (Jargowsky 1996). Nevertheless, this index
still has few disadvantages. As all other correlation ratio-based measures, NSI is essentially
aspatial (Massey and Denton 1988). Put a different way, NSI can not capture some important
theoretically straightforward, the actual calculation of this index is particularly complicated due
Utilizing NSI, Jargowsky (1996) found relatively low levels of economic segregation
among US metropolitan areas in 1970s and 1980s; this finding is consistent to the results form
previous studies regardless of the measured employed. Despite the low levels, however, he
different result from past research. The NSI values for Whites, Blacks, and Hispanics all
increased steadily between 1970 and 1980 and again between 1980 and 1990. During the two
decades, Blacks experienced the largest increases: the NSI increased more than 40 percent
between 1970 and 1990. Furthermore, Jarwgosky (1996) claimed that the trend toward greater
economic segregation was remarkably widespread. For instance, in the 1980s, 108 out of the
111 (about 97.3 percent) metropolitan areas in US had an increase in NSI among Blacks. For
20
Whites, the NSI increased in 253 out of 318 metropolitan areas (about 79.6 percent); and for
Hispanics, the NSI increased in 39 out of 49 (about 79.6 percent) US metropolitan areas.
segregation has been done. Much of the research has focused on the “social distance”
hypothesis and the “assimilation” hypothesis. These two hypotheses are closely related but
conceptually distinct (Jargowsky 1996). The social distance hypothesis indicates that the degree
of spatial separation between racial groups is hypothesized as a direct function of the social
distance between them (Kantrowitz 1973). Based on this hypothesis, racial or ethnic residential
ethnic minorities and majority population (Massey 1981). The hypothesis asserts that as time
goes by, the original language and culture barriers between racial groups will be weakened, and
successful individuals will “move out and eventually find their places in business and in the
professions” (Park 1926). The point is, as Park (1926) argued, that “change of occupation,
personal success or failure – change of economic and social status, in short – tend to be
high-status minorities from the majority should decrease, and thus economic segregation within
Empirical research has not always been supportive of the assimilation hypothesis. Erbe
(1975), for example, focused on the issue of socioeconomic segregation by social class within
racial groups in the Chicago SMSA by using P* on census 1970 data. Three relevant measures
of socioeconomic status were chosen: male occupation, education of adult male and family
21
income. She found that a middle-class Black person was much more likely to encounter in his
neighborhood adults in lower manual occupations than a middle-class White person. The
educational composition of the neighborhood of most highly educated Black person was
high income Blacks were much less isolated from low income Blacks than high income Whites
were from low income Whites. For example, White families with incomes over $25,000 lived
in tracts where 19 percent of the families had similar incomes and additional 31 percent had
incomes between $15,000 and $25,000. In other words, half the White families surrounding
“rich” Whites had incomes over $15,000. On the contrary, a Black family with an income
exceeding $25,000 lived in a tract where only 23 percent of the Black families made over
$15,000. Therefore she concluded that “middle-class Blacks live in much closer propinquity
with the lower class than do middle-class Whites, simply because the Black lower class is
Blacks and Whites in urbanized areas of the United States in 1970. He used educational
attainment of adult males, occupations of employed males and family income as proxies for
social class, and estimated both the index of dissimilarity and correlation ratios. His major
findings are: first, residential segregation by social class, regardless of the proxy variable used,
was low relative to racial residential segregation. Second, racial segregation was nearly
constant across levels of education, and the findings were largely replicated when occupation of
employed men and family income were used as proxies of social class. These findings do not
support the assimilation hypothesis. Farley concluded that “racial residential segregation is not
Farley (1991) used family income to estimate class and then measured the residential
segregation, which is measured by the index of dissimilarity, for a specific category from all
others. The comparison of residential segregation of Blacks and Whites on the basis of income
in Chicago Metropolitan Area between 1970 and 1980 showed income made virtually no
difference in the extent of residentially segregation. Moreover, the increasing income among
Blacks in 1980 did not lead to residential integration. However, controlled for family income,
the 1980 segregation levels were about 10 points lower than 1970 levels, reflecting the general
decline in Black-White segregation during the 1970s. Furthermore, by comparing the indices of
dissimilarity of “rich” Blacks with “rich” Whites, and of “rich” Blacks with poor Blacks, Farley
(1991) found that the racial residential segregation was more thorough than class segregation in
both 1970 and 1980, and prosperous Blacks were much less segregated from poor Blacks than
Although the empirical work reached quite diverse conclusions by employing various
methodologies, two common points can be summarized from their findings. First, as Massey
(1981) summarized, the social distance and assimilation hypotheses are conceptually distinct
but not contradictory; rather, “they simply address different questions.” In short, social distance
terms of varying social distances between groups,” while assimilation theory is “concerned
with the spatial assimilation of ethnic minorities relative to some majority” (Massey 1981).
Moreover, from a certain point of view, social distance hypothesis can be seen as more focus
interpretation of changes in residential segregation over time. The empirical work indicated that
residential assimilation follows from economic advancement has not been supported for
Second, racial segregation is far more extensive than segregation by social class (Erbe
1975; Farley 1977, 1991; White 1987). Just as Kain (1986) concluded, “virtually every
systematic study has concluded that Black and White differences in income and other
socioeconomic variables account for very little of the current and past patterns of racial
segregation.” In other words, “racial segregation is not an artifact of economic segregation: the
primary organizing principles of the metropolis are race and ethnicity, not social class”
(Jargowsky 1996). This indicates that given the very high level of prevalent racial segregation,
Metropolitan areas in the United States have been sprawling to a remarkable degree.
Suburbanization has become a very attractive research topic in urban studies recently. In
phenomenon, and diverse approaches have been employed to characterize the contemporary
suburbanization patterns.
literature. Then Section 3.2 discusses the comprehensive definition of suburbanization from a
land development perspective and the advantages and drawbacks of the composite sprawl index
as more than only a land use aesthetic. Beyond just a fixed set of characteristics of a region’s
area’s growth over time that is driven by a number of political, economic, and social forces in
this society. Section 3.3 thus discusses suburbanization form this uneven development
perspective.
Based upon this perspective, in this study, the contemporary suburbanization patterns
are characterized by six specific indicators instead of treated as a monolith and defined as a
24
25
comprehensive definition or a composite index. The last section – Section 3.4 – introduces
these indicators: the urban density gradient, the population density, the relatively rapid growth
of the peripheral areas, the homogeneity of new growth, the exclusivity of local zoning, and the
inaccessibility of jobs; and discusses the theoretical relationship between each of them and
economic segregation.
The suburbanization process in US has been going on for centuries. According to Berry
and Kasarda (1977: 180), this urban development process can be defined as “the enlargement
territory”. More recently, suburbanization often bears a new name as “sprawl”, which usually
inherits various negative connotations. From a certain perspective, “sprawl” can be viewed as a
pejorative term that describes certain patterns of contemporary suburbanization. Despite large
suburbanization patterns; instead, numerous claims and counter-claims about “sprawl” are
growth patterns of metropolitan areas. For instance, Los Angeles and Atlanta frequently serve
as typical examples in such case (Geddes 1997). Ad hoc examples often express an aesthetic
judgment: “sprawl” is ugly development. For example, Clawson (1962: 94) described “sprawl”
as the “rapid spread of suburbs across the previously rural landscape, tendency to discontinuity,
large closely settled areas intermingled haphazardly with unused areas”. Some other statements
are even harsher. They criticize “sprawl” as unplanned and illogical, dehumanizing, and
socially isolating. For instance, Kunstler (1993) denounced this type of development for
26
agoraphobic-inducing spectacle”.
(Sierra Club 1998), the mismatch between jobs and housing (Orfield 1997), local fiscal
disparities (Burchell et al. 1998), and civic alienation (Popenoe 1979). For instance, low-
density residential land development patterns are inherently difficult to serve with public
automobiles causes air pollution, traffic congestion, and land consumed by highway
something else. For example, Downs (1998) and Black (1996) argued that “sprawl” occurs as a
development, such as low density (Lockwood 1999), leapfrogging (Harvey and Clark 1965),
large-lot single-family residential (Popenoe 1979), physical separation of land use (Cervero
1991), strip commercial (Black 1996) and non-compact (Gordon and Richardson 1997). For
example, Ewing (1997) defined sprawl as “the spread-out, skipped-over development that
characterizes the non-central city metropolitan areas and non-metropolitan areas of the United
States”. He even vividly portrayed the image of sprawl as “one- or two-story, single-family
residential development on lots ranging in size from one-third to one acre, accompanied by
strip commercial centers and industrial parks, also two stories or less in height and with a
similar amount of land takings”. Finally, some researchers suggest that “sprawl” represents a
27
stage in the development process rather than a static condition; or, it represents some range on a
The above statements are just few representatives in a long list of different approaches
of characterizing contemporary suburbanization patterns. The term “sprawl” is one name for
many conditions. “Sprawl” has been denounced on aesthetic, efficiency, equity, and
environmental grounds and defended on choice, equality, and economic grounds; it has become
“the metaphor of choice for the shortcomings of the suburbs and the frustrations of central
3.2 The land use pattern perspective and composite sprawl index approach
The inconsistent and even contradictory notions of “sprawl” do not provide a clear
picture of the current patterns of metropolitan development. To facilitate research on the causes
and consequences of suburbanization patterns and develop policy judgments, Galster et al.
“sprawl”) from a land development perspective, and developed a composite index of “sprawl”
that can be empirically operationalized and compared across a large number of urbanized areas.
exhibits low levels of some combination of eight distinct, objective dimensions: density,
continuity, concentration, clustering, centrality, nuclearity, mixed uses, and proximity. If each
dimension of land use pattern is placed on a continuum, the lower the level, the greater the
extent of “sprawl” on that dimension; urbanized areas with development patterns that score low
on all dimensions experience more “sprawl” than others. This conceptualization understands
“sprawl” as a multidimensional phenomenon, and indicates the possibility that there can be
Based upon the conceptual definition, Galster et al. (2001) selected thirteen large urban
areas1 from different regions of the country and conducted a prototype test of their definition,
for only housing sprawl. More specifically, they computed a Z score for each of the thirteen
urbanized areas (UAs) on each of the six dimensions (density, concentration, clustering,
centrality, nuclearity and proximity). Then they added the Z scores for each UA across all six
dimensions to obtain a composite sprawl index. Some encouraging results that are consistent
with their firsthand knowledge of these areas, as well as the conventional wisdom, are found
from this prototype test. For instance, the urbanized areas with the greatest degree of sprawl (or
the lowest score on the composite index) were Atlanta, followed by Miami, Detroit, and
Denver. The urbanized areas with the lowest degree of sprawl were all older areas in the
Northeast and Midwest: the New York area had the least sprawl, followed by Philadelphia,
Chicago, and Boston. This comports with the pre-20th-century development of their cores.
Atlanta and Miami represent the epitomes of so-called “wet south region sprawl”, where 20th-
Overall, Galster et al. (2001) concluded that their approach makes it possible to inform
policy far more effectively since the conceptual clarity about the dimensions of “sprawl” and
Nevertheless, this methodology still has several limitations. First, even though the eight
distinct dimensions they defined try to cover the aspects of sprawl as completely as possible, at
29
the same time, the composite index is far from transparent. Although the ranking of urban
sprawl indicators may help to compare the relative sprawl degree among metropolitan areas,
the numerical value of composite sprawl index itself lacks intuitive mathematical meaning, nor
can it be interpreted easily. For example, Galster et al. (2001) presented the composite index for
New York was 8.90, while for Chicago it was 2.97. Can we therefore conclude that the sprawl
More importantly, in their prototype test, each of the dimensions is equally weighted in
calculating the composite index. To address this point, Galster et al. (2001, 12: 708) suggested
that “using statistical techniques such as factor analysis to determine the extent to which
dimensions are associated most closely with others and to develop indices that combine some
or all of the dimensions”. However, factor analysis, as well as any other pure statistical
operation, has no inherent analytical logic by itself. The factors are determined by patterns of
correlations in the data, and these correlations may be driven by things other than sprawl, such
as land characteristics, business cycle effects, and so on. Therefore any “objective” indicator
coming from such an operation may or may not measure sprawl per se. An understanding of
more on access to opportunity and the social mobility dimension, may be more useful for
drawing policy implications than summarizing many dissimilar measures into a single number.
1 The thirteen large urban areas included in Galster et al.’s prototype test are: New York,
Philadelphia, Chicago, Boston, Los Angeles, Washington DC, San Francisco, Houston, Dallas,
Denver, Detroit, Miami, and Atlanta.
30
Many other scholars suggest the contemporary suburbanization patterns (or “sprawl”)
need to be understand as more than just a land use aesthetic or environmental problem.
racial and economic inequality, and power and politics. Beyond just a fixed set of
Mollenkopf, and Swanstrom 2001; Jargowsky 2002; Shlay 2002; Squires 2002).
Squires (2002) argued that the spatial patterns of urban development are not only
encouraged by technological innovation such as modern transportation, but are also rooted in a
context of substantial economic restructuring. Decentralization since the 1970s, which often
was accompanied by the loss of thousands of manufacturing jobs and the growth of both high-
paid producer service industries and low-wage personal service positions, has fueled the
economy, whereby production is decentralized while control and administration are centralized,
has furthered these trends. These economic shifts have stimulated the development of
downtown office space, convention centers, and cultural facilities where relatively well-off
professionals work and play, and the suburban communities where they tend to live. As
downtown and outlying suburban development has proceeded, many urban neighborhoods and
Moreover, these spatial and structural changes feed back on each other. Households
with lower incomes and the greatest economic need find it increasingly difficult to afford
housing in suburban communities, where jobs are growing fastest. As jobs become more
31
distant, it is also more difficult to participate in informal networks through which job
placements are often made. As jobs become harder to find, poverty and the concentration of
poverty increase in urban areas, tax revenues and public services decline, physical conditions
deteriorate and resources leave the community, crime and so-called “underclass behaviors”
increase. These inner-city communities become less attractive to private capital. These cycles
are mutually reinforcing. Thus, suburbanization has always involved both the “pull” of
(Jargowsky 2002). Uneven economic, spatial, and social development are all interrelated pieces
of the metropolitan puzzle (Jargowsky 1996; Wilson 1987, 1996; Massey and Denton 1993).
Based upon this uneven development perspective, scholars often argue that the
contemporary suburbanization patterns have been closely associated with emerging income
Disparity between central cities along with their inner-ring suburbs and the newer outer-ring
suburbs in many metropolitan areas are growing (Orfield 1997). For instance, in 1960, per
capita income in cities was 105 percent of their surrounding suburbs. By 1990, that ratio fell to
84 percent (Cisneros 1993). Although during the 1980s both central cities and suburbs
experienced income increases, the income of suburban residents increased at a higher rate
(Cisneros 1993). Besides the growing income inequality that is associated with the spatial
developments, poverty has become much more concentrated. For example, between 1970 and
1990, the number of census tracts in which the poverty rate was 40 percent or greater and the
economic segregation. In a study across all U.S. metropolitan areas, Jargowsky (2002) found a
32
housing unit and the mean income of the households that occupy those units. Neighborhoods
with the most recent median year built that mainly located in the fringes of the metropolitan
area had the highest mean income. Controlling for median year built, central city
neighborhoods have consistently lower mean incomes than the comparable cohort of suburban
neighborhoods. Within suburbs, the standard deviation of the neighborhood mean incomes
generally decreases as the median year built increase; as expected, the lowest values of
standard deviation are found in the newest neighborhoods. Thus, the mean increases and the
variance decreases the more recently the suburb was built. Using another alternative measure of
the variability of neighborhood mean incomes – the coefficient of variation (CV) – he revealed
fundamentally consistent results. For instance, in the suburbs, the CV is 40 percent lower in the
newest neighborhoods compared with the oldest. Clearly, the newer suburban neighborhoods
Additionally, many scholars pointed out that these development patterns are not
inevitable results of technology improvements and economic growth, and they do not flow
naturally or purely from free market forces (Dreier, Mollenkopf, and Swanstrom 2001; Squires
2002). Markets are real and people do make choices; however, markets operate within, and
individual choices are constrained by, public policy and private sector decision-making. For
example, tax policies have long favored new development in outlying suburban communities
over the reinvestment in older urban neighborhoods (Dreier, Mollenkopf, and Swanstrom 2001;
Squires 1994). Exclusionary zoning laws by most suburban municipalities, racial steering by
real estate agents, and redlining by financial institutions created and continue to reinforce
33
segregated housing patterns (Dreier, Mollenkopf, and Swanstrom 2001; Jackson 1985; Orfield
Overall, suburbanization is not only a land development pattern nor a pure spatial
development picture that has been driven by a number of political, economic, and social forces
in this society. The contemporary suburbanization patterns closely associated with a number of
social problems that have long plagued urban communities, such as income inequality,
by the dynamics of demand and supply in the housing market; nevertheless, it also has been
influenced by public policies that pose essential constraints on individuals’ opportunities and
choices.
the contemporary suburbanization patterns based on the uneven development perspective and
relevant literature, and test their effects on economic segregation. By doing so, first of all, it
makes the interpretation easier. Also, it allows me to make specific policy implications later on.
These indicators of suburbanization are the urban density gradient, the population density, the
relatively rapid growth of the periphery, the homogeneity of new growth, the exclusivity of
The urban density gradient has served as a measure of suburbanization for a long
time. Clark (1951) estimated exponential density functions for a variety of metropolitan areas
34
and years, and observed a strong tendency for density functions to flatten through time. The
d x = d 0 e − bx
near the metropolitan center, and b (larger than 0 usually) is the density gradient. As the natural
ln d x = ln d 0 − bx
The urban density gradient describes how the population per square mile of an area
drops off with distance from the center of the metropolitan area (Berry et al. 1974). It is a rate
at which population density decreases as distance from an area’s center increases (Schiller
2004). The density gradient can reveal the extent to which the spatial structure envisioned by
the monocentric city model actually prevails in a given metropolitan area, and changes in the
density gradient can reveal how centralization has changed in a given area over time (Schiller
2004). Numerous classic estimates of density gradients of metropolitan areas around the world
show that centralization has been declining for the past 200 years (Berry et al. 1974; Berry and
Gillard 1977; Berry and Horton 1970; Edmonston 1975; Mills 1972). Lack of centralization,
which is indicated by a density gradient with a low numerical value, is often cited as evidence
the urban density gradient lessens (Mills 1991). Thus, the expected effect of the first
suburbanization indicator in this study – the urban density gradient – on economic segregation
should be negative.
Density is one of the most widely used measures of suburbanization. Despite the revival
of inner cities in 1980s, population density in the center areas of metropolises had continuously
35
diminished as people relocating to outlying areas (Berry et al, 1974; Carruthers 2003;
Edmonston, Goldberg, and Mercer 1985). Well-known images of urban sprawl, suburban
affluence, and central city decline have led many planners and policy makers to conclude that
low-density development patterns promote racial and income segregation. However, the direct
effects of density on income segregation remain unclear. On the one hand, increased density
might lead to a fishbowl effect, where density crams people together, and creates more
opportunities for mixing housing types and incomes within a small area; on the other hand,
high densities also cause the residential sorting effect, where density leads to intense
competition over urban space, and may end up contributing to income segregation rather than
ameliorating it. A study by Pendall and Carruthers (2003) found density affected economic
decreasing rate. This finding basically confirmed the two simultaneous influences of density on
economic segregation.
that the relatively rapid growth at the peripheral areas of a city center is one significant
pattern of contemporary suburbanization. In 1960s, the nation’s population was roughly one-
third urban, one-third rural, and one-third suburban. By 1990, suburbanites were close to half
the population (Schneider 1992) and in 1992, for the first time, they accounted for a majority of
the nation’s voters (Dreier 2000). In addition, while in recent years several downtown areas
have experienced a population increase and project more in the near future (The Brooking
Institute Center on Urban and Metropolitan Policy and the Fannie Mae Foundation 1998), the
overwhelming continuing pattern is the exodus of households from central cities to the suburbs
(Squires 2002). In 1996 alone, about 2.7 million people left a city for a suburb whereas just
36
800,000 made the opposite move (Katz and Bradley 1999). Between 1970 and 2000, the
suburban share of the total metropolitan population increased consistently from 55.1 percent to
Furthermore, during this process, cities tended to gain lower-income residents and lose
upper-income residents (Squires 2002), and the concentrated poverty doubled between 1970
and 1990 (Jargowsky 1996). As the unevenly developments in suburbs and central cities may
exacerbate geospatial separation of population along income lines, the expected effect of this
metropolitan areas with more rapid population growth at the peripheral suburbs relative to the
The exclusivity of local zoning regimes serves as the fourth suburbanization indicator
in this study. The wide latitude and autonomy granted to local governments in US often
(Dreier, Mollenkopf, and Swanstrom 2001). In the competition for favored residents and
investments, local governments use zoning regulations, which divide localities into districts, or
zones, to segregate land uses and to limit access of potential unwanted new residents with
income below the median for their jurisdiction or more requirements for costly services. To
protect property values, local zoning regulations often favor some kinds of housing, such as
single-family housing, over other “noxious” land uses such as apartment buildings (Wright
1983). Local zoning laws also allow municipalities to regulate the location and minimum lot
size for various kinds of housing (Dreier, Mollenkopf, and Swanstrom 2001). Especially,
affluent suburbs have used “snob zoning” to limit housing for the poor. For instance, Many
suburbs set minimum lot sizes (such as one-half acre per home) that increase the cost of
37
housing and rule out the construction of dense housing – not just apartment buildings, but also
exclusionary zoning regulations, which are rooted in the fragmented arrangements of local
jurisdictions in this country, help isolate the higher-income households from the lower-income
families (Dreier, Mollenkopf, and Swanstrom 2001). Thus, the expected influence of this
affluent flight from city centers often entailed relocating to newly developed, low-density areas
at the urban fringe, where land use regulations enforce socioeconomic homogeneity and ensure
the highest possible quality of public services (Carruthers 2003; Squires 2002). The stock aerial
photograph, symbolic of suburban conformity, shows miles after miles of virtually identical
new developments in suburban areas – no matter they are planned or unplanned, ugly or
beautiful, high density or low density – are “exclusionary” by their very nature, because
households with lower incomes typically find these housing units are difficult to afford (Dreier,
Mollenkopf, and Swanstrom 2001; Goldsmith and Blakely 1992; Orfield 1997). “While nobody
expects that the poor will live in the same neighborhoods as the rich”, as Jargowsky (2002: 58)
argued, “the aspect of suburban development that increases economic segregation is that entire
sectors of the metropolitan are devoted exclusively to one type of housing”. Therefore,
metropolitan areas with more homogeneous newly developed suburban neighborhoods are
inaccessibility of jobs. A voluminous literature documented the loss of job access for poor
38
central-city residents as firms and industries relocated to suburbs (Dreier, Mollenkopf and
Swanstrom 2001). For instance, the well-known “spatial mismatch” hypothesis argues “sprawl”
widened the gap between where people live and where jobs are located; the suburbanization
process interacts with racial segregation in housing to create barriers to job access for low-
income people (Kain 1968; Ihlanfeldt and Sjoquist 1998). This situation contributes to the
continuous income segregation within metropolitan areas and the worsen concentration of
central city poverty. Measured by the length of average daily commutes of central city
residents, this indicator is expected to be positively associated with the economic segregation.
In other words, metropolitan areas with longer commute times of their central city residents are
recognized as more suburbanized and anticipated to have higher level of economic segregation.
CHAPTER 4
metropolitan context, structural economic characteristics, and social distance and inequality. As
shown in Figure 4-1, this model investigates the impact of suburbanization on economic
segregation by controlling for three other types of factors that may also have important effects
Suburbanization indicators
Metropolitan context
Economic
Structural economic segregation
characteristics
has inseparable connections with, and may even has nurtured, economic segregation in this
country. In this study, suburbanization is characterized by six indicators: the urban density
gradient, the population density, the relatively rapid growth of the periphery, the homogeneity
of new development, the exclusivity of local zoning, and inaccessibility of jobs. Since the
39
40
theoretical relationship between each indicator and economic segregation has been elaborated
in Chapter 3, Section 4.2 in this Chapter directly summarizes the hypotheses regarding the
segregation and considerable correlations with the suburbanization indicators, should also be
taken into consideration to minimize the potential omitted variable bias. Variables related to
metropolitan context, structural economic characteristics, and social distance and inequality are
included based upon findings from other research. The theoretical relationships between the
variables characterizing these aspects and economic segregation are briefly discussed in the
Numerous studies indicate that many features of metropolitan labor and housing
markets may have impacts on the extent to which high- and low-income individuals are
segregated from other members of their own race/ethnic group. Based upon findings from other
research, the first set of factors – metropolitan context – is characterized in this study by
variables such as division, aged population ratio, old housing stock ratio, large-size household
ratio, recent in-migration rate, internal turnover rate, total population scale, and population
growth rate.
metropolitan features. Metropolitan areas with more elderly residents are likely to have less
economic segregation because people in this group are less mobile, even though they have
accumulated wealth (Pendall and Carruthers 2003). Economic segregation may be lower for
metropolitan areas with more old housing stock, since there are usually more neighborhoods
41
with mixed housing types available (Pendall and Carruthers 2003). Compared to smaller
households, larger households are less mobile and not able to quickly respond to and reflect
rapid shifts in metropolitan economic patterns, rents, and housing prices. Therefore,
metropolitan areas with more large-size households should have less economic segregation
Considering local housing market dynamics, studies suggest that a rapid influx of new
households put pressure on the housing market and will reduce economic segregation, at least
in short run (Jargowsky 1996). Thus metropolitan areas with higher recent in-migration rate
should have lower economic segregation. On the contrary, as a high rate of internal turnover in
the housing market may advance ecological processes, metropolitan areas with more relocating
The total population scale controls for the size of metropolitan area, which is expected
to have a positive effect on economic segregation since large metropolitan areas may have
greater internal differentiation of neighborhoods than do small ones (Hoch 1987). The relative
change in total population over the decade is also included; however, this variable may have a
complicated direct impact on economic segregation. On the one hand, faster population growth
may put more pressure on local housing market in the short run and reduce economic
segregation; on the other hand, metropolitan areas with faster population growth will become
Furthermore, there are many other idiosyncratic features of metropolitan areas, such as
ownership patterns, institutional arrangements, information networks, and so on, that may
42
influence economic segregation. These largely invariant features are all implicitly controlled by
The second set of factors targets on the local opportunity structure, or, the structural
local economic characteristics. Mean household income, percentage change of mean household
income, manufacturing employment ratio, and management and professional related occupation
Overall mean household income of a metropolitan area may have a nonlinear impact on
economic segregation (Jargowsky 1996). Also, the percentage change in household income
over the decade may have an ambiguous effect on economic segregation. An increase in mean
household income may come about in a variety ways; for example, the wealthy households
may become wealthier, or the poor residents may be catching up. The effect of a change in
mean household income would likely depend on the specific pattern generating the change.
structure (Frey and Speare 1988; Kasarda 1985; Kleinberg 1995; Squires 2002), and it may also
manufacturing and a growing share of jobs in the management and professional related
skilled minority individuals away from traditional minority enclaves to jobs in dispersed
locations (Jargowsky 1996). Studies show that deindustrialization has increased inner-city
stress (Galster, Mincy and Tobin 1997), even as the expansion of knowledge-based industries
has helped ameliorate it (Glaeser 1999). Therefore declines in the manufacturing employment
proportion should increase economic segregation as new firms locate in a more dispersed
43
pattern with concurrent adjustments in the residence patterns of employees (Jargowsky 1996).
Since rising skill requirements can accentuate social class differences within racial/ethnic
groups, decrease group cohesion, and increase economic segregation within racial/ethnic
communities (Wilson 1980), higher management and professional related occupation ratios
The final set of control variables, poverty rate within each racial/ethnic group and the
overall gini coefficient, focus on social distance and inequality. The social distance within each
(Wilson 1987), a higher poverty rate within a racial/ethnic group should produce an increase in
economic segregation, because it may encourage more privileged group members to isolate
themselves spatially (Massey and Eggers 1993; Jargowsky 1996). Furthermore, rapid increases
in income inequality since the late 1960s may increase both overall residential segregation
between income groups and economic segregation within racial and ethnic groups (Massey and
Eggers 1993). Thus metropolitan areas with larger value in gini coefficient, which is an index
4.2 Hypotheses
The hypotheses presented below are designed to test the idea that economic segregation
The first set of hypotheses regards the relationship between economic segregation and
H1: Economic segregation is higher for metropolitan areas with lower urban
H3: Economic segregation is higher for metropolitan areas with more local exclusionary
zoning interventions.
H4: Economic segregation is higher for metropolitan areas with more homogenous
H5: Economic segregation is higher for metropolitan areas with more inaccessibility of
H6: Economic segregation is higher for metropolitan areas with more rapid growth
The next set of hypotheses regards the relationship between economic segregation and
metropolitan context variables: division, aged population ratio, old housing stock ratio, large-
size household ratio, recent in-migration rate, internal turnover rate, and population growth
rate.
H8: Economic segregation is lower for metropolitan areas with higher proportions of
aged population.
45
H9: Economic segregation is lower for metropolitan areas with higher proportions of
H10: Economic segregation is lower for metropolitan areas with higher proportions of
larger households.
H11: Economic segregation is higher for metropolitan areas with higher proportions of
recent in-migrants.
H12: Economic segregation is higher for metropolitan areas with higher proportions of
internal movers.
H13: Economic segregation is higher for metropolitan areas with more total population.
The third set of hypotheses regards the relationship between economic segregation and
ratio, management and professional related occupation ratio, and percentage change of mean
household income.
H16: Economic segregation is lower for metropolitan areas with higher proportions of
H17: Economic segregation is higher for metropolitan areas with higher proportions of
H18: Economic segregation varies with percentage change of mean household income.
46
The final set of hypotheses regards the relationship between economic segregation and
the social distance and inequality measurements: poverty rate within each racial/ethnic group
H19: Economic segregation is higher for metropolitan areas with higher poverty rates
H20: Economic segregation is higher for metropolitan areas with higher overall gini
coefficients.
CHAPTER 5
To operationalize the model and test the hypotheses, the primary data sources are the
1990 and 2000 US Census summary files. Most of the variables are calculated based upon the
2000 Summary File 3 and the 1990 Summary Tape File 3. The 1990 Summary Tape File 1
provides the housing value information of specified owner-occupied housing units in 1990.
Additionally, individual household income information is extracted from the 5 percent Public
Use Microdata Samples (PUMS) data in 1990 and 2000 for estimating the characteristics of
The unit of analysis in this study is the metropolitan area. However, there are a number
primary metropolitan areas (PMSAs)3, and consolidated metropolitan areas (CMSAs)4. CMSAs
2 An MSA consists of one or more counties that contain a city of 50,000 or more inhabitants, or
contain a Census Bureau defined urban area and have a total population of at least 100,000
(75,000 in New England).
3 Subreas may be defined within an area that meets the requirements to qualify as an MSA and
also has a population of one million or more. The definition of these subareas, called PMSAs,
requires meeting specified statistical criteria and having the support of local option.
4 An area that meets the requirements to qualify as an MSA and also has a population of one
million or more becomes a CMSA if component parts of the area are recognized as PMSAs.
47
48
are comprised of contiguous PMSAs, such as the Dallas PMSA and the Fort Worth PMSA,
which make up the Dallas-Fort Worth CMSA. Additionally, New England county metropolitan
areas (NECMAs) are county-based alternatives to the city- and town-based MSAs and CMSAs
in the six New England States. In terms of geographic area and population size encompassed,
PMSAs are more comparable to MSAs than other alternatives (Iceland et al, 2002). The
following Table 5-1 illustrates a comparison among MSAs, PMSAs, and CMSAs regarding
their population sizes. Even though both PMSAs and CMSAs encompass larger average
population sizes than MSAs, the national mean population in PMSAs is only about 3 times of
the mean population size in MSAs; in contrast, the national mean population in CMSAs is
almost 14 times of the mean population size in MSAs. Thus in terms of their population sizes,
PMSAs are relatively more similar to MSAs than CMSAs. Besides, a PMSA comprises “a
large urbanized county or a cluster of counties (cities and towns in New England) that
demonstrate strong internal economic and social links in addition to close ties with the central
core of the larger area” (Geographic Reference Manual of the U.S. Census 2000: 13-2). Hence
this study defines both MSAs and PMSAs as metropolitan areas. To allow comparison over
time, 2000 MSAs and PMSAs geographic boundaries are applied consistently to calculate both
1990 and 2000 variables (Jargowsky 2003). Overall this 2000 metropolitan area set includes
Census tracts, which are relatively small units that are defined consistently across the
country, serve as proxies of neighborhoods to build up the metropolitan area level analysis in
this study. Generally, a census tract contains 1,000 to 8,000 people, with an optimum size of
4,000 people (Glossary of Basic Geographic and Repated Terms of U.S. Census 2000). Census
tracts are small statistical subdivisions of a county with generally stable boundaries and
homogeneous population characteristics, economic status, and living conditions at the time they
are established (Ricketts and Sawhill 1988), and thus often serve as neighborhood proxies in
most research on neighborhoods and segregation. Due to population growth and decline, census
tracts boundaries are sometimes adjusted; for instance, usually the Census Bureau split census
tracts as the population grows too large. In this analysis, contemporaneous tract boundaries are
used rather than constant boundaries. In other words, I use the 1990 census tract definitions in
analyzing the 1990 data and the 2000 boundaries for the 2000 data. It is crucial to use
because the constant boundaries approach results in smaller average neighborhood population
Based upon the conceptual model introduced in Chapter 4, the dependent variable in
this study is economic segregation; the independent variables include the suburbanization
indicators that serve as major explanatory variables, and other groups of control variables that
5 This is known as the Modifiable Areal Unit Problem (MAUP), which is a potential source of
error that can affect spatial studies that utilize aggregate data sources (Unwin 1996). Briefly
speaking, when data are presented using totals for arbitrary areas such as administrative units,
the patterns that they show may be simply the effect of the administrative units rather than
genuine patterns among the underlying population.
50
characteristics, and social distance and inequality. These factors need to be controlled since
they are correlated with suburbanization and yet may have independent effects on economic
segregation. The following sub-sections discuss how these variables are operationalized in
detail.
introduced in Chapter 2, NSI simply is the ratio of between-tract standard deviation of tract
mean household income and the total standard deviation of household income. Values of NSI
vary in the range of 0 and 1; values closer to 1 indicate higher levels of economic segregation
(Jargowsky 1996).
Among alternative measures of economic segregation, NSI is chosen to use in this study
because of its certain desirable properties. Based upon deviations from the mean household
income, it implicitly controls for the overall income level. It also controls for income inequality
Contrary to other measures based on breaking the income distribution into fixed classes,
such as the index of dissimilarity based on specific income thresholds (Massey and Eggers
1990), NSI is not a statistical artifact that shifts with the shifting of income distribution
(Jargowsky 1996). For example, suppose residents within a metropolitan area did not change
their residential patterns at all during a certain period of time and yet only their income
distribution shifted up, the value of the Massey and Eggers’ index of dissimilarity would
unavoidably change, but the value of NSI will remain the same. Hence, NSI is technically
51
stable with respect to changes in the mean and variance of the income distribution.
Furthermore, as a correlation ratio-based measure, NSI also has conceptual advantages since it
gives more emphasis to areal units that differ sharply from the mean (Zoloth 1976). From a
policy perspective, NSI is a more suitable measure because often the concern is with the most
segregation by social class in this country (Erbe 1975; Farley 1977, 1991; White 1987); the
primary organizing principles of the US metropolis are race and ethnicity, not income or social
class (Jargowsky 1996). Thus it makes sense to study economic segregation within racial and
ethnic groups. In this study both 1990 and 2000 NSI values are calculated for each
MSA/PMSA within each of the three racial and ethnic groups: non-Hispanic Whites and others,
In 1997, the Office of Management and Budget (OMB) revised the old standards for the
classification of federal data on race and ethnicity6. Race categories in the Census 2000 data are
defined based upon the new standards and are not directly comparable to the ones in 1990
Census. Also there is no practical method to perfectly match these racial and ethnic categories
over time based upon the aggregate data provided in Census. Given this situation, to still allow
6 The old standards for the classification of federal data on race and ethnicity were issued by
OMB in 1977. In these standards, four racial categories were established: American Indian or
Alaskan Native, Asian or Pacific Islander, Black, and White. In addition, two ethnicity
categories were established: Hispanic origin and Not of Hispanic origin. Considering the
significant changes in the racial and ethnic makeup of U.S. during 20 years, OMB revised the
old standards and announced the new standards in October 1997. Under these new standards
the race categories are: American Indian or Alaska Native; Asian; Black or African American;
Native Hawaiian or Other Pacific Islander; White; and Some Other Race. And repondents are
allowed to pick more than one race. There are also two minimum categories for ethnicity:
Hispanic or Latino and Not Hispanic or Latino. Hispanics and Latinos may be of any race.
52
comparison between 1990 and 2000 and to best serve the purpose of this study, I define three
principal racial and ethnic groups: non-Hispanic Whites and others, Blacks, and Hispanics. For
Census 1990 data, the Black group includes households with a Black householder, and the
Hispanic group comprises households with a householder of Hispanic origin. Then the
difference between the total number of households belonging to these two groups and the
overall total household number regardless race becomes the third group – non-Hispanic Whites
and others. For Census 2000 data, the Black group contains households with a householder who
is Black or African American alone. The Hispanic group consists of households with a
White alone and is not Hispanic or Latino, and households with a householder who is Asian
alone, American Indian and the Alaska Native alone, or Native Hawaiian and Other Pacific
Islander alone, are all combined together and defined as the non-Hispanic Whites and others
group. This configuration of race and ethnic groups provides the best match of inconsistent
categories in the 1990 and 2000 Census data on income by race at the Census tract level7.
Moreover, to capture the patterns of economic segregation over last decade, both 2000
NSI and 1990 NSI for each MSA/PMSA are calculated. Census 2000 SF3 and Census 1990
STF3 data are used to compute the numerator of NSI – the between-tract standard deviation of
tract mean household income. More importantly, since the necessary information about
individual household income are not published by the Bureau of the Census, I use household
income information from the 5 percent PUMS data to estimate the overall standard deviation of
7 Note these racial/ethnic groups are still not strictly equivalents between 1990 and 2000. Also
note that by this method, the counts of non-Hispanic Whites and others, Blacks, and Hispanics
are not unduplicated. That is, the Black figures in both years and the 2000 White and other
figures include some non-White Hispanics, who may be of any race.
53
individual household income (the denominator of NSI)8. To allow comparison over time,
Empirical results indicate that the economic segregation, measuring by NSI, declined
from 1990 to 2000. The next chapter (Chapter 6) discusses this observed trend and other
gradient, the population density, the exclusivity of local zoning, the homogeneity of new
development, the inaccessibility of jobs, and the relatively rapid growth of the peripheral
suburbs.
The first suburbanization indicator in this study – the urban population density
gradient is estimated based upon the natural logarithm density function (Clark 1951):
ln d x = ln d 0 − bx
8 Individuals in PUMS areas that span MSA/PMSA boundaries were included in all the
spanned MSA/PMSAs, with sample weights adjusted proportional to the population
proportions in the spanned MSA/PMSAs, based on estimates obtained from the Master Area
Block Level Equivalency (MABLE) Geographic Correspondence Engine.
(Source: http://www.oseda.missouri.edu/plue/geocorr/)
9 The 1989 household income information reported in Census 1990 STF3 and PUMS data is
converted into 1999 dollars by multiplying an inflation multiplier of 1.34. Based on the
Consumer Price Index (CPI) for all urban consumers for all items with a base period of 1982-
84, this multiplier is calculated as 1999 annual CPI (166.6) over 1989 annual CPI (124.0).
54
Using Census SF3 data, gross population density10 ( d x ) is computed for each tract
within an MSA/PMSA. Often Census Bureau defines multiple central cities in a single
metropolitan area. In such cases, the distance of a census tract’s centroid from the closest
central city is defined as the distance from a neighborhood to the metropolitan center (x),
resulting in a synthetic distance scale. Because of the oblique spheroid shape of the Earth, this
distance is calculated by the Great Circle Distance Formula11. Finally, based upon the tracts’
densities and the distances between tracts and the closest central city, the density gradient (b),
as well as the log of central density ( ln d 0 ), are estimated by regression analyses, for each
MSA/PMSA in both 2000 and 1990. The density gradient is named as ‘DENGRAD’ for
Empirical results show that the density gradients are larger than zero for the vast
majority of US metropolitan areas in both 1990 and 2000; the only exception is Jersey City,
which exists in a continuously urbanized area in the shadow of New York. Thus population
density usually drops off with distance from the center of the metropolitan area in this country,
be more suburbanized as the density gradient lessens (Mills 1991), this indicator is expected to
10 The gross population density is simply defined as total population over total land areas.
11 To apply the Great Circle Distance Formula, the latitude and longitude values of any
centroid are converted from decimal degrees to radians first, then distance between any two
points equals to: 3963.0*arcos[sin(latitude of point1)*sin(latitude of point2) + cos(latitude of
point1)*cos(latitude of point2)*cos(longitude of point2 – longitude of point1)]. This formula
measures distance in miles.
55
measured as the total population over the total land area12 of a certain MSA/PMSA using
Census SF3 data. Empirical calculation shows strong regional differences of urban density
among US metropolitan areas. Density is an average across the whole metropolitan area,
whereas the density gradient represents the slope from the center to the edge.
segregation (Pendall and Carruthers 2003). Thus both gross population density (‘DEN’) and its
square term (‘SQDEN’) are included in the later regression analyses. This indicator is expected
to have non-linear impact on the dependent variable; however, the specific signs remain
unknown.
Due to lack of national data for local exclusionary zoning, two alternative proxy
variables – total number of local government fragmentations and number of local government
regimes, which is the third suburbanization indicator in this study. The “Tiebout Hypothesis”
indicates that the consumer-voters try to move to the place where the local government’s
revenue-expenditure pattern for goods and services fit their preference the best (Tiebout 1956).
This voting-by-moving behavior of residents, and the wide latitude and autonomy granted to
competition by adopting exclusionary zoning policies that filter out potential residents with
income below the median for their jurisdiction or who require more costly services (Dreier,
12 Land area is calculated from the specific boundary recorded for each land entity in the U.S.
Census Bureau’s geographic database. It includes intermittent water and glaciers, which appear
on census maps and in the TIGER® file as hydrographic features. The accuracy of the area
56
Mollenkopf, and Swanstrom 2001). A metropolitan area with a larger number of either total
local governments or local governments per 100,000 households is considered having more
exclusionary zoning restrictions, and thus having higher level economic segregation. Using
Census SF3 data, the total number of local governments13 (‘GOVT’) is summarized for each
MSA/PMSA in both 1990 and 2000; and the number of local governments per 100,000
households, ‘GOVT_SCL’, is simply a rescaled version of the ‘GOVT’. The expected impact
operationalize this indicator, within each MSA/PMSA, suburban census tracts14 are ranked by
the median year structure built information of their housing units; a smaller ranking number
indicates an older tract. Then the last 10 percent of these tracts whose housing units were most
recently built are defined as the “newest developed suburban neighborhoods”. The ratio of
upper value quartile over lower value quartile for specified owner-occupied housing units15 is
computed for each of these newest developed suburban neighborhoods. Finally, mean of these
quartile ratios, weighted by the number of total housing units, is calculated for each
metropolitan area. A lower value on this indicator means a metropolitan area has more
homogenous housing types in its newest developed suburban neighborhoods, though it does not
measurement is limited by the inaccuracy inherent in the mapping of the various boundary
features in the TIGER® file.
13 Number of census places serves as proxies of number local governments.
14 Census tracts sometimes span city boundaries. A census tract with more than 50 percent of
its total population living in a central city is recognized as a “central-city-tract”; otherwise it is
considered as a “suburban tract”.
15 Specified owner-occupied housing units include only owner-occupied, one family houses on
less than 10 acres without a business or medical office on the property.
57
say whether the housing stock is high value or low value. Therefore the expected impact of this
variable is endogenous. However, as discussed in the paragraph above, this variable actually
focuses only on the newest housing. Thus it tests the role of rapidly suburbanizing areas
relative to the other 90 percent existing stock. Moreover, housing value is not exactly identical
to income although they are correlated. Plus, when compare over time, the effect of this
variable depends on homogeneity of both origin and destination neighborhoods. So clearly the
Besides Census SF3 data, Census 1990 STF1 data is also utilized to provide sufficient
housing value information in 1990. Despite the differences existing between STF1 and STF3
data, combining them in the 1990 calculation of this indicator is considered as a legitimate
approach, since Census Bureau claims that the long form estimates in summary file 3 match the
summary file 1 counts for larger geographic areas and are essentially the same for medium and
large cities. Furthermore, appropriate inflation adjustment is made for the 1990 calculation to
length of average daily commutes of employed central city residents who are age sixteen or
16 The 1990 housing value information reported in Census 1990 STF1 data is converted into
2000 dollars by multiplying an inflation multiplier of 1.33. Based on the Consumer Price Index
(CPI) for all urban consumers for all items with a base period of 1982-84, this multiplier is
calculated as 2000 April CPI (171.3) over 1990 April CPI (128.9). Instead of 1999 and 1989
annual CPI used to adjust household income inflation, 2000 and 1990 April CPI are employed
58
older and work outside home. By examining commutes times only among central city residents,
I eliminate suburban commutes to the core. Generally, the longer commute times of central city
residents indicates a more spread out pattern of suburban development. Hence this variable,
variable. Using Census SF3 data, the average commuting time of central city residents,
measured in minutes, has become slightly longer over the last ten years in most of metropolitan
areas.
The relatively rapid growth at the peripheral areas of a city center serves as the last
indicator of suburbanization in this study. Based on Census SF3 data, variable ‘RAPIDGW’ is
calculated as the difference in population growth rates between central cities and peripheral
suburbs for each MSA/PMSA during the study period17. The larger value this variable obtains,
the more rapid growth experienced at the periphery suburbs relative to the city center in a
positive. Empirical results found that the suburban areas experienced relatively higher
population growth relative to central city for the majority of US metropolitan areas.
segregation are summarized in Table 5-7. Moreover, Chapter 7 discusses the basic descriptive
statistics, regional differences, and other observed patterns of each indicator in detail.
here because the reported housing values in Census were estimated by owners by the time they
turned in the Census forms. Note the difference between these two multipliers is negligible.
17 Because central city boundaries change and the Census Bureau changes the list of cities
considered “central”, I use 2000 central cities as the base and project their boundaries onto the
1990 map using GIS.
59
Variables operationalizing metropolitan context are: division, aged population ratio, old
housing stock ratio, large-size household ratio, recent in-migration rate, internal turnover rate,
Nine dummy variables categorize nine census divisions based on Census geographic
information. Census divisions are groupings of states and the District of Columbia that
subdivide the United States for the presentation of census data. Established by the Census
Bureau in 1910, there are nine census divisions: New England (‘NEWENG’), Middle Atlantic
(‘MIDATL’), East North Central (‘ESTNCT’), West North Central (‘WSTNCT’), South
Atlantic (‘STHATL’), East South Central (‘ESTSCT’), West South Central (‘WSTSCT’),
Mountain (‘MOUNTN’), and Pacific (‘PACIFC’); and these nine divisions can be grouped into
four regions: Northeast, Midwest, South, and West (see Figure 5-1). For every MSA/PMSA,
each of the nine division dummy variables is coded as 1 if this metropolitan area is located in
that division; otherwise it is coded as 0. For those metropolitan areas that comprise multiple
divisions, the division where most of their population lived in is considered as the dominant
one, and only the dummy variable corresponding to that dominant division is coded as 1. Thus
each metropolitan area only belongs to one division. Table 5-2 demonstrates the distribution of
metropolitan areas by census division. These division dummy variables control for the potential
regional differences of economic segregation as a group, although the specific direct impact of
Aged population ratio, old housing stock ratio, and large-size household ratio are all
operationalized based upon Census SF3 data. The variable ‘AGEPOPRAT’ measures aged
population ratio as the percentage of the total population over 65 in an MSA/PMSA. This
61
variable is expected to be negatively associated with the dependent variable, since the higher
proportion of aged population within a metropolitan area, the less mobility its residents have.
The old housing stock ratio, ‘AGEHSRAT’, is calculated as the percentage of housing units
built before 1939 in a metropolitan area. This variable is also expected to be negatively related
to the dependent variable, because a city with higher proportion of older housing stock is
considered to have more mixed housing types available and thus less economic segregation
observed. The large-size household ratio, ‘HHSSIZ’, is computed as the number of households
that include more than 4 members over the total number of households in a metropolitan area.
Since large households are considered as less mobile than small ones, a metropolitan area with
higher proportion of large households is expected to have lower level of economic segregation.
Again, the expected impact of this variable on the dependent variable is negative.
Two variables focus on the housing market dynamics: recent in-migration rate
calculated as the newly in-migrant population over the total population in a metropolitan area.
The newly in-migrant population is defined as the metropolitan residents who are age five or
older and lived outside the metropolitan area five years before the base year of the decade. This
variable is expected to be negatively associated with the dependent variable, since more recent
in-migration introduces more disturbances in local housing market and reduces economic
segregation in short term. Internal turnover rate, on the other hand, is expected to be positively
related to the dependent variable, because higher proportion of internal movers in local housing
market advances the ecological process and increases economic segregation. This variable is
computed as the number of metropolitan area residents (except for the in-migrants) who are age
five or older and moved within the metropolitan area in the previous five years over the number
62
of total metropolitan population who are not in-migrants. Both variables are measured based
Using the 100 percent count of population data in Census SF3 files, variable
‘POPSCAL_N’ summarizes the log of total population within metropolitan area in 1990 and
2000. The expected sign of the coefficient for this variable is positive, since larger metropolitan
‘POPGWRAT’, standing for population growth rate, is simply calculated as the difference
between 1990 and 2000 total population over the 1990 total population in a metropolitan area,
based upon the Census 100 percent count of population information. The national mean of
population growth rate is about 13.44 percent over the last decade. As discussed in Chapter 4,
this variable is considered having complicated direct effects on economic segregation; therefore
The following Table 5-3 demonstrates the national means of aged population ratio, old
housing stock ratio, large-size household ratio, recent in-migration rate, internal turnover rate,
and total population scale in MSA/PMSAs in 1990 and 2000. The expected direct impacts of
these metropolitan context variables on economic segregation are summarized in Table 5-7.
18 In this study, 2000 MSAs and PMSAs geographic boundaries are utilized consistently to
calculate both 1990 and 2000 variables. However, the actual metropolitan areas’ boundaries
have changed during the ten years. Thus the consistent boundary method may introduce
potential possibility of measurement errors into the calculations of recent in-migration rate and
internal turnover rate using 1990 data. Although it is hard to exactly determine the scale of
these possible measurement errors, it is reasonable to assume that the probability of these
potential measurement errors is proportional to the percentage of total population living within
the geographic areas that the metropolitan boundary changes occurred. To control for these
potential measurement errors, an adjustment factor that measures the proportion of total 1990
metropolitan population that lived in the metropolitan boundary changed areas is introduced
into the regression analyses. This adjust factor is eventually left out from the estimation models
63
Mean household income is simply calculated as the aggregate household income over
the total household number for each MSA/PMSA based upon Census SF3 data. To allow
comparison over time, appropriate inflation adjustment is made for the 1990 mean household
nonlinear effect on economic segregation (Jargowsky 1996). Thus both overall mean household
income (‘MHHSINC’) and its square term (‘SQMHI’) are included in the later regression
presented in later chapters, because empirical results show that it has no significant effect at all,
indicating the potential measurement errors are not significant either.
19 The 1989 household income information reported in Census 1990 STF3 data is converted
into 1999 dollars by multiplying an inflation multiplier of 1.34. Based on the Consumer Price
Index (CPI) for all urban consumers for all items with a base period of 1982-84, this multiplier
is calculated as 1999 annual CPI (166.6) over 1989 annual CPI (124.0).
64
analyses. This variable is expected to have non-linear impact on the dependent variable,
ratio and management and professional related occupation ratio. Variable ‘MANUFCTRAT’
represents manufacturing employment ratio and is computed as the percentage of the total
employed civilian metropolitan population (16 years and over) who work in manufacturing
segregation as new firms locate in a more dispersed pattern, this variable is expected to be
negatively associated with the dependent variable. Management and professional related
occupation ratio (‘MNGPRFRAT’), on the other hand, is expected positively related to the
dependent variable, since higher proportion of employees who have management and
professional occupation may emphasize class differences within racial/ethnic groups and
increase economic segregation. This variable is calculated as the proportion of the total
employed civilian metropolitan population (16 years and over) who are engaged in
management, professional, and related occupations. Both variables are measured based upon
Census 2000 SF3 and Census 1990 STF3 data: manufacturing employment ratio is derived
from the industry information, and management and professional related occupation ratio is
20 The definitions of industry and occupation categories are significantly different in 1990 and
2000 Census data. These differences were caused by the revision to the Standard Occupational
Classification (SOC) and to the North American Industry Classification System (NAICS). To
minimize the potential measurement errors and to allow comparison over time, the 1990
calculation is adjusted by the using the preliminary crosswalk factors reported by the Census
Bureau. Specifically, to calculate the 1990 manufacturing employment ratio, the corresponding
1990 industry categories are converted to 2000 “manufacturing” category by using the
preliminary industry tabulation crosswalk factors; and to compute the 1990 management and
65
Based on the Census SF3 data, percentage change of mean household income,
‘PCCHGMINC_T’, is simply computed as the difference between 1990 and 2000 mean
household income over the 1990 mean household income in every metropolitan area. As
economic segregation; therefore the expected sign of its coefficient remains unclear.
employment ratio and management and professional related occupation ratio in MSA/PMSAs
in 1990 and 2000. Table 5-5 shows the national means of metropolitan areas’ mean household
income in 1990 and 2000 and of its percentage change over the last decade. The expected direct
impacts of all the variables characterizing structural local economy on economic segregation
Table 5-5. National means of mean household income and percentage change
of mean household income
Structural economic Mean
characteristics 1990 2000 Percentage change
Mean household income 54191.51* 59360.98* 9.93**
*The national means of mean household income are weighted by the number of total
households in MSA/PMSAs, and measured in 2000 dollars.
**The national mean of the percentage changes of mean household income is weighted
by the average of 1990 and 2000 total households number in MSA/PMSAs, and
expressed in percentage form.
professional related occupation ratio, the corresponding 1990 occupation categories are
converted to 2000 “management, professional, and related occupations” category by using the
preliminary occupation tabulation crosswalk factors.
66
Variables operationalizing social distance and inequality are poverty rate within each
Based upon Census SF3 data, for each racial/ethnic group, group’s poverty rate
(‘POVRAT’) is calculated as the population with income below the applicable federal poverty
threshold over the total population for whom poverty status is determined.
The gini coefficient is an index between zero and one that measures the degree of
inequality in the income distribution of a given society. It is derived based on the Lorenz curve,
a cumulative frequency curve that compares the distribution of hosehold income with the
uniform distribution that represents equality. As shown in Figure 5-2, the equality distribution
is represented by the diagonal line; and the greater the deviation of the Lorenz curve from this
line, the greater the inequality. The gini coefficient is measured as the area A over the summary
of areas A and B. It registers zero for a society in which each household receives exactly the
same income, and equals to one if one household gets all the income and the rest get nothing. In
this study, variable ‘GINI_I’ is calculated based upon the individual household income
information from the 5 percent PUMS data. To allow comparison over time, appropriate
21 The 1989 household income information reported in PUMS data is converted into 1999
dollars by multiplying an inflation multiplier of 1.34. Based on the Consumer Price Index (CPI)
for all urban consumers for all items with a base period of 1982-84, this multiplier is calculated
as 1999 annual CPI (166.6) over 1989 annual CPI (124.0).
67
As measures of overall inequality and within group inequality, both the groups’ poverty
rate and the overall gini coefficient are expected to be positively associated with the dependent
variable (see Table 5-7). The following Table 5-6 demonstrates the national means of the
poverty rates for each racial/ethnic group and of the overall gini coefficients in 1990 and 2000.
Table 5-6. National means of poverty rates for each racial group
and overall gini coefficient
Mean
Social distance and inequality variables 1990 2000
Whites and others 8.00 7.75
Group’s poverty rate Blacks 27.78 23.94
Hispanics 22.47 21.35
Overall gini coefficient 0.42 0.45
*All the national means are weighted by total household number
in MSA/PMSAs, and the values of groups’ poverty rate are
expressed in percentage form.
68
Economic segregation increased in the United States from 1970 to 1990. Not only did
inequality among regions widen, but also economic segregation between cities and suburbs,
and income segregation among suburbs grew during this period (Dreier, Mollenkopf, and
Swanstrom 2001; Lucy and Phillips 2000). Meanwhile, the concentration of urban poverty
that poverty areas grew in terms of the number of tracts, in total population, as a percentage of
the overall population, in the percentage of poor persons living in them, and in geographical
size. Furthermore, this problem was mostly clustered in older industrial cities of the Northeast
The increasing trend of economic segregation from 1970 to 1990 was also found at the
neighborhood level. Using census tract data and measuring economic segregation by the
indices of dissimilarity and isolation, Abramson, Tobin, and VanderGoot (1995) found the
dissimilarity of the poor increased by 11 percent, and the isolation of the poor rose by 9 percent
between 1970 and 1990. They further claimed that income segregation increased more in the
Northeast and Midwest than in the South and West areas. Another study by Jargowsky (1996)
used the neighborhood sorting index (NSI) to measure the economic segregation, and also
found a pronounced and nearly ubiquitous trend toward increased economic segregation within
69
70
each racial or ethnical group from 1970 to 1990. As illustrated in the following Table 6-1, the
mean NSI values for all Whites, Blacks, and Hispanics kept raising during the twenty years.
The increases are largest for Blacks: the mean NSI rose more than 40 percent between 1970
and 1990 – from 0.341 to 0.480. Most of this increase happened in the 1980s. Economic
segregation among Whites also increased about 10 percent in both the 1970s and 1980s. For
Hispanics, the mean NSI increased from 0.384 to 0.419 (roughly 9 percent increasing) in the
1970s, and further rose to 0.487 (about 16 percent increasing) during the 1980s.
Table 6-1. Neighborhood Sorting Index (NSI) by racial and ethnic group:
U.S. metropolitan areas (MSA/PMSAs), 1970 to 1990
White Black Hispanic
Mean MSA/PMSAs Mean MSA/PMSAs Mean MSA/PMSAs
Year NSI Number NSI Number NSI Number
1970 0.310 228 0.341 76 0.384 30
1980 0.343 318 0.395 111 0.419 49
1990 0.374 336 0.480 131 0.487 68
Note: includes metropolitan areas with 10,000 or more households for each racial
or ethnic group indicated; means are weighted by number of households for each
racial or ethnic group.
Source: Jargowsky, Paul A. 1996. “Take the Money and Run: Economic Segregation
in U.S. Metropolitan Areas” American Sociological Review Vol. 61
(December. 984-998).
Using 1990 and 2000 Census and PUMS data, this study examines economic
segregation, which is measured by the Neighborhood Sorting Index (NSI), over the last decade
in all US metropolitan areas. Empirical results indicate that economic segregation decreased
significantly from 1990 to 2000, reversing the earlier trend. Section 6.2.1 below discusses the
empirical findings about this reversing trend, followed by the Section 6.2.2 that tries to explore
6.2.1 Empirical findings of declining economic segregation over the last decade
segregation. The following Table 6-2 shows the weighted mean NSI for non-Hispanic Whites
and other racial/ethnic groups, Blacks, and Hispanics. Figures in the upper panel are for all
metropolitan areas in the US that had at least 10,000 households for that group in that year; the
lower panel demonstrates the weighted mean for a constant set of metropolitan areas. For
instance, the 1990 NSI for Blacks is 0.479, indicating the standard deviation of the distribution
of neighborhood mean incomes is about half of the standard deviation of overall income
Note that the 1990 weighted mean NSI values presented in Table 6-2 are not necessarily
comparable to those in Table 6-1 in Section 6.1 due to different calculation methods engaged.
The metropolitan areas investigated in these two tables are not exactly the same. As discussed
in previous Chapter 5, 2000 MSAs and PMSAs geographic boundaries are applied consistently
to calculate both 1990 and 2000 NSI in this study. This constant boundary method recognizes
331 MSAs and PMSAs altogether. Among these metropolitan areas, six of them are excluded
from this study since five metropolitan areas contain no “central city”22 and one has no
“suburbs”23. Based upon the 325 MSA/PMSAs actually included in this study, Table 6-2
further excludes those containing less than 10,000 households for each racial/ethnic group; thus
the MSA/PMSAs number reported in Table 6-2 and Table 6-1 are different. More important,
22 The five metropolitan areas having no “central city” are Bergen, NJ; Brazoria, TX;
Middlesex, NJ; Monmouth, NJ; and Nassau, NY.
23 The metropolitan area having no real “suburbs” is Anchorage, AK, where all of its residents
live in the central city.
72
since the necessary information about individual household income are not published by the
Bureau of the Census, Jargowsky (1996) used the Pareto extrapolation method to estimate the
total variance of household income – which is the denominator of NSI24, whereas this study
directly extracts individual household income information from the 5 percent PUMS data.
Table 6-2. Neighborhood Sorting Index (NSI) by racial and ethnic group:
U.S. metropolitan areas (MSA/PMSAs), 1990 to 2000
White and others Black Hispanic
Sample Mean MSA/PMSAs Mean MSA/PMSAs Mean MSA/PMSAs
and Year NSI Number NSI Number NSI Number
All
MSA/PMSAs
1990 0.454 324 0.479 130 0.501 69
2000 0.377 324 0.408 143 0.400 99
Constant
Set of
MSA/PMSAs
1990 0.454 324 0.479 130 0.501 69
2000 0.377 324 0.408 130 0.398 69
*Note: includes metropolitan areas with 10,000 or more households for each racial or
ethnic group indicated; means are weighted by number of households for each racial or
ethnic group.
As shown in the upper panel of Table 6-2, the NSI values declined for all racial/ethnic
groups from 1990 to 2000. For non-Hispanic Whites and others, the NSI decreased by about 17
percent over the last decade — from 0.454 to 0.377. For Blacks, the NSI decreased from 0.479
to 0.408 (15 percent) from 1990 to 2000. Hispanics experienced an even larger decline: the NSI
for Hispanics decreased by about 20 percent over the ten years — from 0.501 to 0.400. These
findings are not driven by the inclusion of more metropolitan areas in 2000; as shown in the
lower panel of Table 6-2, indices for a constant set of metropolitan areas illustrate a virtually
0.55
0.50
0.45
0.40
Hispanic
0.35
Mean NSI
Black
0.30 White*
0.25
0.20
0.15
0.10
0.05
0.00
1960 1970 1980 1990 2000 2010
Year
The above Figure 6-1 helps to illustrate the patterns of changes in economic segregation
from 1970 to 2000 more clearly. The weighted mean NSI for Whites and others, Blacks, and
Hispanics in 1970 and 1980 are based upon Jargowsky’s study (1996); the 1990 and 2000
numbers are from the empirical results of this study. The upward sloping lines connecting the
1970, 1980, and 1990 mean NSI values demonstrate the continuously increasing economic
segregation for all the three racial groups in those two decades. The turning points are 1990.
From this year till 2000, the line for every racial/ethnic group becomes negative sloping, clearly
74
reflecting the new trend of decreasing economic segregation during the last ten years. The
scopes of these declines are quite considerable for all these three racial/ethnic groups;
nevertheless, the mean NSI values in 2000 after the ten-year diminishing are still higher than
their counterparts in 1980 for Whites and others and Blacks and certainly higher than the
figures in 1970 for all the three groups. For example, the mean NSI for Whites and others
dropped to 0.377 in 2000, which is still about 10 percent higher than the value in 1980 and
Furthermore, the trend toward lower economic segregation was remarkably widespread.
As shown in Table 6-3, 309 out of 324 metropolitan areas (95 percent) had a decrease in NSI
for Whites and others in the 1990s. For Blacks, among 130 metropolitan areas, there were 108
metropolitan areas (90 percent) had higher NSI values in 1990 compared to 2000.
Additionally, the NSI values for the ten largest metropolitan areas in US, as presented in
Table 6-4, demonstrate the same pattern of decreasing economic segregation. Over the last
decade, the NSI decreased for all the non-Hispanic Whites and other racial/ethnic groups,
Blacks, and Hispanics in all these ten largest metropolitan areas. More specifically, Houston
had the largest decline of economic segregation for Whites and others. The NSI for Whites and
75
others in this metropolitan area changed from 0.596 in 1990 to 0.460 in 2000, dropping about
23 percent. As to Blacks, Detroit experienced the most significant decreasing in the NSI value
– from 0.552 to 0.421 or about 24 percent. The most dramatic income segregation diminishing
for Hispanics happened in Atlanta. Over the last ten years, the NSI for Hispanics dropped in
toward decreasing economic segregation is found in the 1990s for all racial and ethnic groups
in U.S. This finding is largely consistent with the dramatic decline of concentrated poverty
during the same period documented by numerous studies. For instance, Kingsley and Pettit
(2003) found that the number of poor residents living in high-poverty neighborhoods in US
metropolitan areas declined slightly from about 7.1 millions in 1990 to nearly 6.7 millions in
2000. As a result, the share of metropolitan poor living in high-poverty neighborhoods declined
during the ten years – from about 31 percent to 26 percent. Jargowsky (2003) also reported that
the number of people living in high-poverty neighborhoods (where the poverty rate is 40
76
percent or higher) declined by a dramatic 24 percent, or 2.5 million people, in the 1990s.
Furthermore, he found the concentrated poverty, which is the share of the poor living in high-
poverty neighborhoods, declined among all racial and ethnic groups, especially for African
Americans.
First of all, the strong economic boom in the 1990s may be the primary contributor to
the declining economic segregation and concentrated poverty that discussed earlier. During the
1990s boom, which is the longest economic expansion on record, the US unemployment rate
fell from 6.8 percent in March 1991 to 4.3 percent in March 2001. Over those ten years, the
economy created almost 24 million jobs – an average of 200,000 jobs per month – and pushed
the jobless rate to 4.5 percent or below for about 34 months (Rodgers 2005). Moreover, the
robust economy not only led to substantial real-wage and income growth for low-income
families and even narrowed wage dispersion in the bottom half of the distribution (Katz 2002),
but also substantially improved the absolute and relative economic position of African
Americans (Rodgers 2005). For example, by the end of this boom, the Black unemployment
rate dropped below 10 percent for a sustained period; and the earnings of African Americans,
the economic prosperity induced pervasive central city revitalization and gentrification progress
in the metropolitan areas in this country. By reinvesting in city centers and deteriorating
neighborhoods, these redevelopment programs changed the essential characters and flavors of
residents.
77
Besides the robust economy, the significant public policy changes in US over the last
decade may also contribute toward the diminishing economic segregation and concentration of
poverty. For instance, the Earned Income Tax Credit (EITC), which is the largest federal aid
program targeted to low-income working individuals and families, was sharply expanded by
the Clinton Administration during the 1990s (Berube 2005). In 1998, the EITC lifted about 4.3
million people out of poverty, half of whom were children25. Meyer and Rosenbaum (1999)
also found that the expansion of EITC contribute to about 37 percent of the increase in the
employment rate of single mothers from 1992 to 1996. Another significant change in the
nation’s welfare system was the replacement of the old Aid to Families with Dependent
Children (AFDC) with the new program called Temporary Assistance to Needy Families
(TANF) in mid-1990s. The TANF required the recipients to undertake constructive activities
motivations of job-seeking and moving behaviors on the recipients. Some studies indicated that
the welfare reform contributed to the sustained and unprecedented drop in the poverty rates for
single mothers and Black children (Rector 2002). For example, Ellwood (1999) concluded that
about 50 percent of the increase in the employment rate of single mothers during the 1990s is
Furthermore, the decentralized housing policy changes during 1990s may serve as
positive factors contributing to lower level of income segregation and concentrated poverty as
well. For instance, the Clinton Administration substantially expanded the Housing Choice
Voucher Program (formerly known as Section 8) and provided very low-income families with
25 This statement with specific numbers is from the JCPR Legislative Research Briefing by the
Joint Center for Poverty Research of the Northwestern University and the University of
78
more opportunities to rent safe, privately-owned apartments or houses and move into better
communities with higher income and lower poverty and crime (Weiss 2002). In addition,
during the last decade, more local governments and jurisdictions adopted the inclusionary
zoning policy, which aims to promote mixed-income, diverse, and integrated communities,
provide housing for a diverse labor force, and connect residents in high poverty neighborhoods
residential developments available to low- and moderate-income households, this policy links
the production of affordable housing with private market development and expands the supply
for affordable housing by dispersing affordable units throughout a city or county (Fox and Rose
2003).
Overall, it is true that economic segregation did decrease significantly over the last
decade, because of certain reasons such as the health of the US economy and few essential
public policy changes discussed above. Nevertheless, suburban development might still have
been contributing to increasing income segregation. The Figure 6-1 in Section 6.2.1 indicates
clearly that although the mean NSI values in 2000 are much lower than the values in 1990, they
are still higher than their counterparts in 1980 and 1970. From a long-term perspective, the
business cycles kept fluctuating while suburbanization progressed steadily during the last thirty
more years. It is reasonable to consider the hypothesis that if there was no continuing
suburbanization, the economic segregation might have decreased even further. Therefore, the
6.3 Descriptive statistics and histogram of NSI and the changes in NSI in the 1990s
controlling for other variables measuring urban scenarios of metropolitan context, structural
economic characteristics, and social distance and inequality. The Neighborhood Sorting Index
(NSI) is the measurement of economic segregation in this study and computed based upon
Census and PUMS data. For the cross-sectional estimations, the NSI are first calculated in both
1990 and 2000 for non-Hispanic Whites and other racial/ethnic groups, Blacks, and Hispanics
respectively; then the six resulting sets of NSI are pooled together as the dependent variable. As
to the first-difference fixed effects analyses, the changes in NSI are computed for all the three
racial/ethnic groups in the 1990s; then the three resulting sets of changes in NSI are pooled and
regressed against all the independent variables. The following Table 6-5 illustrates the basic
descriptive statistics of the pooled NSI and the pooled changes in NSI.
Table 6-5. Descriptive statistics of pooled NSI and pooled changes in NSI
Mean Std. Dev. Minimum Maximum
Pooled NSI* 0.419 0.094 0.129 0.986
Pooled changes in NSI** -0.077 0.047 -0.470 0.233
*Mean of pooled NSI is weighted by the number of total households in
MSA/PMSAs in 1990 or 2000.
**Mean of pooled changes in NSI is weighted by average number of total
households in MSA/PMSAs in1990 and 2000.
The national mean value of pooled NSI is about 0.419. After pooled the NSI for every
racial/ethnic group in both 1990 and 2000 together, the minimum numeric value of 0.129 is
found in Rocky Mount, North Carolina in 2000 for Blacks, whereas the maximum numeric
value of 0.986 is found in Detroit in 1990 for Hispanics. Regarding the pooled changes in NSI,
the national mean value is about -0.077, hinting the trend of decreasing economic segregation
80
discussed earlier. The minimum numeric value of pooled changes of NSI is -0.470, found in
Pittsburgh for Blacks. Despite the widespread declines of income segregation over the last
decade, some metropolitan areas still experienced increasing economic segregation for some
groups. For example, the maximum numeric value of pooled changes of NSI is 0.233, which is
Furthermore, the following Figure 6-2 and Figure 6-3 demonstrate the histograms of the
pooled NSI and pooled changes in NSI respectively. For pooled NSI, as shown in Figure 6-2,
most of the observations fall within the range of positive and negative two standard deviations
around the mean. This distribution has one peak and looks largely symmetrical with few
positive outliers. Figure 6-3 illustrates that the absolute majority of the pooled changes in NSI
fall within the range of positive and negative three standard deviations around the mean. This
distribution also has only one peak and seems mainly symmetrical with few negative outliers.
Overall, both the distributions of pooled NSI and pooled changes in NSI are not very skewed
120
90
Frequency
60
30
0
The first suburbanization indicator is the urban density gradient, which describes how
population per square mile of an area drops off with distance from the center of the city (Berry
et al, 1974). The basic descriptive statistics of this indicator are shown in Table 7-1. Most
importantly, the national mean of urban density gradients of all MSA/PMSAs falls from 0.141
in 1990 to 0.132 in 2000; a drop about 6.38 percent. This decline is also quite widespread.
Among the 325 metropolitan areas26, 250 of them (about 76.85 percent) experienced declines in
The observed decline in urban density gradient over the last decade is consistent with
the historical trend documented by numerous studies. Berry and Horton (1970) recognized the
phenomenon that density gradient falls over time as the second axiom in the intra-urban
structure and growth, and demonstrated the diminishing density gradients in London from 1801
to 1941 and in Chicago from 1860 to 1950. Mills (1972) reported the average population
density gradients of Baltimore, Milwaukee, Philadelphia, and Rochester dropped from 1.22 in
1880 all the way down to 0.31 in 1963. Berry et al (1974) showed the density gradients of the
26 There are totally 331 MSA/PMSAs in 2000 Census data; however, 5 of them (Bergen, NJ;
Brazoria, TX; Middlesex, NJ; Monmouth, NJ; and Nassau, NY) contain no “central city”.
Besides, metropolitan area Anchorage, AK has no real “suburbs” since all its residents live in
its central city. Thus the total number of MSA/PMSAs included in this study is 325.
82
83
eight US urban regions27 they investigated all declining from 1950 to 1970. Furthermore, some
studies with larger scope replicated the same conclusion. For example, Edmonston (1975)
found that the mean population density gradient was 0.65 for 170 US urbanized areas in 1950,
0.47 for 193 urbanized areas in 1960, and eventually 0.38 for 236 urbanized areas in 1970. By
employing the two-point estimation method to examine the Canadian Census Metropolitan
Areas (CMA), scholars also observed significant diminishing of average density gradients from
a value about 1.03 in 1941 to 0.42 in 1976 (Edmonston, Goldberg, and Mercer 1985). All this
literatures, together with the current study, vividly depicts a picture of continuing
differences in the areas studied and alternative computational methodologies, the trend of
flattening urban density gradient is unanimous. As Berry and Gillard (1977: 1) indicated,
“counter-urbanization has replaced urbanization as the dominant force shaping the nation’s
settlement patterns.”
As indicated in classic literature, the urban density gradients for all US MSA/PMSAs
are found positive, with only one exception of Jersey City, New Jersey in 2000. In 1990, the
density gradient of Jersey City is around 0.003, which is already a very small value that
27 These eight urban regions are: Akron, Birmingham, Chicago, Cincinnati, Denver, Detroit,
Seattle, and Washington.
84
demonstrating very similar population densities in its central cities and surrounding areas. In
2000, this value further drops to -0.040, which indicates the population density slightly
increasing with distance from central cities28. One possible reason behind this special case may
be the extreme high density in this metropolitan area. With the highest population density in
this country, the very limited land area forces the relatively large population spreading out all
over the entire place. Another probable cause may be the deteriorating central city
neighborhoods that often accused for gangs behaviors, drug sales, and other repellent features
induce residents moving away from the metropolitan cores. Since the magnitudes of density
gradients for Jersey City in both 1990 and 2000 are essentially moderate and their change over
time is not very dramatic either compared with the national mean, this outlier is assumed not a
interesting regional differences. As shown in Table 7-2, in both 1990 and 2000, West North
Central division and New England division bear the largest urban density gradients among the
nine census divisions. On the contrary, Pacific division has the least average density gradients,
indicating metropolitan areas within this region are generally more suburbanized. For instance,
Los Angeles is the largest metropolitan area in this division and has a density gradient as low as
28 As an approximate verification, I examine the central city and suburbs population density
for Jersey City in 2000. The whole metropolitan area has 2 central cities: Bayonne city and
Jersey City city. The population density for Bayonne city in 2000 is about 10992 persons per
square mile; while its surrounding suburban neighborhoods have a higher mean population
density as 19711 people per square mile. Also, the population density for Jersey City city is
16094 people per square mile in 2000; while its surrounding suburbs have a much higher mean
population density as 33592 persons per square mile. Hence both central cities in Jersey City
metropolitan areas are less dense compared to their surrounding suburbs in 2000.
29 The influence test for Jersey City appears insignificant, and the regression results reveal no
essential change by excluding this outlier.
85
0.08 in 2000. Many other large metropolitan areas, such as San Francisco, Riverside, and
Orange County in California and Seattle in Washington, all bear density gradients no more than
0.10 in 2000.
In terms of change over time, the average urban density gradient of each division
diminishes from 1990 to 2000, with South Atlantic division experiencing the biggest decline of
14.49 percent. A typical case within this division is Atlanta, where the density gradient dropped
from 0.092 in 1990 to 0.079 in 2000, approximately 13.51 percent. With neither natural
geographical boundaries hemming in outward expansion nor state or local policy restrictions,
Atlanta gained its fame as “the city without limits” because of the phenomenal development
and population growth mainly in its suburbs and peripheral areas (Jaret 2002).
areas. Table 7-3 illustrates clearly that the larger metropolitan areas have smaller average
density gradients value in both 1990 and 2000. This finding is consistent with results from
many earlier studies. For example, by investigating the Standard Metropolitan Statistical Areas
(SMSAs) in US from 1950 to 1975, researchers found the mean density gradients inversely
86
connecting with the population size of the central city (Edmonston, Goldberg, and Mercer
1985). Table 7-3 also indicates the unanimous decline of mean density gradients in all different
sized US metropolitan areas over the last decade. Compared to large and extremely small cities,
the middle size metropolitan areas that contain households between 250,000 and 1,000,000
Table 7-4, the gross population densities among US metropolitan areas vary in a wide range. In
1990, Jersey City, New Jersey contains the highest density in this country of approximately
11855 people per square mile, while Flagstaff, Arizona/Utah has only about 4 persons per
square mile. In 2000, Jersey City, New Jersey remains the densest place of US with about
13044 persons per square mile, and the least dense city is still Flagstaff, Arizona/Utah where
Table 7-4. Descriptive statistics of the gross population density: 1990 to 2000
Gross population density Mean Std. Dev. Minimum Maximum
1990 1001.606 1644.454 4.500 11855.27
2000 1064.831 1752.25 5.412 13043.62
*Note: mean of gross population density is weighted by the number of total
households in MSA/PMSAs.
87
The gross population density in US also reveals variations by region and size of
metropolitan areas. As shown in Table 7-5, small metropolitan areas in Middle Atlantic and
New England divisions have the highest mean population densities in 2000, while their
counterparts in West North Central, West South Central, Mountain, and East South Central
divisions contain the least values of gross population density. Especially, the mean density of
small metropolitan areas in Middle Atlantic is more than five times higher than the one in West
North Central. As to medium sized metropolitan areas in US, the highest three mean densities
are found in Pacific, South Atlantic, and Middle Atlantic divisions. West North Central division
still possesses the least mean density, and Mountain division has the second smallest value. The
mean population densities of large metropolitan areas in Middle Atlantic, New England, East
North Central, and Pacific divisions are significantly higher compared to other divisions. On
the contrary, large metropolitan areas in Mountain division are least dense in this country. In
general, metropolitan areas in east and west costal divisions, such as New England, Middle
Atlantic, and Pacific divisions, are much denser compared to their counterparts in other interior
divisions. Regardless of their location, larger metropolitan areas usually tend to have relatively
Regarding the change of population density in US metropolitan areas over the last
decade, Table 7-4 seems to indicate an increasing trend: the national mean rises from about
1002 people per square mile in 1990 to roughly 1065 persons per square mile in 2000.
However, this increase actually is only an artifact that is caused by the consistent metropolitan
boundary approach used in calculation. By assigning the fixed 2000 metropolitan boundaries to
1990 data, the land area in every metropolitan area actually remains constant. Thus the
increasing in national mean of population density is more a reflection of the urban population
88
growth during the last ten years, rather than a sign of greater metropolitan crowdedness
Nevertheless, the gross population density based upon the consistent metropolitan
boundary approach is still appropriate for cross-sectional analyses, since it efficiently captures
the existing differences of population concentration between metropolitan areas at a static time
point. For instance, Dallas with 569 people per square mile in 2000 is considered much less
dense than New York where 8159 persons living in every square mile at the same time.
However, for the first-difference fixed-effects estimations that focus on changes of variables
over time, this computation method may lead to potential measurement bias and misleading
interpretation. The estimated central city density is regarded as a better measure of density
change and thus is utilized as a substitute to gross population density in the first-difference
fixed-effects analyses.
89
The estimated central city density is generated by the urban density gradient calculation.
As shown in Table 7-6, the simple average values of central city densities for the whole nation
reveal a moderate decline from 1990 to 2000. Among the 325 metropolitan areas, 183 of them
(about 56.31 percent) experienced declines in estimated central city density. This trend is
generally consistent with the findings about diminishing central city density in the course of
time from many earlier studies (Berry et al, 1974; Edmonston, Goldberg, and Mercer 1985).
However, the picture becomes a bit blurred when the mean of central city density are weighted
by the metropolitan size that measured as total households number. The last column in Table 7-
6 demonstrates two weighted means of central city density for 1990 and 2000 that are almost
the same.
Table 7-6. Mean values of the estimated central city density: 90 to 2000
Estimated central city density Simple mean Weighted mean*
1990 3432.585 6065.905
2000 3380.869 6085.590
*Note: the weights are total numbers of households in MSA/PMSAs.
In fact, this vague picture reflects the complicated impacts of metropolitan size on
central city density change. On the one hand, growing population in a metropolitan area usually
leads to higher density in its central cities. On the other hand, as a metropolitan area expands,
people tend to move to peripheral areas and leave the central cities less crowded. These two
simultaneous effects largely “wash out” each other and make the observed tendency of
declining central city density fairly weak. The following Figure 7-1 illustrates a graphic version
of the above discussion. The fitted lines30 with positive slope clearly indicate that estimated
30 The fitted lines shown in Figure 7-1 are predicted by a simple OLS regression as following:
Pooled estimated central city density = B0 + B1*Year + B2*Log(population); where the Year
dummy variable is coded as 1 for 2000, and 0 for 1990. The coefficient of the Log(population)
90
central city density increase with total metropolitan population in both 1990 and 2000;
meanwhile, holding metropolitan size constant, the 2000’s estimated central city density
appears smaller than the 1990’s, although the difference is not statistically significant. Thus,
when examining the change in density for a particular city, growing population leads to an
increase in core density that offsets the systemic shift toward lower central density.
9.5
Estimated Central City Density
9
11 12 13 14 15 16
log(population)
lack of national data for exclusionary zoning policies by local governments, total number of
local governments within a certain metropolitan area and number of local governments per
100,000 households serve as two alternative proxy variables measuring zoning regime in this
is 0.364 and significant at 1 percent confidence level. The coefficient of the Year dummy
91
study. The latter one is just a rescaled version of the earlier one. Table 7-7 presents the basic
Both variables vary in a wide range. For example, Chicago had 316 total local
governments in 1990, while there was only one in San Angelo, Texas. As to the number of
local governments per 100,000 households, the minimum value of 2.55 was found in New York
in 1990, and Joplin, Missouri had the highest value as 79.05. Apparently, the range of total
number of local governments is relatively wider. Moreover, the national mean of total number
of local governments increased slightly from 85.7 in 1990 to 88.4 in 2000; whereas the national
mean of number of governments per 100,000 households declined moderately from 14.7 in
1990 to 13.7.
Table 7-7. Descriptive statistics of the local zoning regimes: 1990 to 2000
Zoning regime Mean Std. Dev. Minimum Maximum
Total number of governments
1990 85.659 79.555 1 316
2000 88.362 78.799 3 319
Number of governments
per 100,000 households
1990 14.716 9.639 2.555 79.051
2000 13.714 9.261 2.596 75.195
*Note: means are weighted by the number of total households in MSA/PMSAs.
manner. As shown in Table 7-8, the means of total number of local governments positively
relate with the sizes of metropolitan areas. Both 1990 and 2000 data indicate that larger
metropolitan areas have more local governments overall. In contrast, the means of number of
local governments per 100,000 households decrease as metropolitan sizes increase in both
years. These two opposite patterns together reveal a nonlinear relationship between number of
local governments and size of metropolitan areas: the number of overall local governments
The means of the local government variables increased slightly regardless of size of
metropolitan areas. Relatively small metropolitan areas that contain total households between
62,500 and 125,000 experienced the largest rise of 9.75 percent. The means of number of local
governments per 100,000 households, on the contrary, declined a little from 1990 to 2000 for
every group of metropolitan areas. The sharpest drop, which is about 9.70 percent, happened in
Table 7-8. Total number of governments and number of governments per 100,000 households
by size of metropolitan areas: 1990 to 2000
Mean of number of governments Total number of Number of governments
governments per 100,000 households
Number of households % %
in an MSA/PMSA in 2000 1990 2000 Change 1990 2000 Change
More than 1,000,000 161.3 164.5 1.98 10.10 9.12 -9.70
Between 500,000 and 1,000,000 84.28 89.43 6.11 12.53 11.81 -5.75
Between 250,000 and 500,000 52.12 54.56 4.68 15.96 14.56 -8.77
Between 125,000 and 250,000 27.72 29.57 6.67 17.53 16.41 -6.39
Between 62,500 and 125,000 16.61 18.23 9.75 21.25 20.08 -5.51
Less than 62,500 12.69 13.73 8.20 29.21 28.35 -2.94
*The means are weighted by the number of total households in MSA/PMSAs.
Additionally, Table 7-9 demonstrates the regional variations of these two proxy
variables. The highest numbers of local government per MSA/PMSAs are observed in West
North Central, East North Central, and Middle Atlantic divisions in 1990 and 2000.
Metropolitan areas in Mountain division have the least mean of number of local governments in
both years; however, they experienced the most dramatically raise about 29.16 percent over the
ten years. On the other hand, metropolitan areas in West North Central and East South Central
93
divisions have the largest means of per 100,000 households local governments number in both
1990 and 2000. The least two mean values of this variable are found in Pacific and New
England divisions for both years. Again, for every division, the mean values of overall local
governments number increased over the last decade, whereas the mean values of per 100,000
discussed in Chapter 5, the variable measuring this dimension is the weighted mean of housing
value quartile ratios in the most recently developed suburban neighborhoods in MSA/PMSAs.
As illustrated in Table 7-10, this indicator obtains very similar national mean values in 1990
and in 2000. Also, 2000 calculation yields smaller numbers in both range and standard
deviation, representing the most newly developed suburban neighborhoods between different
metropolitan areas in 2000 are more similar to each other in terms of the homogeneity of their
housing units.
94
neighborhoods by Census divisions. The fairly small mean value in every division indicates
that throughout the whole nation the newly built suburban neighborhoods in metropolitan areas
are pretty homogeneous. The regional variations are quite moderate in both 1990 and 2000,
with only East South Central and West South Central divisions obtaining slightly higher mean
The picture of how the homogeneity of new development varies with metropolitan area
size is neither very clear. Table 7-12 exhibits fairly moderate differences in the mean values of
this variable among metropolitan areas with various sizes. At most, it seems the relatively small
metropolitan areas tend to have more heterogeneous newly built suburban neighborhoods
95
relative to the medium and large ones. For instance, metropolitan areas that have less than
125,000 total households obtain higher mean values around 1.7 and 1.8, compared with mean
the mean household income levels of metropolitan areas. All MSA/PMSAs are categorized into
four groups based upon their mean household income in 2000. In general the metropolitan
areas belonging to the upper quartiles are considered wealthier or better-off regarding their
economic wellbeing than the ones falling into the lower quartiles. As demonstrated in Table 7-
13, in both 1990 and 2000, the mean values of this variable lessen when moving upward along
the quartile rank. In other words, most recently built suburban neighborhoods in richer
metropolitan areas tend to be more homogeneous than the ones in cities with lower mean
measured by the length of average daily commuting time of the central city residents31 in
metropolitan areas in this study. Table 7-14 indicates that the national mean of daily
commuting time for central city residents was 21.6 minutes in 1990. After ten years, the
national mean value becomes 24.5 minutes, increasing about 13.23 percent. This rising trend is
quite widespread as well; among the 325 metropolitan areas, 322 of them (about 99.08 percent)
experienced increasing in daily commuting length. Furthermore, the central city residents of
Bismarck, North Dakota and Grand Forks, North Dakota/Minnesota had the least daily travel
time as 11.4 minutes in 1990 and 13.3 minutes in 2000 respectively. For both years, New York
central city residents suffered for the longest average commutes of 36.4 and 39.9 minutes daily.
Additionally, Table 7-15 below explores potential regional variations of this indicator.
Among the nine Census divisions, Middle Atlantic and Pacific divisions possessed the longest
two average daily commuting times in both 1990 and 2000; while West North Central and East
South Central divisions, in contrast, had the shortest two daily travel times in both years. For all
divisions, the mean travel time per day for central city residents increased over the ten years,
but with different speed. For instance, New England division had the fastest raise as 17.49
percent, whereas West North Central division experienced the least increase by only 9.62
percent.
Furthermore, the average daily commuting time of central city residents also varies by
the size of metropolitan areas. Like indicated in Table 7-16, in both 1990 and 2000, the mean
values of daily travel length become bigger when the sizes of metropolitan areas increase. On
average, the central city residents of super large metropolitan areas that contain more than
1,000,000 households need to spend roughly ten more minutes per day on their way to work,
compared with their counterparts who live in really small metropolitan areas where less than
62,500 households dwell in. Additionally, the mean daily travel length became longer over the
98
last decade for central city residents in all different metropolitan size groups; and the increasing
rates among different metropolitan size groups were pretty similar in general.
The last suburban indicator is the relatively rapid growth of the peripheral suburbs,
which examines the differences of population growth rates between central cities and suburban
areas for each MSA/PMSA. Over the last ten years, the national average population growth rate
of US central cities is only about 8 percent (see Table 7-17), while the national mean suburban
population growth rate reaches 18.2 percent. In fact, there are many central cities even declined
during this period (Kingsley and Pettit 2002). Among the 325 metropolitan areas examined in
this study, 250 of them (about 76.92 percent) have faster growth in their suburban areas. On
average, American population growth in suburbs is about ten percent faster than in central
cities.
Besides the national mean, Table 7-18 shows that this relatively rapid growth of the
periphery indicator varies quite widely. The standard deviation is about 13 percentage points.
Some metropolitan areas experienced dramatic suburban population growth relative to their
central cities during the past ten years. For example, population in suburbs in Naples, Florida
increased by 73.76 percent, while its central city population only grew by 7.5 percent, granting
it the fastest relative growth rate (66 percent) in the whole country as.
Table 7-18. Descriptive statistics of the relatively rapid growth rate (90-2000)
Mean Std. Dev. Minimum Maximum
Relatively rapid growth 10.207 13.126 -152.485 66.223
*Note: mean of relatively rapid growth rate is weighted by the average of
1990 and 2000 total households numbers in MSA/PMSAs, and expressed
in percentage form.
Furthermore, Table 7-19 tries to explore the variation patterns of this indicator by
summarizing it by Census divisions and metropolitan areas’ sizes. However, the whole picture
is not very clear. Small metropolitan areas in East South Central division had the fastest relative
population growth in their suburbs on average, whereas their counterparts in Pacific division
experienced a slight decline in their mean relatively rapid growth rate. Medium metropolitan
areas in South Atlantic and West South Central division obtained the highest and lowest
average relative population growth rates respectively. As to large metropolitan areas, in terms
of mean values, those located in Mountain division had most rapidly grown suburbs and the
others located in Pacific, New England, and Middle Atlantic divisions had the most slowly
Overall, based upon the discussions above, the six suburbanization indicators coarsely
portrayed a picture of suburban development in the metropolitan areas in this country during
the last decade. The urban density gradients and the estimated central city densities for most
metropolitan areas declined. The length of average daily commuting time of the central city
residents increased slightly. In addition, suburbs in most of metropolitan areas have faster
The discussions in earlier sections also reveal that these suburbanization indicators have
diverse regional variations; however, all of them have strong relationships with the size of
metropolitan size introduce potential multicollinearity problem into the regression analyses that
is discussed later in the Chapter 8, and make it harder to net out the impacts of suburbanization
on economic segregation.
CHAPTER 8
test the hypotheses discussed in Chapter 4, three series of regression models based upon
different estimation techniques are employed. The first two sets of cross-sectional regressions
emphasize the static differences between metropolitan areas, whereas the later one set of fixed-
effects estimations focus on the within metropolitan area changes over time. The following
Sections 8.1 and 8.2 elaborate the estimation procedures and results of the cross-sectional
First, I employ two different sets of cross-sectional regressions to test the hypotheses
listed in Chapter 4. In the first set of cross-sectional regressions, I calculate the Neighborhood
Sorting Index (NSI) in both 1990 and 2000 for non-Hispanic Whites and other racial/ethnic
groups, Blacks, and Hispanics respectively, for all metropolitan areas in US that had at least
10,000 households for that group in that year. Considering the limited degrees of freedom, the
six resulting sets of NSI32 are pooled together as the dependent variables, and regressed against
a set of dummy variables for race, year, and their interaction terms. Then variables measuring
32 The six resulting sets of NSI are: Whites and others, 1990; Blacks, 1990; Hispanics, 1990;
Whites and others, 2000; Blacks, 2000; Hispanics, 2000.
101
102
However, only five suburbanization indicators33 are employed in the first set of cross-
sectional models; the sixth indicator – the relatively rapid growth of the periphery – is excluded
from these regressions. The relatively rapid growth indicator measures the differences between
suburban and central city population growth rates over the 1990s. Since the dependent variable
in the first set of regressions includes 1990 NSI values, it would be inappropriate to use the
relatively rapid growth indicator that measures change over the period as a regressor to explain
the different economic segregation levels at the beginning of the period. Of course, one could
attempt to calculate this variable over the 1980s; but projecting 2000 MSA/PMSA and central
city boundaries back onto the 1980 data is very problematic due to boundary changes.
the effect of the relatively rapid growth of the periphery on economic segregation. In these
regressions, only the 2000 NSI values for non-Hispanic Whites and other racial/ethnic groups,
Blacks, and Hispanics are pooled together as the dependent variables, and then regressed
against a set of dummy variables for race, all the six suburbanization indicators, and other
regimes – is measured by two alternative proxy variables: total number of local governments
and number of local governments per 100,000 households. For the first set of cross-sectional
33 These five suburbanization indicators are: the urban density gradient, the population density,
the homogeneity of new growth, the exclusivity of local zoning, and the inaccessibility of jobs.
103
regressions where 1990 and 2000 NSI from different racial/ethnic groups are pooled together as
the dependent variable, I firstly use the number of local governments per 100,000 households
and present the estimation results in Table 8-1, then replicate the regression models by using
the total number of local governments and report the results in Appendix D. Similarly, for the
second set of cross-sectional models where only the pooled 2000 NSI from different
racial/ethnic groups are dependent variable, the number of local governments per 100,000
households is firstly used in the regressions presented in Table 8-2 and then is substituted by
To deal with the heteroskedasticity associated with size of the metropolitan area, all
regressions are weighted by the square root of the number of households entering into the
calculation of the dependent variable, which is NSI in this study. Additionally, because of the
pooling method, some observations from the same metropolitan area may not be independent to
each other; the standard errors are adjusted to allow for covariance of these observations.
The following variables are used in the first set of cross-sectional estimates highlighted
Dependent variable:
34 Only two race dummy variables are included in the cross-sectional regression analyses; the
non-Hispanic White and others is omitted as the base case.
104
Suburbanization indicators:
Metropolitan context:
AGEHSRAT = Old housing stock ratio: percentage of housing units built before
1939.
Most of the variables listed above are also used in the second set of cross-sectional
regressions presented in Table 8-2 in this Section, except for YR, BLK_YR, and HSP_YR.
Since only the pooled 2000 NSI values serve as dependent variable in the second set of models,
the year dummy variable and the interaction terms between year and race become unnecessary
and are excluded from the regressions. The dependent variable and the relatively rapid growth
All the variables discussed above are calculated at the metropolitan area level. The
specific computation methods of these variables are thoroughly discussed in preceding Chapter
5; especially, the expected signs of the coefficients for independent variables are shown in
Table 5-7. In addition, the basic descriptive statistics of the independent variables included in
the first set of cross-sectional models are presented in Appendix F, and the basic descriptive
statistics of the independent variables included in the second set of cross-sectional regressions
The following six equations correspond to the first series of six cross-sectional models
shown in Table 8-1 in this Section. In Equation 1, the pooled NSI is regressed against a set of
dummy variables for race, year, and their interaction terms only. This regression, in effect,
recreates the upper panel in Table 6-2 in preceding Chapter 6, with the coefficients representing
unconditional differences between the cells of that panel. Then the cross-sectional model is re-
estimated by adding five suburbanization indicators39 (see Equation 2). After that, variables
characteristics, and social distance and inequality are added sequentially into the estimations
areas is closely correlated with all the suburbanization indicators and many other control
variables. Thus the log of total population is excluded from earlier equations and finally added
39 These five suburbanization indicators are: the urban density gradient, the population density,
the homogeneity of new growth, the exclusivity of local zoning, and the inaccessibility of jobs;
only the relatively rapid growth of the periphery is excluded from the regressions.
108
scenario.
Equation 1
NSI mry
=β + 0 1
(BLK) + 2
(HSP) + 3
(YR) + 4
(BLK_YR) + 5
(HSP_YR) + u i
Equation 2
NSI mry
= β + β ( BLK ) + β ( HSP) + β (YR) + β ( BLK _ YR) + β ( HSP _ YR)
0 1 2 3 4 5
+ 6
(DENGRAD) + 7
(DEN_SCL) + 8
(SQDEN_SCL)
+ 9
(GOVT_SCL) + 10
(HSVVARRAT) + 11
(AVTM_CC) + u i
Equation 3
Equation 4
NSI mry
= β + β ( BLK ) + β ( HSP) + β (YR) + β ( BLK _ YR) + β ( HSP _ YR)
0 1 2 3 4 56
Equation 5
NSI mry
= β + β ( BLK ) + β ( HSP) + β (YR) + β ( BLK _ YR) + β ( HSP _ YR)
0 1 2 3 4 5
+ 29
(POVRAT)+ 30
(GINI_I)+ u i
109
Equation 6
NSI mry
= β + β ( BLK ) + β ( HSP) + β (YR) + β ( BLK _ YR) + β ( HSP _ YR)
0 1 2 3 4 5
+ β ( POVRAT ) + β (GINI _ I ) +
29 30 31
(POPSCAL_N) + u i
Similarly, the six equations below correspond to the second set of six cross-sectional
regressions presented in Table 8-2 in this Section. In Equation 1, only the pooled 2000 NSI
serves as the dependent variable and is regressed against a set of dummy variables for race.
Then the model is re-estimated by adding all the six suburbanization indicators (see Equation
2). After that, variables controlling for metropolitan context (except population scale),
structural economic characteristics, and social distance and inequality are added sequentially
into the estimations shown in Equation 3, 4, and 5. Since the size of metropolitan areas is
closely correlated with all the suburbanization indicators and many other control variables, the
log of total population is excluded from earlier equations and finally added in Equation 6 to
Equation 1
NSI mr
=β + 0 1
(BLK) + 2
(HSP) + u i
Equation 2
NSI mr
= β + β ( BLK ) + β ( HSP ) +
0 1 2 3
(DENGRAD)
+ 4
(DEN_SCL) + 5
(SQDEN_SCL ) + 6
(GOVT_SCL)
+ 7
(HSVVARRAT ) + 8
(AVTM_CC) + 9
(RAPIDGW) + u i
110
Equation 3
+ β ( RAPIDGW ) +
9 10
(NEWENG) + 11
(MIDATL) + 12
(ESTNCT)
+ 13
(WSTNCT) + 14
(STHATL) + 15
(ESTSCT) + 16
(WSTSCT)
+ 17
(MOUNTN) + 18
(AGEPOPRAT) + 19
(AGEHSRAT) + 20
(HHSSIZ)
+ 21
(INMGRRAT) + 22
(INTRNRAT) + u i
Equation 4
+ β ( INMGRRAT) + β ( INTRNRAT) +
21 22 23
(MHHSINC_SCL)
+ 24
(SQMHI_SCL) + 25
(MANUFCTRAT) + 26
(MNGPRFRAT) + u i
Equation 5
+ β ( MANUFCTRAT) + β ( MNGPRFRAT) +
25 26 27
(POVRAT) + 28
(GINI_I) + u i
Equation 6
+ 29
(POPSCAL_N) + u i
111
Table 8-1 presents the results for the first set of six cross-sectional Weighted Least
Squares regression models. Model 1 includes only the dummy variables for race and year, and
their interactions. Therefore, the constant in this estimation represents the weighted mean of
NSI for Whites and other racial/ethnic groups in 1990 – 0.454. Compared to Whites and others,
both Blacks and Hispanics had significantly higher level within-group economic segregation in
1990. The weighted mean NSI for Blacks was 0.479 and the weighted mean NSI for Hispanics
was about 0.501. In this simple model, the coefficient of the year dummy variable represents
the component of the change in economic segregation common to all these four racial/ethnic
groups from 1990 to 2000 – a decrease of 0.077. There was no significant difference between
Blacks and the non-Hispanic White and other racial/ethnic groups in terms of magnitudes of the
decreases. However, economic segregation among Hispanics declined by 0.024 more than the
common decrease. The findings parallel those in the upper panel of Table 6-2 in preceding
Chapter 6.
for the number of local governments per 100,000 households, all of the coefficients of these
indicators demonstrate expected directions and are statistically significant. The urban density
metropolitan areas with less steep urban density gradients have higher levels of economic
segregation. This negative effect is statistically significant. The gross population density affects
coefficient of density is positive while the coefficient of its square term appears negative,
indicating economic segregation tends to rise when density increases, but in a decreasing rate.
112
Both of these coefficients are significant. The homogeneity of new development is measured by
the weighted mean of housing value quartile ratios in the most recently developed suburban
The highly significant negative coefficient of this variable indicates metropolitan areas with
more homogeneous newly built suburban neighborhoods have higher level of economics
and highly significant. The average daily commuting time of the central city residents increases
by 1 more minute in a metropolitan area, its economic segregation level rises by 0.008.
As a proxy of the exclusivity of local zoning policies, the number of local governments
per 100,000 households has an unexpected negative effect on economic segregation. Despite its
significance, the magnitude of this effect is very small; holding others constant, 1 more local
government per 100,000 households in a metropolitan area affect its economic segregation
level by only 0.001. As discussed in the previous Chapter 7 and shown in Appendix B, the
number of local governments per 100,000 households has a significant negative correlation
with the size of metropolitan areas, and this negative correlation possibly affects the
performance of this local zoning proxy. When the log of total population, which is a measure of
metropolitan size, is introduced in Model 6, the coefficient of the number of local governments
per 100,000 households becomes insignificant and extremely small in magnitude. Additionally,
the later discussions in this Section will show that the performances of two proxies of the
exclusivity of local zoning indicator actually are neither consistent nor stable in these cross-
sectional models. Thus I would not draw any conclusion regarding the effect of exclusionary
zoning regime on economic segregation based upon the cross-sectional estimations; rather,
113
more accurate interpretations about this suburbanization indicator are discussed based on the
of the year dummy variable dropped from -0.077 to -0.100 and was still highly significant. This
change demonstrates that the decrease in economic segregation would have been greater if the
suburbanization indicators had been constant. Additionally, the coefficients for the dummy
variables for Blacks and Hispanics decrease in magnitude and become insignificant after the
Model 3 includes variables to control for the geographic and spatial context of the
metropolitan area. A set of dummy variables for census division captures aspects of
metropolitan area economic structure and housing markets that vary by region. All of these
coefficients are positive and most of them are significant. For instance, compared to the Pacific
division that is the omitted category, the Middle Atlantic division had economic segregation
level that was 0.035 higher, assuming others constant. The highest coefficients appeared for the
West North Central division. Compared to the Pacific division, the West North Central division
As expected, cities with high proportions of aged population had lower levels of
economic segregation. Also, metropolitan areas with high proportion of old housing stock had
smaller NSI values. The coefficient of larger household ratio variable is also negative. As
previous literature indicated, cities with high proportions of households including 4 members
and more had lower income segregation levels, since larger households tend to have lower
mobility. However, none of these three coefficients is significant. Furthermore, the two
variables focusing on the housing market dynamics performed as expected as well; the recent
114
in-migration rate had a negative coefficient and the internal turnover rate had a positive
coefficient, and both of these coefficients are significant. Metropolitan areas with high
proportions of residents moving into the area had smaller NSI values; however, cities with high
proportions of residents moving within the metropolitan area had more economic segregation.
metropolitan area’s average household income affected the economic segregation level in a
nonlinear fashion40. The coefficient for mean household income is positive whereas the
coefficient for its square term is negative, indicating economic segregation rose when average
household income increased, but in a decreasing rate. In this model neither of these coefficients
is significant; however, the coefficients for both average household income and its square term
remained in the expected directions and became significant in the succeeding estimations.
Moreover, both of the variables focusing on the decentralization process had significant
impacts in the expected directions. Metropolitan areas with large share of jobs in manufacturing
sector had smaller values in NSI. On the contrary, cities with large share of jobs in professional
Model 5 adds two variables controlling for social distance and inequality: the group’s
poverty rate reflecting within-group social distance and the gini coefficient measuring overall
income inequality. Both had the expected sign (positive), but only the coefficient for gini
40 Recall that NSI is based on deviations from the mean. Thus, the effect noted here is a
substantive effect of income on social behavior rather than a statistical artifact of the measure’s
construction.
115
So far, Model 5 had included the major explanatory variables and all the control
variables except the total population scale. Until this model, most of the coefficients for
suburbanization indicators still remained the expected directions, and almost every coefficient
for the control variables had the anticipated signs as well. From Model 2 to Model 5,
continuously adding variables controlling for essential urban aspects, such as metropolitan
context, structural economic characteristics, and social distance and inequality, reduced the
potential left out variables bias. However, at the same time, this procedure also inevitably
introduced the multicollenearity problem into the regressions and increased the difficulty of
estimating coefficients for the variables included. For example, although the coefficients for
most of the suburban indicators consistently kept the expected signs through all these four
models, they gradually lost their individual significance. Even so, an F-test41 indicated that
although most of the suburbanization indicators did not have significant effects individually, as
Finally, Model 6 adds the log of total population as a proxy of metropolitan area size.
As expected, the coefficient for this variable was positive and highly significant, indicating
large metropolitan areas encompass greater differentiation of neighborhoods and have higher
significantly correlated with the metropolitan area size42, the introduction of the total
population scale made the coefficients for the urban density gradient, the population density,
and the inaccessibility of jobs suddenly switch to the opposite signs. To solve this problem, the
first-difference fixed-effects estimations discussed in the next Section are employed as a better
41 The F-test for joint significance for suburbanization indicators is presented in Appendix A.
116
approach to implicitly control for the metropolitan area size and the features associated with it
and help to net out the effects of suburbanization indicators on economic segregation more
clearly.
Furthermore, Table 8-2 presents the results for the second set of six cross-sectional
Weighted Least Squares regression models, where only the 2000 NSI values for different
racial/ethnic groups are pooled as the dependent variable and all the six suburbanization
indicators are included as the major independent variables. These regressions help to
investigate the effect of the relatively rapid growth of the periphery indicator on economic
segregation. As shown in Model 2, the relatively rapid growth of the periphery has a positive
coefficient that is highly significant, indicating a metropolitan area obtains higher income
segregation level as its suburbs experience more rapid growth relative to its city center.
However, since Model 3, the coefficient for this indicator lost its significance. The performance
of other suburbanization indicators remain largely the same as in the regressions presented in
Table 8-1. Except for the number of local governments per 100,000 households, all other
suburbanization indicators have the expected directions and are significant initially in Model 2.
However, as other control variables added from Model 3 to Model 5, they lost their
significance gradually. Even so, by an F-test43, I found that although most of the
suburbanization indicators did not have significant effect individually, as a group, they still had
measurements were significantly correlated with the metropolitan area size, the introduction of
the log of total population in Model 6 made the coefficients for the urban density gradient, the
42 Correlations between the suburbanization indicators and log of total population are
demonstrated in Table B-1 in Appendix B.
117
population density and its square term, and the inaccessibility to jobs switch signs. Relative to
the cross-sectional regressions, the fixed-effects estimations discussed in the next Section
provide a better approach to implicitly control for the metropolitan area size and illustrate the
Additionally, Appendix D and Appendix E use the total number of local governments as
the proxy of local exclusionary zoning regimes to substitute the number of local governments
per 100,000 households, and replicate the cross-sectional regressions shown in Table 8-1 and
Table 8-2 respectively. The estimation results of the alternative models shown in Appendix D
are consistent with the ones in Table 8-1 discussed previously, except the coefficients for the
total number of local governments became positive as expected. Put a different way, the sign of
coefficient for the exclusivity of local zoning indicator flipped when I alternated these two
Appendix E and Table 8-2. The inconsistency between the directions of coefficients for these
two alternative zoning proxies is related to the cross-sectional estimation structure and the fact
that both of these variables are very closely correlated with the metropolitan area size. Since
zoning laws are generally controlled by local jurisdictions and no adequate nation-wide zoning
data available among primary research in this field currently, these two proxy variables are still
employed for now. The succeeding Section in this Chapter illustrates that when estimated with
the fixed-effects models, alternating between these two proxies of exclusionary zoning regimes
yields consistent results. Thus I hesitate to draw any conclusion regarding the effect of
43 The F-test for joint significance for suburbanization indicators is presented in Appendix C.
118
estimations; rather, more accurate interpretations about this suburbanization indicator are
discussed based on the fixed-effects regression analyses in the next Section 8.2.
Table 8-1. Neighborhood Sorting Index: Pooled Cross-sectional Weighted Least Squares
regression Results, U.S. Metropolitan Areas, 1990 and 2000
Variables Model1 Model2 Model3 Model4 Model5 Model6
Constant 0.454*** 0.388*** 0.206** –0.024 –0.516*** –0.697***
BLK 0.025** –0.001 –0.004 –0.003 –0.007 –0.010
HSP 0.047*** 0.009 0.008 0.009 0.005 0.002
YR –0.077*** –0.100*** –0.097*** –0.115*** –0.143*** –0.130***
BLK_YR 0.006 0.008 0.009 0.010 0.010 0.011
HSP_YR –0.024*** –0.018* –0.016* –0.015 –0.016* –0.016*
DENGRAD –0.101** –0.098** –0.054 –0.040 0.075**
DEN_SCL 0.002* 0.004*** 0.003** 0.001 –0.0004
SQDEN_SCL –0.003*** –0.004*** –0.002*** –0.001* –0.00001
GOVT_SCL –0.001*** –0.001** –0.001** –0.001* –0.0004
HSVVARRAT –0.043*** –0.018 –0.011 –0.030** –0.027**
AVTM_CC 0.008*** 0.006*** 0.002 0.001 –0.003*
NEWENG 0.022 0.009 0.005 –0.003
MIDATL 0.035* 0.044*** 0.043*** 0.025**
ESTNCT 0.033** 0.050*** 0.051*** 0.043***
WSTNCT 0.067*** 0.063*** 0.063*** 0.045***
STHATL 0.017 0.026** 0.027*** 0.014
ESTSCT 0.025 0.049*** 0.027* 0.025*
WSTSCT 0.036*** 0.054*** 0.037*** 0.032***
MOUNTN 0.053*** 0.063*** 0.065*** 0.059***
AGEPOPRAT –0.0004 0.003* 0.002 0.001
AGEHSRAT –0.0002 –0.0004 –0.0003 –0.0004
HHSSIZ –0.002 0.001 –0.00002 –0.0007
INMGRRAT –0.001* –0.003*** –0.003*** –0.001**
INTRNRAT 0.005*** 0.006*** 0.005*** 0.002*
MHHSINC_SCL 0.004 0.009*** 0.008***
SQMHI_SCL –0.001 –0.004** –0.003**
MANUFCTRAT –0.001** –0.0003 –0.001
MNGPRFRAT 0.002* 0.001 –0.0003
POVRAT 0.0001 0.0002
GINI_I 1.037*** 0.989***
POPSCAL_N 0.031***
R2 0.1830 0.5299 0.6075 0.6630 0.6883 0.7026
N 1089 1089 1089 1089 1089 1089
Note: * p < 0.10 ** p < 0.05 *** p < 0.01 (two-tailed tests)
119
Table 8-2. Neighborhood Sorting Index: Pooled Cross-sectional Weighted Least Squares
Regression Results, U.S. Metropolitan Areas, 2000 only
Variables Model1 Model2 Model3 Model4 Model5 Model6
Constant 0.377*** 0.347*** 0.063 –0.170 –0.531*** –0.672***
BLK 0.031*** 0.009 0.010 0.011 0.010 0.011
HSP 0.023** –0.005 –0.006 –0.005 –0.006 –0.007
DENGRAD –0.094** –0.122*** –0.067* –0.061* 0.042
DEN_SCL 0.002*** 0.003*** 0.002* –0.00002 –0.001
SQDEN_SCL –0.003*** –0.004*** –0.002** –0.001 0.0004
GOVT_SCL –0.002*** –0.001*** –0.001*** –0.001** –0.001*
HSVVARRAT –0.049** 0.002 –0.008 –0.029 –0.029
AVTM_CC 0.005*** 0.005*** 0.002 0.001 –0.002
RAPIDGW 0.001*** 0.0003 –0.0001 –0.00002 0.00004
NEWENG 0.012 0.011 0.008 0.005
MIDATL 0.024 0.044*** 0.043*** 0.026*
ESTNCT 0.019 0.042*** 0.044*** 0.037***
WSTNCT 0.062*** 0.061*** 0.064*** 0.046***
STHATL 0.007 0.026** 0.026** 0.013
ESTSCT 0.009 0.040** 0.021 0.018
WSTSCT 0.022* 0.044*** 0.031** 0.025**
MOUNTN 0.038** 0.050*** 0.053*** 0.045***
AGEPOPRAT –0.0001 0.004** 0.002 0.001
AGEHSRAT 0.0005 0.0001 0.00001 –0.0001
HHSSIZ –0.001 0.002 0.001 0.001
INMGRRAT 0.0003 –0.002** –0.002** –0.0002
INTRNRAT 0.005*** 0.006*** 0.005*** 0.002
MHHSINC_SCL 0.003 0.006*** 0.006***
SQMHI_SCL –0.001 –0.003 –0.002
MANUFCTRAT –0.001* –0.0004 –0.001
MNGPRFRAT 0.003** 0.002* 0.0004
POVRAT –0.0001 –0.0001
GINI_I 0.850*** 0.814***
POPSCAL_N 0.029***
R2 0.0224 0.4412 0.5279 0.6054 0.6315 0.6464
N 566 566 566 566 566 566
Note: * p < 0.10 ** p < 0.05 *** p < 0.01 (two-tailed tests)
120
a further set of tests on the hypotheses listed in Chapter 4. First, I calculate the metropolitan-
level changes in NSI for non-Hispanic Whites and other racial/ethnic groups, Blacks, and
Hispanics in the 1990s. The three resulting sets of NSI44 are pooled and regressed against a set
of dummy variables for race. Then variables measuring the metropolitan-level changes in
and social distance and inequality are added sequentially. Furthermore, many idiosyncratic
features of metropolitan areas that may influence economic segregation are unchanging over
the period in question (such as geophysical configuration, room for expansion, accumulated
housing stock, and historical ownership patterns) or not quantifiable (such as institutional
structure, information networks, and cultural aspects). The fixed-effects estimation structure
helps to minimize the potential left out variable bias associated with these variables because it
To examine the stability of the estimates, both of the alternative instrument variables
measuring the exclusivity of local zoning regimes are explored: the changes in number of local
governments per 100,000 households is included in regressions shown in Table 8-3, and the
estimation results using the changes in total number of local governments is presented in Table
I-1 in Appendix I. In addition, the regressions in Table 8-3 and Table I-1 in Appendix I are re-
estimated by using the relative population growth rate instead of absolute change in total
44 The three resulting sets of NSI are: non-Hispanic Whites and other racial/ethnic groups,
1990-2000; Blacks, 1990-2000; Hispanics, 1990-2000.
121
population scale as the measure of the change in metropolitan area size, and the results are
To deal with the heteroskedasticity associated with size of the metropolitan area,
Weighted Least Squares (WLS) estimations are employed and regressions are weighted by the
square root of the average number of 1990 and 2000 total households in metropolitan areas.
Additionally, because of the pooling method, some observations from the same metropolitan
area may not be independent to each other; the standard errors are adjusted to allow for
Dependent variable:
Suburbanization indicators:
45 Only two race dummy variables are included in the first-difference, fixed-effects regression
analyses; the non-Hispanic White and other racial/ethnic groups is omitted as the base case.
122
Metropolitan context:
46 The reason of using the estimated central city density and its square term to substitute the
gross population density of the entire metropolitan area in first-difference, fixed-effects
estimations are discussed in Section 7.2 in previous Chapter 7.
47 GOVTCHG_SCL is used in regressions shown in Table 8-3 and Table I-2 in Appendix I;
GOVTCHG is used in estimations shown in Table I-1 and Table I-3 in Appendix I.
48 POPSCLCHG_N is used in regressions shown in Table 8-3 and Table I-1 in Appendix I;
POPGWRAT is used in estimations shown in Table I-2 and Table I-3 in Appendix I.
123
All the variables listed above are calculated at metropolitan areas level. The expected
signs of the coefficients for these independent variables are shown in Table 5-7 in preceding
Chapter 5; and the basic descriptive statistics of them are demonstrated in Appendix J.
The following five equations correspond to the five models shown in Table 8-3 in later
part of this Section. In Equation 1, the pooled changes in NSI are regressed against a set of
dummy variables for racial and ethnic groups only. Then the model is re-estimated by adding
the changes in suburbanization indicators (see Equation 2). After that, variables controlling for
49 Mean household income in both 2000 and 1990 are measured in 2000 US Dollars.
124
the changes in metropolitan context, structural economic characteristics, and social distance
and inequality are added sequentially into the first-difference, fixed-effects estimations shown
in Equation 3, 4, and 5.
Equation 1
∆ NSI mr
=β + 0 1
(BLK) + 2
(HSP) + u i
Equation 2
Equation 3
+ β ( AVTMCHG _ CC ) +
8 9
(AGEPOPCHG ) + 10
(HHSSIZCHG )
+ 11
(POPSCLCHG _N) + 12
(INMGRCHG) + 13
(INTRNCHG) + u i
Equation 4
∆NSI mr
= β + β ( BLK ) + β ( HSP ) + β ( DGCHG ) + β ( ESTCCDENCH G )
0 1 2 3 4
+ 14
(PCCHGMINC _T) + 15
(MANUFCTCH G) + 16
(MNGPRFCHG ) + u i
Equation 5
+ 17
(POVRATCHG) + 18
(GINI_ICHG) + u i
125
Note that the relatively rapid growth of the periphery indicator is not included in the
fixed-effects models. In a sense it is a change measure, since it compares rates of growth, and it
seems plausible that it should be included. However, the logic of fixed-effects model requires
that all variables in the model should be entered as changes between one period and the next.
Thus, I would have to compute the difference in the relative rates of growth between central
cities and suburbs in the 1980s and the 1990s. This would require computing the measure for
population changes in the 1980s. As noted previously, such a calculation could be difficult and
Table 8-3 presents the results for the five first-difference, fixed effects Weighted Least
Squares regression models. Model 1 includes only the dummy variables for race and ethnic
groups. The constant in this regression represents the component of the change in economic
segregation common to all these four racial/ethnic groups in the 1990s – a decrease about
0.075. There was no significant difference between Blacks and Whites and others in terms of
magnitudes of the decreases. However, economic segregation for Hispanics declined by 0.032
more than the common decrease. These findings largely parallel those in the lower panel of
patterns. All of the coefficients of these indicators demonstrate expected directions and most of
them are statistically significant. The changes in the urban density gradient are negatively
associated with the changes in economic segregation; and the coefficient is significant. As the
urban density gradient lessens, a metropolitan area is considered to be more suburbanized, and
smaller decrease (or larger increase) in economic segregation is observed over the last decade.
126
As expected, density affects the changes in economic segregation in a non-linear fashion: the
coefficient for the changes in estimated central city density is positive whereas the coefficient
for its square term appears negative; both of these coefficients are highly significant.
The change in the number of local governments per 100,000 households is used as the
proxy of change in the exclusivity of local zoning regimes in Model 2. In contrast to the cross-
sectional models, the coefficient for this variable is positive as anticipated and significant.
Holding the other variables constant, one more local government per 100,000 households in a
metropolitan area leads to an increase in economic segregation about 0.001. Furthermore, the
significant negative coefficient for the changes in the homogeneity of new development
indicates that economic segregation will increase more or decrease less if the most recently
developed neighborhoods in suburbs are more homogeneous in terms of their housing values.
Finally, the coefficient for the change in the inaccessibility of jobs is positive as expected.
Economic segregation within metropolitan areas increase by 0.003 as the average daily
commuting time of central city residents becomes one minute longer, holding others constant.
However, this coefficient is not statistically significant. Interestingly, the introduction of the
changes in suburbanization indicators made the constant term drop from -0.075 to -0.085,
indicating the decrease in economic segregation would have been larger if the suburbanization
expected, cities with larger increases in the proportions of aged population over the decade had
larger decreases in the NSI. Also, metropolitan areas with further increases in the proportions
of large-size households over the 1990s had larger decreases in economic segregation. Both of
the coefficients for these variables are significant. The difference between log total population
127
in 2000 and 1990 measured the absolute change in metropolitan area size. This variable had a
significant positive coefficient, indicating metropolitan areas with larger population expansions
over the decade had smaller decreases in the NSI. As anticipated, cities with higher proportions
of residents moving into the area had larger decreases in economic segregation over the 1990s.
The change in proportion of residents moving within the metropolitan area, however, was not
The effect of the percent change in mean household income over the decade does not have a
manufacturing sector50, which were the norm, are associated with increases in income
segregation, even after controlling for the percent change in mean income over the decade. The
effect of increases in the share of jobs in professional and management related occupations51,
Finally, Model 5 adds two more variables controlling for the changes in social distance
and inequality. Changes in the group’s poverty rate have the expected positive coefficient that
is statistically significant. Increased poverty within the racial and ethnic group thus produced
smaller decreases (or larger increases) in economic segregation, supporting the hypothesis of
middle-class flight (Wilson 1987). The coefficient for gini coefficient is negative and
estimations and my expectation that increased overall income inequality leads to increases in
50 Table 5-3 in Chapter 5 shows the national mean of manufacturing employment ratio
declined from 16.40 percent in 1990 to 13.18 percent in 2000.
51 Table 5-3 in Chapter 5 presents the national mean of management and professional related
occupation ratio increased from 30.94 percent in 1990 to 34.86 percent in 2000.
128
economic segregation within racial/ethnic groups. One possible explanation is that the
unexpected sign on this coefficient reflects a short-run disequilibrium condition. The annual
income probably changes faster than do residential patterns. If the overall variance in the
distribution of household income changes faster than persons can change neighborhoods to
reflect their new economic status, the between-neighborhood proportion of that variance may
temporarily dip. Additionally, in this full model, the coefficients for all suburbanization
indicators still remain the expected directions. However, only the coefficients for the estimated
central city density and its square term are slightly significant, and the coefficient for changes
indicators did not have significant effect individually, as a group, they still had significant joint
In addition, the five regression models shown in Table 8-3 are re-estimated by using the
changes in total number of local governments instead of the changes in number of local
governments per 100,000 households as the proxy of the change in the exclusivity of local
zoning regimes. The regression results are presented in Table I-1 in Appendix I. Unlike in the
regime yields largely consistent results from the fixed-effects regressions. As shown in Table I-
1, the coefficient for the changes in total number of local governments also appears the
expected positive sign, but it is only statistically significant in Model 2. Increases in local
exclusionary zoning regimes helped produce economic segregation, regardless which proxy is
52 The F-test for joint significance for changes in suburbanization indicators is presented in
Appendix H.
129
used. The directions of coefficients for all other independent variables and their significance in
Table I-1 in Appendix I are almost identical to the regression results shown in Table 8-3.
replicated the two sets of regressions models shown in Table 8-3 and Table I-1 in Appendix I
by using the percent change in total population instead of the difference between 2000 and
1990 log population. As alternative measures for the change in metropolitan area size, the
percent change in total population represents the relative population growth rate over the
decade, while the difference between log populations emphasizes the absolute change in total
population scale. The results of these re-estimations are presented in Table I-2 and Table I-3 in
Appendix I. Compared with the coefficients for the absolute changes in log population in Table
8-3, the coefficients for the relative population growth rate shown in Table I-2 share the same
directions (positive) and remain significant in all the models. Furthermore, the coefficients for
all other independent variables in regression models in Table 8-3 and Table I-2 have identical
signs, very similar significance levels, and even almost the same magnitudes for most of them.
The same statements can be drawn by comparing the estimated results in Table I-1 and Table I-
metropolitan size produces consistent and stable estimation results. Overall, the four sets of
regressions shown in Table 8-3 and tables in Appendix I indicate that the fixed-effects
CONCLUSION
not only transforming the geospatial landscape of US metropolitan areas, but also triggering
variety of changes in the social, economic, and political arenas. Suburbanization has been
suspected to contribute to the spatial separation of households along income lines, yet whether
or not the current suburban development patterns are significant contributors to the changes in
To provide some answers to this question, this nationwide study has portrayed the trend
of changes in economic segregation and the patterns of suburbanization process during the last
decade, and modeled the relationship between these two using U.S. Census and PUMS data.
significantly for all racial and ethnic groups during the last decade, reversing the earlier
increasing trend from 1970 to 1990. This finding is largely consistent with the dramatic decline
of concentrated poverty during the same period documented by numerous studies (Jargowsky
2003; Kingsley and Pettit 2003). As to the potential contributors to this ubiquitous trend of
diminishing economic segregation, the strong economic boom in the 1990s may be the primary
one. The significant public policy changes in US over the last decade, such as the expansion of
131
132
the EITC53 program, the replacement of the AFDC54 with the new TANF55 program, and a new
emphasis on decentralization in housing policy, may also help reduce economic segregation
and the concentration of poverty. In the face of these strong factors reducing economic
effect, the question is whether the declining in economic segregation may have been even
indicators that are identified based on the relevant literature. These suburbanization indicators
are the urban density gradient, the population density, the relatively rapid growth of the
periphery, the homogeneity of new growth, the exclusivity of local zoning, and the
inaccessibility of jobs. Together, they portrayed a picture of suburban development: both the
urban density gradients and the estimated central city densities for most metropolitan areas
declined; the length of average daily commuting time of the central city residents increased
slightly; and suburbs in most of metropolitan areas grew faster relative to their inner cities. In
addition, these suburbanization indicators have diverse regional variations, and all of them
suggest that contemporary suburbanization patterns increase economic segregation. In every set
have significant effects on economic segregation initially; however, as more control variables
are included, these coefficients gradually lost their significance. Eventually, most of the
suburbanization indicators had no significant effects individually in the full models, but as a
group, they still had significant joint impact on economic segregation. Overall, even though
these results are not definite proof, they are still strongly suggestive.
In addition, the estimates from both sets of cross-sectional models were plagued by the
significant correlations between the suburbanization indicators and the sizes of metropolitan
areas that were measured by the log of total metropolitan population. Also, due to the
limitations associated with the cross-sectional estimation structure, alternating between the two
proxies of the exclusivity of local zoning regimes yielded inconsistent results on this
provided more robust results. The fixed-effects estimation structure implicitly controls for the
metropolitan area size and the features related to it and nets out the effects of suburbanization
indicators more clearly. Unlike the cross-sectional regressions, the fixed-effects models yield
highly stable estimates that are consistent with expectations when switching between the two
alternative proxies of the exclusivity of local zoning regimes or between the two alternative
Given the strong correlations between metropolitan-level variables and the relatively
small sample sizes, strong conclusions are probably not justified. The results are suggestive,
needed in the future. The empirical results basically suggested that as a group, the
segregation; however, the individual effect of each indicator was unclear in the fully controlled
models. Thus, further careful investigation will be necessary before any specific policy
formulation and enforcement. Additionally, the more significant empirical results for other
control variables indicates that more aspects of metropolitan development other than
suburbanization should be take into account when considering the changes in economic
segregation.
With these caveats in mind, a few specific policy implications can be made based upon
the empirical results in this study. For instance, the well-known images of urban sprawl,
suburban affluence, and central-city decline have led many planners and policy makers to jump
Therefore, many organizations that seek to curb sprawl through smart growth (or
favored by their policies will have better socioeconomic integration (Talen 2002). This
perception is partially driven by significant empirical evidence that Whites move away from
central-city neighborhoods where Blacks and/or other racial minorities are concentrated and
relocate to newly developed, low-density areas at the urban fringe (Carruthers 2003).
However, the empirical results from this study confirm that density actually has a
coefficients for the gross population density and its square term are positive and negative
respectively. In the fixed-effects models, the coefficients for the estimated central city density
and its square term not only obtain the expected signs, but also remain significant even in the
135
full models. Higher density does not necessarily lead to lower economic segregation; actually,
it increases economic segregation in a decreasing rate. Although this finding might seem
counterintuitive, it indeed suggests the connection between density and income segregation is
more complex than is typically thought. Thus, the increasingly popular smart growth programs
aimed at shaping a denser, more compact urban form may not be the optimal solution to greater
socioeconomic equity, though they may be helpful to promote public transit and address
environmental concerns.
indicator in the cross-sectional regressions. The coefficient for this indicator always obtains the
anticipated sign (negative) in every cross-sectional model, even when the size of metropolitan
areas is controlled. In the first set of cross-sectional regressions where the 1990 and 2000
observations are pooled together, the coefficients for this indicator remain individually
significant in the full models. This finding indicates that the aspect of current suburban
development that entire sectors of the metropolitan are devoted exclusively to similar types of
housing helps to increase economic segregation. As to the fixed-effects models, besides the
estimated central city density, the exclusivity of local zoning is the only indicator that has
significant effect on economic segregation individually in the full models. This finding
supports the argument that the exclusivity of local zoning policies helps to isolate the higher-
Based upon these results, compared to smart growth programs, inclusionary zoning
ordinances seem to be a better choice to address economic segregation concerns than just
focusing on density. Inclusionary zoning programs aim to provide affordable housing through
the leveraging of both public and private resources and stimulate social and economic
136
2005; The PolicyLink 2004). These programs require developers to make a certain percentage
zoning variances, and/or expedited permits, that reduce construction costs. By linking the
programs expand the supply for affordable housing to a diverse labor force by dispersing
affordable units throughout a city or county, and help to connect residents in high poverty
Inclusionary zoning, sometimes also called “inclusionary housing”, can take various
forms. Some inclusionary zoning programs are mandatory, whereas others are voluntary or
housing units in exchange for development rights, while the incentive-based programs allow
construct affordable dwelling units within the development, whereas others allow affordable
housing units to be built in another location. While some programs require developers to
actually build the units, some other communities allow developers to contribute to an
Recently, inclusionary zoning programs have become a common tool for local
Massachusetts, New Jersey, and Colorado. Many metropolitan areas, such as Washington DC,
Santa Fe, New Mexico, and Burlington, Vermont, have also adopted these policies (The
PolicyLink 2004). Especially, the state of California is home to the most inclusionary zoning
137
programs: there are 107 jurisdictions employ these policies to produce affordable housing. Till
2003, more than 34,000 units of affordable housing had been constructed in California (The
PolicyLink 2003). Moreover, a recent case study in New York city indicates that a carefully
development without dampening market-rate development in the big city (Bagley and Meltzer
2005).
The results of this study also indicate that as much as economic segregation is the
primary concern, the thing that really matters is not so much about the fact that metropolitan
areas are spreading out; rather, it is how the cities are spreading out. The empirical work in this
research illustrates that the urban density gradients declined over the last decade, indicating
cross-sectional and fixed-effects analyses, this indicator has no significant effects on economic
segregation in fully controlled models. On the contrary, the homogeneity of new development
and the exclusivity of local zoning are the two most robust indicators that have significant
models respectively. These findings suggest that the fast development at urban fringe with
spread-out land use patterns does not necessarily produce income segregation. But during the
economic segregation.
Economic segregation is a significant social problem that deserves attention from policy
makers in all levels of governments. The spatial isolation of the lower-income households not
only affects poverty and inequality in the short-run, but also threatens equality of opportunity
138
and social mobility in the long run (Jargowsky 2002). The poor are less likely to find out about
remote suburban job opportunities. Even if they do, it is increasingly difficult for them to
obtain one because of inefficient public transportation systems. Isolated and disadvantageous
neighborhoods suffer for rampant crime and other problem behaviors, lack necessary public
services, and diminish the life chances of their residents. In addition, the negative impacts of
segregation also may increase inequality of education, hinder the development of human and
cultural capital in the next generation, and contribute to the inter-generational transmission of
poverty (Farkas 1996; Levy 1995). Furthermore, spatial separation of income groups weakens
the collective ability to respond to the challenges of poverty and inequality (Jargowsky 2002).
Certainly, this research has limitations that should be improved in the future. For
instance, both the total number of local governments and the number of local governments per
100,000 households are relatively weak proxies of the exclusivity of local zoning regimes; yet
they are still employed due to lack of adequate nation-wide zoning data. Better proxy
measurements for this indicator should be explored in future research. Also, comparative study
of multiple states could be considered. For example, Texas and California are both large states;
however, they have very different local zoning history: while California adopts lots of zoning
regulations, Texas has really few. Comparison between these two states would be helpful to
Ha: otherwise.
Since the P-value is obviously smaller than α = 0.01 , the null hypothesis is rejected.
139
APPENDIX B
140
APPENDIX C
Ha: otherwise.
Since the P-value is obviously smaller than α = 0.01, the null hypothesis is rejected.
141
APPENDIX D
Table D-1. Neighborhood Sorting Index: Pooled Cross-sectional Weighted Least Squares
regression Results, U.S. Metropolitan Areas, 1990 and 2000
Variables Model1 Model2 Model3 Model4 Model5 Model6
Constant 0.454*** 0.409*** 0.205** –0.068 –0.548*** –0.809***
BLK 0.025** 0.001 –0.003 –0.002 –0.005 –0.009
HSP 0.047*** 0.017 0.009 0.010 0.007 0.002
YR –0.077*** –0.094*** –0.096*** –0.116*** –0.144*** –0.133***
BLK_YR 0.006 0.009 0.009 0.010 0.010 0.011
HSP_YR –0.024*** –0.021** –0.017* –0.016 –0.017* –0.017*
DENGRAD –0.063 –0.062 –0.035 –0.022 0.088***
DEN_SCL 0.003*** 0.005*** 0.003** 0.001 –0.001
SQDEN_SCL –0.003*** –0.005*** –0.003*** –0.001* 0.0002
GOVT 0.0002* 0.0001 0.00001 0.00003 –0.0001
HSVVARRAT –0.068*** –0.029* –0.017 –0.035*** –0.024*
AVTM_CC 0.006*** 0.006*** 0.003 0.001 –0.002
NEWENG 0.035* 0.019 0.011 –0.0002
MIDATL 0.031 0.044*** 0.041*** 0.028**
ESTNCT 0.027 0.049*** 0.048*** 0.048***
WSTNCT 0.047*** 0.054*** 0.054*** 0.050***
STHATL 0.012 0.026** 0.025** 0.016
ESTSCT 0.019 0.046*** 0.024 0.028*
WSTSCT 0.030** 0.051*** 0.034*** 0.033***
MOUNTN 0.052*** 0.063*** 0.065*** 0.059***
AGEPOPRAT –0.001 0.003** 0.002 0.001
AGEHSRAT –0.001 –0.001 –0.001 –0.0004
HHSSIZ –0.002* 0.001 –0.0002 –0.0006
INMGRRAT –0.002* –0.003*** –0.003*** –0.001**
INTRNRAT 0.005*** 0.006*** 0.005*** 0.002*
MHHSINC_SCL 0.004 0.009*** 0.008***
SQMHI_SCL –0.001 –0.004** –0.003**
MANUFCTRAT –0.001 –0.0002 –0.001
MNGPRFRAT 0.002** 0.001 –0.00004
POVRAT 0.00003 0.0002
GINI_I 1.076*** 0.976***
POPSCAL_N 0.038***
R2 0.1830 0.5267 0.6038 0.6597 0.6870 0.7040
N 1089 1089 1089 1089 1089 1089
Note: * p < 0.10 ** p < 0.05 *** p < 0.01 (two-tailed tests)
142
APPENDIX E
Table E-1. Neighborhood Sorting Index: Pooled Cross-sectional Weighted Least Squares
Regression Results, U.S. Metropolitan Areas, 2000 only
Variables Model1 Model2 Model3 Model4 Model5 Model6
Constant 0.377*** 0.364*** 0.055 –0.260** –0.605*** –0.821***
BLK 0.031*** 0.011 0.011 0.011 0.012 0.012
HSP 0.023** –0.001 –0.005 –0.004 –0.005 –0.006
DENGRAD –0.059 –0.087* –0.055 –0.050 0.061*
DEN_SCL 0.003*** 0.004*** 0.002* 0.00004 –0.001**
SQDEN_SCL –0.003*** –0.004*** –0.002** –0.0006 0.001
GOVT 0.0001 0.0001 –0.0001 –0.00004 –0.0001**
HSVVARRAT –0.078*** –0.010 –0.012 –0.033* –0.026
AVTM_CC 0.005*** 0.005*** 0.003* 0.002 –0.001
RAPIDGW 0.001*** 0.0004 –0.0001 –0.00002 0.00002
NEWENG 0.026 0.022 0.016 0.009
MIDATL 0.021 0.048*** 0.045*** 0.030**
ESTNCT 0.015 0.045*** 0.045*** 0.044***
WSTNCT 0.044** 0.059*** 0.061*** 0.053***
STHATL 0.002 0.028** 0.027** 0.016
ESTSCT 0.002 0.040** 0.020 0.022
WSTSCT 0.015 0.043*** 0.029** 0.026**
MOUNTN 0.035** 0.050*** 0.054*** 0.045***
AGEPOPRAT –0.0003 0.004** 0.003* 0.002
AGEHSRAT –0.0001 –0.0002 –0.0003 –0.0002
HHSSIZ –0.002 0.002 0.001 0.001
INMGRRAT 0.0003 –0.002*** –0.002*** –0.0002
INTRNRAT 0.005*** 0.007*** 0.006*** 0.002
MHHSINC_SCL 0.003 0.006*** 0.006***
SQMHI_SCL –0.001 –0.003 –0.002
MANUFCTRAT –0.001 –0.0001 –0.001
MNGPRFRAT 0.003*** 0.002** 0.001
POVRAT –0.0002 –0.0002
GINI_I 0.883*** 0.795***
POPSCAL_N 0.038***
R2 0.0224 0.4268 0.5174 0.5997 0.6277 0.6500
N 566 566 566 566 566 566
Note: * p < 0.10 ** p < 0.05 *** p < 0.01 (two-tailed tests)
143
APPENDIX F
144
145
146
147
conducted after the first-difference, fixed-effects regression shown as Model 5 in Table 8-3.
Ha: otherwise.
( RSS R − RSSU ) m
The F statistics: F = = 2.07
RSSU (n − k )
Since the P-value is obviously smaller than α = 0.10 , the null hypothesis is rejected.
Thus, as a group, the changes in suburbanization indicators have significant effects on the
148
APPENDIX I
149
150
152
BIBLIOGRAPHY
Abramson, Alan J., Mitchell S. Tobin, and Matthew R. VanderGoot. 1995. “The Changing
Geography of Metropolitan Opportunity: The Segregation of the Poor in U.S.
Metropolitan Areas, 1970-1990.” Housing Policy Debate 6(1): 45-72.
Bagley, Nick, and Rachel Meltzer. 2005. “A Guarded Optimism: The Viability of Inclusionary
Zoning in New York City.” Presented at the Urban Affairs Association 35th Annual
Research Conference.
Bain, Read. 1969. In A Dictionary of the Social Sciences, edited by Julius Gould and William
L. Kolb. New York: The Free Press.
Bell, Wendell. 1954. “A Probability Model for the Measurement of Ecological Segregation.”
Social Forces 32: 357-64.
Berry, Brian J. L., Andrew J. Bruzewicz, Douglas B. Cargo, James B. Cummings, Donald C.
Dahmann, Peter G. Goheen, Charles P. Kaplan, Dorothy B. Koopman, Richard F.
Lamb, Lee F. Margerum, Marvin W. Mikesell, David J. Morgan, Jack P. Mrowka, John
P. Piccininni, and Judith A Soisson. 1974. Land Use, Urban Form, and Environmental
Quality. The University of Chicago, Department of Geography, Research Paper No. 155
for the Office of Research and Development, Environmental Protection Agency.
Berry, Brian J. L., and Quentin Gillard. 1977. The Changing Shape of Metropolitan America:
Commuting Patterns, Urban Fields, and Decentralization Processes, 1960-1970.
Cambridge, MA: Ballinger Publishing Company.
Berry, Brian J. L., and Frank E. Horton. 1970. Geographical Perspectives on Urban Systems:
With Integrated Readings. Englewood Cliffs, NJ: Prentice Hall, Inc.
Berry, Brian J. L., and J. D. Kasarda. 1977. Contemporary Urban Ecology. New York:
Macmillan Publishing Co., Inc.
Black, Thomas J. 1996. “The Economics of Sprawl.” Urban Land 55(3): 6-52.
153
154
The Brooking Institute Center on Urban and Metropolitan Policy and the Fannie Mae
Foundation. 1998. “A Rise in Downtown Living.” Washington, DC: The Brooking
Institute Center on Urban and Metropolitan Policy and the Fannie Mae Foundation
(November).
Burchell, Robert W., George Lowenstein, William R. Dolphin, Catherine C. Galley, Anthony
Downs, Samuel Seskin, Katherine G. Still, and Terry Moore. 1998. The Costs of
Sprawl—Revisited. Transit Cooperative Research Program Report 39, sponsored by The
Federal Transit Administration. Transportation Research Board of National Research
Council. Washington, DC: National Academy Press.
Calhoun, Craig. 2002. In Dictionary of the Social Sciences, edited by Craig Calhoun. Oxford
University Press.
Carruthers, John I. 2003. “Growth at the Fringe: The Influence of Political Fragmentation in
United States Metropolitan Areas.” Papers in Regional Science 82: 475-99.
Cervero, Robert. 1991. “Land Use and Travel at Suburban Activity Centers.” Transportation
Quarterly 45(4): 479-91.
Cisneros, Henry G. 1993. “Interwoven Destinies: Cities and the Nation.” In Interwoven
Destinies: Cities and the Nation, edited by Henry G. Cisneros. New York: W. W.
Norton & Company.
Clark, C. 1951. “Urban Population Densities.” Journal of the Royal Statistical Society Series A
114: 490-96.
Clawson, Marion. 1962. “Urban Sprawl and Speculation in Suburban Land.” Land Economics
38(2): 94-111.
Crane, Jon. 1991. “Effects of Neighborhood on Dropping Out of School and Teenage
Childbearing.” In The Urban Underclass, edited by Christopher Jencks and Paul E.
Peterson (299-320). Washington, DC: Brookings Institution.
Downs, Anthony. 1998. “How America’s Cities Are Growing: The Big Picture.” Brookings
Review 16(4): 8-12.
——. 1999. “Some Realities about Sprawl and Urban Decline.” Housing Policy Debate 10(4):
955-74.
Dreier, Peter. 2000. “Sprawl’s Invisible Hand.” The Nation (February 21): 6-7.
Dreier, Peter, John Mollenkopf, and Todd Swanstrom. 2001. Place Matters: Metropolitics for
the Twenty-first Century. Lawrence, KA: University Press of Kansas.
155
Duncan, Otis Dudley, and Beverly Duncan. 1955a. “A Methodological Analysis of Segregation
Indexes.” American Sociological Review 20: 210-17.
Edmonston, Barry, Michael A. Goldberg, and John Mercer. 1985. “Urban Form in Canada and
the United States: an Examination of Urban Density Gradients.” Urban Studies 22: 209-
17.
Ellwood, David. 1999. “The Impact of the Earned Income Tax Credit and Social Policy
Reforms on Work, Marriage and Living Arrangements.” Working Paper. Harvard
University.
Erbe, Brigitte M. 1975. “Race and Socioeconomic Segregation.” American Sociological Review
40: 801-12.
Ewing, Reid. 1997. “Is Los Angeles-Style Sprawl Desirable?” Journal of the American
Planning Association 63(1): 107-26.
Farkas, George. 1996. Human Capital or Cultural Capital? Ethnicity and Poverty Groups in an
Urban School District. Hawthorne, NY: Aldine de Gruyter.
Farley, Reynolds. 1977. “Residential Segregation in Urbanized Areas of the United States in
1970: An Analysis of Social Class and Racial Differences.” Demography 14: 497-517.
——. 1991. “Residential Segregation of Social and Economic Groups Among Blacks, 1970 to
1980.” In The Urban Underclass, edited by C. Jencks and P. E. Peterson. Washington,
DC: The Brookings Institution.
Farley, Reynolds, and William H. Frey. 1994. “Changes in the Segregation of Whites from
Blacks During the 1980s: Small Steps Toward a More Integrated Society.” American
Sociological Review 59: 23-45.
Fox, Radhika J., and Kalima Rose. 2003. Expanding Housing Opportunity in Washington DC:
The Case for Inclusionary Zoning. Oakland, CA: PolicyLink.
Frey, William H., and Alden Speare, Jr. 1988. Regional and Metropolitan Growth and Decline
in the United States. New York: Russell Sage Foundation.
Galster, George, Royce Hanson, Michael R. Ratcliffe, Harold Wolman, Stephen Coleman, and
Jason Freihage. 2001. “Wrestling Sprawl to the Ground: Defining and Measuring an
Elusive Concept.” Housing Policy Debate 12(4): 681-717.
156
Galster, George, Ronald Mincy and Mitchell Tobin. 1997. “The Disparate Racial
Neighborhood Impacts of Metropolitan Economic Restructuring.” Urban Affairs
Review 32: 797-824.
Geddes, Robert. 1997. “Metropolis Unbound: The Sprawling American City and the Search for
Alternatives.” American Prospect 8(35): 40.
The Geographic Areas Reference Manual of U.S. Census 2000. U.S. Census Bureau.
Glaeser, Edward L. 1999. “Are Cities Dying?” Journal of Economic Perspectives 12: 139-60.
The Glossary of Basic Geographic and Repated Terms of U.S. Census 2000. U.S. Census
Bureau.
Goldsmith, William W., and Edward J. Blakely. 1992. Separate Societies: Poverty and
Inequality in U.S. Cities. Philadelphia: Temple University Press.
Gordon, Peter, and Harry W. Richardson. 1997. “Are Compact Cities a Desirable Planning
Goal?” Journal of the American Planning Association 63(1): 89-106.
Hartman, Chester. 2002. Review Comments. In Urban Sprawl: Causes, Consequences and
Policy Responses, edited by Gregory D. Squires. Washington, DC: The Urban Institute
Press.
Harvey, Robert O., and William A. V. Clark. 1965. “The Nature and Economics of Sprawl.”
Land Economics 41(1): 1-9.
Heinlich, R., and W. Andersen. 2001. Development at the Urban Fringe and Beyond: Impacts
on Agriculture and Rural Land. ERS Agriculture Economics Report No. 803.
Washington, DC: U.S. Department of Agriculture.
Helling, Amy. 2002. “Transportation, Land Use, and the Impacts of Sprawl on Poor Children
and Families.” In Urban Sprawl: Causes, Consequences and Policy Responses, edited
by Gregory D. Squires (119-40). Washington, DC: The Urban Institute Press.
Hoch, Irving. 1987. “City Size and U.S. Urban Policy.” Urban Studies 24: 570-86.
Iceland, John., Daniel H. Weiberg, and Erika Steinmetz. 2002. Racial and Ethnic Residential
Segregation in the United States: 1980-2000. U.S. Census Bureau, Census Special
Report, CENSR-3. Washington, DC: U.S. Government Printing Office.
157
Ihlanfeldt, Keith R., and David L. Sjoquist. 1998. “The Spatial Mismatch Hypothesis: A
Review of Recent Studies and Their Implications for Welfare Reform.” Housing Policy
Debate 9(4): 849-92.
——. 2000. “The Geographic Mismatch between Jobs and Housing.” In The Atlanta Paradox,
edited by David L. Sjoquist (116-27). New York: Russell Sage Foundation.
Jackson, Kenneth T. 1985. Crabgrass Frontier: The Suburbanization of the United States. New
York: Russell Sage Foundation.
Jaret, Charles. 2002. “Suburban Expansion in Atlanta: ‘The City without Limits’.” In Urban
Sprawl: Causes, Consequences and Policy Responses, edited by Gregory D. Squires
(165-206). Washington, DC: The Urban Institute Press.
Jargowsky, Paul A. 1996. “Take the Money and Run: Economic Segregation in U.S.
Metropolitan Areas.” American Sociological Review 61(December): 984-998.
——. 1997. Poverty and Place: Ghettos, Barrios, and the American City. New York: Russell
Sage Foundation.
——. 2002. “Sprawl, Concentration of Poverty, and Urban Inequality.” In Urban Sprawl:
Causes, Consequences and Policy Responses, edited by Gregory D. Squires (39-71).
Washington, DC: The Urban Institute Press.
——. 2003. “Stunning Progress, Hidden Problems: The Dramatic Decline of Concentrated
Poverty in the 1990s.” Living Cities Census Series, Center on Urban and Metropolitan
Studies. Washington, DC: The Brookings Institution.
——. May 19, 2003. Public Lecture, “Stunning Progress, Hidden Problems: Concentrated
Poverty in the 1990s.” Washington, DC: The Brookings Institution.
Jencks, Christopher, and Susan E. Mayer. 1990. “The Social Consequences of Growing Up in a
Poor Neighborhood.” In Inner-City Poverty in America, edited by Laurence E. Lynn, Jr.
and Michael G. H. McGeary (111-86). Washington, DC: National Academy Press.
——. 1986. “The Influence of Race and Income on Racial Segregation and Housing Policy.” In
Housing Desegregation and Federal Policy, edited by J. M. Goering. Chapel Hill, NC:
University of North Carolina Press. pp. 99-118.
——. 2000. “Racial and Economic Segregation in US Metropolitan Areas.” In The Oxford
Handbook of Economic Geography, edited by Gordon L. Clark, Maryann P. Feldman
and Meric S. Gertler. Oxford University Press.
158
Kantrowitz, Nathan. 1973. Ethnic and Racial Segregation in the New York Metropolis. New
York: Prager.
Kasarda, John. 1985. “Urban Change and Minority Opportunities.” In The New Urban Reality,
edited by Paul E. Peterson. Washington, DC: Brookings Institution.
——. 1988. “Jobs, Migration, and Emerging Urban Mismatches.” In Urban Change and
Poverty, edited by Michael G. H. McGeary and Laurence E. Lynn, Jr. (148-98).
Washington, DC: National Academy Press.
Katz, Bruce, and Jennifer Bradley. 1999. “Divided We Sprawl.” The Atlantic Monthly 284(6):
26-42.
Kingsley, G. Thomas, and Kathryn L.S. Pettit. 2002. “Population Growth and Decline in City
Neighborhoods.” In the Neighborhood Change in Urban America Series. Washington,
DC: The Urban Institute.
——. 2003. Concentrated Poverty: A Change in Course. Washington, DC: The Urban Institute.
Kunstler, James Howard. 1993. The Geography of Nowhere: The Rise and Decline of
America’s Man-Made Landscape. New York: Simon and Schuster.
Levy, Frank. 1995. “The Future Path and Consequences of the U.S. Earnings/Education Gap.”
Economic Policy Review 1(1): 35-41.
Lucy, William H., and David L. Phillips. 2000. Confronting Suburban Decline. Washington,
DC: Island Press.
Mallach, Alan. 1984. Inclusionary Housing Programs. New Brunswick, NJ: Rutgers University
Center for Urban Policy Research.
——. 1996. “The Age of Extremes: Concentrated Affluence and Poverty in the Twenty-first
Century.” Demography 33(4): 399.
159
Massey, Douglas S., and Nancy A. Denton. 1987. “Trends in the Residential Segregation of
Blacks, Hispanic, and Asians: 1970-1980.” American Sociological Review 52: 802-25.
——. 1993. American Apartheid: Segregation and the Making of the Underclass. Cambridge,
MA: Harvard University Press.
Massey, Douglas S., and Mitchell L. Eggers. 1990. “The Ecology of Inequality: Minorities and
the Concentration of Poverty, 1970-1980.” American Journal of Sociology 95: 1153-88.
——. 1993. “The Spatial Concentration of Affluence and Poverty During the 1970s.” Urban
Affairs Quarterly 29: 299-315.
Meyer, Bruce D., and Dan T. Rosenbaum. 1999. “Welfare, the Earned Income Tax Credit, and
the Labor Supply of Single Mothers.” JCPR Working Paper Series. World Wide Web
page
<www.jcpr.org/wp/wpdownload.cfm?pdflink=wpfiles/eitc_meyer_rosenbaum_update2
2001.pdf>
Mills, Edwin S. 1972. Studies in the Structure of the Urban Economy. Baltimore, MD: The
Johns Hopkins Press.
Musterd, Sako, and Wim Ostendorf. 1998. “Segregation, Polarisation and Social Exclusion in
Metropolitan Areas.” In Urban Segregation and the Welfare State: Inequality and
exclusion in Western Cities, edited by Sako Musterd and Wim Ostendorf (1-14). New
York: Routledge.
Orfield, Myron. 1997. Metropolitics: A Regional Agenda for Community and Stability.
Washington, DC: Brookings Institution, and Cambridge, MA: Lincoln Institute of Land
Policy.
Park, Robert E. 1926. “The Urban Community as a Spatial Pattern and a Moral Order.” In The
Urban Community, edited by E. W. Burgess. Chicago, IL: University of Chicago Press.
pp. 3-18.
Pendall, Rolf, and John I. Carruthers. 2003. “Does Density Exacerbate Income Segregation?
Evidence from U.S. Metropolitan Areas, 1980 to 2000.” Housing Policy Debate 14(4):
541-89.
160
The PolicyLink. 2003. “Expanding Housing Opportunity in Washington, DC: The Case for
Inclusionary Zoning.” A PolicyLink Report. Oakland, CA.
Popenoe, David. 1979. “Urban Sprawl: Some Neglected Sociology.” Sociology and Social
Research 31(2): 181-88.
Rector, Robert. 2002. “Housing Policy and Welfare Reform.” Policy Research and Analysis.
The Heritage Foundation. World Wide Web page
<http://www.heritage.org/Research/Welfare/Test050102.cfm>
Regional Energy Profile. 2001. The Energy Information Administration. World Wide Web
page <http://www.eia.doe.gov/emeu/reps/maps/us_census.html>
Ricketts, Erol R., and Isabel V. Sawhill. 1988. “Defining and Measuring the Underclass.”
Journal of Policy Analysis and Management 7(2): 316-25.
Rodgers, William M. III. 2005. “Black Workers Need More Than An Economic Boom.” The
Heldrich Center for Workforce Development Working Papers. World Wide Web page
<www.heldrich.rutgers.edu/Resources/Publication/166/WP_05.pdf>
Schenk, Robert. 2002. “Measuring Income Distribution.” World Wide Web page
<http://ingrimayne.saintjoe.edu/econ/AllocatingRationing/MeasuringIncomeDist.html>
Schiller, Timothy. 2004. “Sprawl: What’s in a Name?” Business Review Q4: 26-38.
Schneider, William. 1992. “The Suburban Century Begins.” The Atlantic Monthly 270(1): 33-
44.
Shlay, Anne B. 2002. Review Comments. In Urban Sprawl: Causes, Consequences and Policy
Responses, edited by Gregory D. Squires. Washington, DC: The Urban Institute Press.
Sierra Club. 1998. “Sprawl: The Dark Side of the American Dream.” World Wide Web page
<http://www.sierraclub.org/sprawl/report98/what.html>
Squires, Gregory D. 1994. Capital and Communities in Black and White: The Intersections of
Race, Class, and Uneven Development. Albany: SUNY Press.
——. 2002. “Urban Sprawl and the Uneven Development of Metropolitan America.” In Urban
Sprawl: Causes, Consequences and Policy Responses, edited by Gregory D. Squires.
Washington, DC: The Urban Institute Press.
161
Staley, Samuel R. 1999. “The Sprawling of America: In Defense of the Dynamic City.” Policy
Study No. 251. Reason Public Policy Institute. World Wide Web page
<http://www.rppi.org/ps251.html>
Taeuber, Karl E., and Alma F. Taeuber. 1965. Negroes in Cities: Residential Segregation and
Neighborhood Change. Chicago: Aldine.
Talen, Emily. 2002. “The Social Goals of New Urbanism.” Housing Policy Debate 13(1): 165-
88.
Tumin, Melvin M. 1969. In A Dictionary of the Social Sciences, edited by Julius Gould and
William L. Kolb. New York: The Free Press.
U.S. Bureau of the Census. 1992. Studies in the Distribution of Income. Series P60-183.
Washington, DC: U.S. Government Printing Office.
——. 1993. Money Income of Households, Families, and Persons in the United States: 1992.
Current Population Reports, Series P60-184. Washington, DC: U.S. Government
Printing Office.
U.S. Department of Housing and Urban Development. 2000. The State of the Cities 2000.
Washington, DC: U.S. Department of Housing and Urban Development.
Unwin, D. J. 1996. “GIS, Spatial Analysis and Spatial Statistics.” Progress in Human
Geography 20(4): 540-441.
Weiss, Marc A. 2002. “National Housing Policy in the U.S. for the 21st Century.” Prague
Institute for Global Urban Development. Prague, Czech Republic and Washington, DC.
——. 1987. American Neighborhoods and Residential Differentiation. New York: Russell Sage
Foundation. (Census Monograph Series.)
162
Wilson, William J. 1980. The Declining Significance of Race: Blacks and Changing American
Institutions. 2nd ed. Chicago, IL: University of Chicago Press.
——. 1987. The Truly Disadvantaged: The Inner City, The Underclass, and Public Policy.
Chicago, IL: University of Chicago Press.
——. 1996. When Work Disappears: The World of the New Urban Poor. New York: Alfred A.
Knopf.
Wirth, Louis. 1954. In Encyclopaedia of the Social Sciences. Volumes XIII-XIV. The
MacMillan Company.
Wright, Gwendolyn. 1983. Building the Dream. Cambridge, MA: MIT Press.
Zoloth, Barbara S. 1976. “Alternative Measures of School Segregation.” Land Economics 52:
278-93.
VITA
Rui Yang was born in Beijing, China May 25th, 1977. She completed primary
education in 1995. She earned her Bachelor of Arts in Economics in 2000 at Peking University
in Beijing, China. She earned her Masters of Sciences in Applied Economics in 2003 at the
University of Texas at Dallas. In August 2000 she began her doctoral studies in public policy
and political economy at the University of Texas at Dallas. She worked as a teaching assistant
and research assistant for the School of Social Sciences at the University of Texas at Dallas
from August 2000 till now. After graduation, she will join the Workers Compensation