Separate When Equal? Racial Inequality and Residential ...public.econ.duke.edu/~hf14/WorkingPaper/segregation/seg1_old.pdf · Residential Segregation Patrick Bayer Hanming Fang Robert

Separate When Equal? Racial Inequality and

Residential Segregation�

Patrick Bayer Hanming Fang Robert McMillan

January 13, 2005

Abstract

Conventional wisdom suggests that residential segregation will fall when racial di¤erences

in education and other sociodemographics decline. This paper argues that this prediction may

not be accurate. We identify, both theoretically and empirically, a forceful mechanism that may

lead to persistent and even increasing residential segregation as racial di¤erences in sociodemo-

graphics decline. The key feature of the mechanism relates to the emergence of middle-class

black neighborhoods as the sociodemographics of blacks improve. We show that, middle-class

black neighborhoods are in short supply given the current black sociodemographics in many

U.S. metropolitan areas, forcing wealthy blacks either to live in white neighborhoods with high

levels of neighborhood amenities or in more black neighborhoods with lower amenity levels.

Increases in black sociodemographics in the metropolitan areas will lead to the emergence of

new middle-class black neighborhoods, relieving the prior neighborhood supply constraint and

leading to increases in residential segregation. We present cross-MSA evidence from the 2000

Census indicating that this mechanism does in fact operate: as the proportion of highly educated

blacks in an MSA increases, the segregation of educated blacks, and all blacks more generally,

goes up. We also present evidence against leading alternative hypotheses.

Keywords: Segregation, Racial Sorting, Racial Inequality, Neighborhood Formation.

JEL Classi�cation Numbers: H0, J7, R0, R2.

�We are grateful to Caroline Hoxby, Kim Rueben, and Jacob Vigdor as well as conference/seminar participants

at the NBER and Yale for helpful comments and suggestions. We are responsible for all remaining errors. Contact

Addresses: Bayer and Fang are at the Department of Economics, Yale University, P.O. Box 208264, New Haven, CT

06520-8264; McMillan is at the Department of Economics, University of Toronto, 150 St. George Street, Toronto,

Ontario M5S 3G7, Canada.

1 Introduction

There is a large literature on the causes and consequences of racial segregation (Massey and

Denton [22], Cutler and Glaeser [9], Cutler, Glaeser and Vigdor [10]). Many researchers have

attempted to evaluate the contributions of various socioeconomic variables to the racial segrega-

tion. For example, Bayer, McMillan and Reubin [2] used restricted-access 2000 Census microdata

to show that a set of sociodemographic variables including education, income and language can

explain 30 percent of Black segregation and 93 percent of Hispanic segregation in the Bay Area

housing market.1 The basic idea in this literature is simple. Given that demand for housing and

neighborhood quality increases with income and that whites earn more than blacks in general,

neighborhood with higher quality houses will be primarily occupied by white households. More

generally, racial di¤erences in socioeconomic characteristics a¤ect households�residential choices,

some racial segregation would emerge even in the absence of any sorting on the basis of race. The

implicit conclusion from this literature is that reductions in racial di¤erences in income and other

important sociodemographics would decrease the level of residential segregation.

In this paper, we argue that the conventional wisdom may not be accurate. We argue that

improvements in the blacks�socioeconomic characteristics (proxied by their level of education) in

an MSA may lead more segregation of educated blacks and all blacks more generally. Our analysis

is motivated by three empirical observations about the current state of racial segregation in the U.S.

(see Section 2). First, in almost every metropolitan-area, there are few, if any, neighborhoods with

both high fractions of blacks and highly-educated households. The shortage of such middle-class

(i.e., with high average education) black neighborhoods forces highly-educated black households to

choose between predominantly black neighborhoods with low average education and predominantly

white neighborhoods with high levels of education. Second, faced with limited choice set of neigh-

borhood alternatives, highly-educated blacks do live in a very diverse set of neighborhoods: while

a fraction live in neighborhoods with very few other black households and many college-educated

neighbors, many live in neighborhoods that have a high fraction of black households and very few

other college-educated households. Third, most of the middle-class black neighborhoods are located

in metropolitan areas with a signi�cant proportion of highly-educated blacks.

The above empirical observations suggest the following. First, neither race nor average ed-

ucation of the neighborhood is all-important in the location decisions of highly-educated blacks

1See Miller and Qingley [24] for an earlier study using similar methodology. Sethi and Somanathan [29] proposed

a di¤erent method for the decomposition of segregation measures into one component that can be attributed to the

e¤ect of racial income disparaties alone, and another component that combines the e¤ects of neighborhood preferences

and discrimination.

1

(we present more evidence of preferences over race and neighborhood characteristics in Section

3). This further implies that highly-educated blacks may have preferred to live in highly-educated

majority-black neighborhoods � alternatives most of the metropolitan areas in the U.S. do not

currently. Second, middle-class majority-black neighborhoods, currently only existing in a handful

of metropolitan areas, are more likely to emerge when the number of highly-educated blacks in-

creases in the metropolitan area. The emergence of middle-class black neighborhoods will then lead

to more residential segregation because these neighborhoods are highly attractive to middle-class

black households. Section 3 presents a simple and stylized model of household residential choice

and neighborhood formation that captures the essence of the above mechanism.2

Our empirical analysis takes seriously the mechanism of decentralized residential sorting and

neighborhood formation, by examining how changes in the structure of the population within a

metropolitan area a¤ect the way that households sort on the basis of race and education.3 Our

central hypothesis relates to whether the relative exposure of highly educated black households to

blacks is an increasing or decreasing function of the education level of blacks in the metro area.4

Using 2000 Census data from 277 US metropolitan-areas (MSA or CMSAs) and summarized at the

tract level, the empirical analysis involves a series of regressions that relate the racial-educational

composition of an individual�s tract to an individual�s own race-education category, a set of MSA

�xed e¤ects, and interactions of individual and MSA race-education characteristics. The results

reveal that relative to other households in the MSA, highly-educated blacks are increasingly exposed

to other blacks as the education level of blacks in the metropolitan area increase. This change is

driven primarily by a large relative increase in exposure to other highly-educated blacks and is

more than completely o¤set by a decrease in exposure to highly-educated whites. These changes

are likely to result in a slight decrease in the average level of education in the neighborhoods in

which highly-educated blacks reside. At the same time, highly-educated blacks are also increasingly

exposed to less-educated blacks and vice-versa. This e¤ect is consistent with the comparative-statics

prediction of our model when highly-educated blacks in an MSA increase from moving from a low

to a moderate proportion in an MSA.

In a nutshell, Section 4 established a positive correlation at the MSA level between the edu-

cational attainment of blacks and the segregation of highly-educated blacks.5 Section 5 attempts

2A recent paper by Sethi and Somanathan [30] presents a di¤erent model in which they show that racial segregation

and income inequality does not exhibit a monotonic relationship. Further discussions of this paper is in Section 3.3 In our analysis, we use education as a proxy for socioeconomic characteristics more generally.4Note that the inclusion of MSA �xed e¤ects in the regression absorbs any mechanical increase in same-race

exposure due to the changing population composition.5Our empirical analysis exploits cross-MSA variations in the proportion of highly-educated blacks; as such, our

2

to empirically distinguish our �ndings from several other plausible mechanisms. We �rst rule out

reverse causality by appealing to the �ndings from Cutler and Glaeser [9].6 They found that blacks

aged 20-30 in more segregated areas have signi�cantly worse outcomes, including educational out-

comes, than blacks of the same age group in less segregated areas. It is worth emphasizing that our

�ndings do not necessarily contradict Cutler and Glaeser [9]. First, they focused on the educational

outcomes of blacks aged 20-30 while we focus on the educational attainment of all blacks in the

MSA; second, we use individual-speci�c exposure rates as our measure of segregation while they

use MSA-level dissimilarity index.7 We can also rule out statistical discrimination in the hous-

ing market or mortgage lending as the explanation to our empirical �ndings. Standard models of

statistical discrimination generally predict that, as the average characteristics of blacks (average

educational attainment, for example) in the MSA improve, highly-educated blacks would less likely

be discriminated against in both the housing market and in mortgage applications. To the extent

that statistical discrimination in the housing market and in mortgage application still exist, the

mechanism we identify may in fact be operating more forcefully than our estimates would imply. A

third alternative explanation is the selection bias where the primary concern is that highly-educated

blacks that select into MSAs with a higher fraction of educated blacks have a stronger taste for

segregation. We address this concern by decomposing the current metropolitan sociodemographic

composition for each household into a lagged measure corresponding to the MSA where the house-

hold lived �ve years ago and the di¤erence between the current and lagged measure. Including

both the lagged and di¤erence measures in an analogous set of regressions reveals that the active

selection related to movement over the past �ve years actually weakens rather than strengthens our

main �ndings, implying that selection across MSAs is not likely to be a signi�cant factor driving

results are not directly related to time-series segregation measures conducted by Cutler, Glaeser and Vigdor [10] and

Glaeser and Vigdor [15]. However, our �nding is consistent with Glaeser and Vigdor�s [15] �nding that �Segregation

decline most sharply in places that ... blacks made up a small portion of the population in 1990. Segregation remains

extreme in the largest metropolitan areas."6Earlier related literature in this line include Ihlanfeldt and Sjoquist [17] and O�Regan and Quigley [25].7The dissimilarity index was proposed by Duncan and Duncan [?] measures the fraction of blacks that would

have to switch areas to achieve an even racial distribution citywide (see Cutler, Glaeser and Vigdor [10] for more

discussions). It is an aggregate-level measure. The exposure rate has the advantage over dissimilarity index in that

it can be speci�c to individuals. It construction can be illustrated simply as follows (see Bayer, McMillan and Ruben

[2]). Let rij be a set of indicator variables that take value 1 if household i is of race j and 0 otherwise. Let Rik be the

fraction of households of race k in household i0s neighborhood (the Census tract, for example). The average exposure

of households of race j to households of race k; denoted by Ejk; is

Ejk =

Pi r

ijR

ikP

i rij

:

3

our results. The �nal alternative explanation relates to omitted variable bias and selection bias,

the concern being that historic patterns of black settlement, migration, and segregation in the U.S.

might give rise to a spurious correlation between metropolitan sociodemographic composition and

segregation. While this explanation can not be addressed to complete satisfaction due to data

limitations, we show that our main �ndings are completely robust to the inclusion of a full set of

interactions with metropolitan size and region.

Our results have several important implications. First, they suggest caution in the conventional

wisdom that racial segregation will necessarily decrease with the decline of the racial di¤erences

in socioeconomic characteristics.8 The mechanism uncovered in our analysis in fact suggests that,

overall increase in blacks�educational attainment may lead to a decrease in residential segregation

if highly-educated blacks are dispersed in many, instead of concentrated in few, metropolitan areas.

This echoes Glaeser and Vigdor�s [15] �nding that segregation is lowest among growing cities in

the West where there is not yet a high concentration of highly-educated blacks. Our �nding is also

related to Wilson [34]. He argues that reductions in institutional discrimination in the housing

market in the middle of the 20th century led to large-scale reductions in the exposure of less-

educated to more-educated blacks, as more-educated blacks left the inner city neighborhoods to

where they were formerly restricted. Our �ndings suggest that this trend may not have been severe

in cities in which the black population was more educated initially and may partially reverse itself

as the black population becomes relatively more educated over time.

The remainder of the paper is structured as follows. Section 2 empirically documents the neigh-

borhood choice sets in U.S. metropolitan areas; Section 3 presents a simple model of neighborhood

formation that highlights the key features of the mechanism underlying our empirical results; Sec-

tion 4 presents our main empirical �nding that the exposure of highly-educated blacks and all

blacks more generally to other blacks increases as the proportion of highly-educated blacks in the

metropolitan area increases; Section 5 evaluates leading alternative hypotheses; �nally, Section 6

concludes.

2 Neighborhood Choice Sets in US Metropolitan Areas

In this section we present some empirical facts about the neighborhood choice sets in U.S.

metropolitan areas using Census Tract Summary Files from 2000 Census. In our analysis, a �neigh-

borhood�corresponds to a �census tract,�which typically contains 3,000 to 5,000 individuals. The

Census Tract Summary Files provides information on the distribution of education by race for each

8This conventional view was embraced by many scholars in this literature (see Durlauf [12], Wilson [34] and Mayer

[23]).

4

Census tract. We characterize the race and educational attainment of households as that of the

head of household and focus speci�cally on non-Hispanic black and non-Hispanic white households

throughout our analysis.9 Educational attainment of the head of the household is used to proxy

the socioeconomic status of the household more generally. We characterize each neighborhood

in a metropolitan area in two dimensions: the fraction of black households, and the fraction of

highly-educated (i.e. educational attainment is college or above) households.

We establish four empirical facts about neighborhood choices sets:

� Highly-educated black households constitute a small fraction of the population living in typicalmetropolitan areas;

� Neighborhoods that combine high fractions of both college-educated and black householdsare in extremely short supply in almost every metropolitan area;

� College-educated blacks choose to live in a very diverse set of neighborhoods in each metropol-itan area;

� Middle-class black neighborhoods are concentrated in few metropolitan areas with sizeable

highly-educated black households.

[Table 1 About Here]

Table 1 describes the joint distribution of education and race for black and white households.

Based on our race de�nitions, black and white households respectively constitute 11.1 and 69.5

percent of the U.S. population that reside in metropolitan areas. Among black households, 15.4

percent are headed by an individual with a college degree, while the comparable number for white

households is 32.5 percent, and for all U.S. households, 27.7 percent.


Table 2 documents the number of tracts in the U.S. by the percentage of households with

a college degree and the percentage of households that are black and white, respectively. Panel

1 describes the number of tracts in which more than 0, 20, 40, 60, and 80 percent of head of

households are college-educated, respectively. Panel 2 reports the number of tracts in each of these

categories that contain a minimum fraction of black households equal to 20, 40, 60, and 80 percent,

9The vast majority of households that checked two races can be characterized as either Hispanic or non-Hispanic

Asian or Paci�c Islander. Other households that checked two or more races - a very small fraction overall - were

dropped from this analysis.

5

respectively. As the corresponding numbers show, a much smaller fraction of the tracts with a high

fraction of black households have a high fraction of households with a college degree. For example,

while 23 percent of all tracts are at least 40 percent college educated (a number comparable to the

fraction of U.S. households with a college degree), only 2.5 percent of tracts that are at least 40

percent black are at least 40 percent college educated, and only 1.1 percent of tracts that are at

least 60 percent black are at least 40 percent college educated. Panel 2 of Table 2 shows analogous

numbers for white households, reporting the number of tracts in the U.S. that meet the education

criterion described in each column heading subject to a minimum fraction of white households equal

to 20, 40, 60, and 80 percent, respectively. The corresponding numbers show a markedly di¤erent

pattern of neighborhood choices for white households, with a greater fraction of neighborhoods

with at least 40, 60, and 80 percent white households meeting each education criterion.


While Table 2 revealed a paucity of neighborhoods with high fractions of both black and college-

educated households in the U.S. as a whole, Table 3 further shows that such tracts, to the extent

that they exist, are in fact concentrated in only a handful of metropolitan areas, most notably

Washington, DC. This implies that the supply of such neighborhoods in most metropolitan areas is

even more limited. Table 3 illustrates, for example, that of the 44 tracts (less than 0.1 percent of all

tracts) that are at least 60 percent black and 40 percent college-educated, 13 are in the Washington

DC PMSA, 8 in Detroit, 6 in Los Angeles, and 5 in Atlanta. Almost 75 percent of these tracts can

thus be found in one of only four PMSAs. Of the 142 tracts that are at least 40 percent black and

40 percent college-educated, almost two-thirds are in the PMSAs listed above along with Chicago

and New York.

Tables 2-3 taken together show clearly that while neighborhoods that combine high fractions of

both college-educated and white households are amply supplied in all metropolitan areas, neighbor-

hoods that combine high fractions of both college-educated and black households are in extremely

short supply. This suggests that college-educated black households in most metropolitan areas may

face a trade-o¤ between living with other blacks versus other college-educated neighbors.

[Figure 1 About Here]

To graphically demonstrates the potential trade-o¤ faced by highly-educated black households,

Figure 1 shows the scatterplots of neighborhoods in four metropolitan areas: Boston, Dallas,

Philadelphia, and St. Louis. In each scatterplot, a circle represents a Census tract and its co-

ordinates represent the fraction of highly-educated households (vertical axis) and the fraction of

6

black households (horizontal axis) in the tract. The diameter of the circle is proportional to the

number of highly-educated black households in the tract, thus the largest circles correspond to the

tracts where highly-educated blacks are most likely to live.10 For these four metropolitan areas, the

scatterplots demonstrate the short supply of neighborhoods that combine high fractions of both

highly-educated and black households, neighborhoods that would have appeared in the north-east

corner of the plot. Consequently, highly-educated black households may face a trade-o¤ when

making their residential choices.

Figure 1 also demonstrates that, facing the constrained choice set, highly-educated black house-

holds do choose to live in a diverse set of neighborhoods: while a sizeable fraction of highly-educated

blacks in each of the �ve MSAs chooses neighborhoods with few black and many highly-educated

neighbors �neighborhoods in the north-western corner of the plots, another sizeable fraction choose

neighborhoods with many black and few highly-educated neighbors �neighborhoods in the south-

western corner of the plots.


Panel A of Table 4 summarizes the characteristics of neighborhoods chosen by highly-educated

blacks in metropolitan areas throughout the U.S. We �rst rank highly-educated black households

in each metropolitan area by the fraction of blacks in their Census tract and assign households

to their corresponding quintile of this distribution. This corresponds to drawing four vertical

lines in the scatterplot for each metropolitan area such that an equal number of highly-educated

black households are in each of the quintiles. Panel A of Table 4 then summarizes the average

fractions of black and Highly-educated households in the tract corresponding to the quintiles of

this distribution, averaged over all U.S. metropolitan areas. The �rst column of Panel A, for

example, the 20 percent college-educated black households with the smallest fraction of other

black households in their metropolitan are (the bottom quintile) live in neighborhoods that are

on average 5.7 percent black and 38.0 percent highly-educated. The numbers corresponding to

di¤erent quintiles show a clear trade-o¤ for highly-educated black households between the fraction

of their neighbors that are black and the fraction that are highly-educated: the average fraction of

highly-educated neighbors falls from 38.0 percent for those college-educated blacks living with the

smallest fraction of black neighbors to 13.8 percent for those living with the largest fraction. Panel

B of Table 4 reports analogous numbers for college-educated white households. The comparison

of Panels A and B reveal that, college-educated blacks in each metropolitan area who live in the

bottom quintile of tracts in terms of the smallest fraction of other blacks have roughly the same

10Note that tracts that do not contain any highly-educated blacks do not show up in these scatterplots.

7

fraction of college-educated neighbors as college-educated whites do on average; however, college-

educated blacks living in the top quintile of tracts with the greatest fraction of other blacks have

only about one-third of that fraction of highly-educated neighbors.

We argue that such a high fraction of college-educated black households in U.S. metropolitan

areas choose segregated neighborhoods with relatively low average education attainments suggest

that race remains an important factor in the location decisions of a large number college-educated

black households. In Section 3 we present more evidence for race preferences and propose a simple

mechanism based on the emergence of highly-educated black neighborhoods to argue that such

racial preferences may lead to more segregation of highly-educated black households as the average

education of the block population in a metropolitan area increases. Speci�cally, we argue that

when the number of highly-educated black households in the metropolitan area increase, a greater

supply of neighborhoods that combine high fractions of both black and college-educated neighbors

may form. Figure 2 depicts the scatterplots of neighborhoods in Atlanta, Chicago, Detroit and

Washington, D.C. As we showed in Table 3, these metropolitan areas contain a sizeable number of

college-educated black households. Figure 2 shows that they supply substantially greater number

of neighborhoods combining relatively high fractions of both black and highly-educated house-

holds, and thus relax the constraint of neighborhood choice set for highly-educated blacks. As the

neighborhood supply constraint relaxes for highly-educated blacks, highly-educated blacks may be

increasingly exposed to other black households.

3 A Model

In this section, we present a simple model of residential choice and endogenous neighborhood

emergence. This simple model formalizes our idea that the set of neighborhood available to highly-

educated blacks will expand as the fraction of highly-educated blacks increase in a metropolitan

area, leading to more segregation. Sethi and Somanathan [30] presented an alternative model

in which they show that low racial inequality is consistent with extreme and even rising levels

of segregation in cities where the minority population is large. Their model does not explicitly

emphasize the idea of neighborhood emergence since they �x the total number of neighborhoods

exogenously, where our model emphasize the emergence of new middle-class neighborhoods, a fact

consistent with the empirical facts documented in Section 2.

Basic Ingredients. Before describing the details we highlight three key ingredients of the model

that drive our results. The �rst key ingredient is that black households� preferences are such

8

that, on net, they prefer to live near others of the same race and education level.11 There is

ample empirical evidence that individuals prefer to live in neighborhoods where they own race is in

majority.12 For example, in the Multi-City Survey of Urban Inequality (MCSUI) respondents were

shown a card representing a neighborhood with �fteen houses (in three parallel rows of �ve houses

each), and were asked to illustrate the racial composition of their �ideal� neighborhoods where

they were presumed to live in the house located at the center of the middle row. Using MCSUI

conducted between 1993-1994 in Atlanta, Detroit, and Los Angeles metropolitan areas, Ihlanfeldt

and Sca�di [17] found that, 35-43 percent of blacks designates the all-black neighborhood or the

mostly black neighborhood (eleven blacks and four whites) as their top choice; and 81-92 percent

of the blacks chose all black or mostly black neighborhoods as one of their top two choices.13

Our assumption that black households prefer neighbors with the same educational attainment

needs more explanation. Indeed intuitive all households may prefer to live with highly-educated

neighbors, due to the positive externality in human capital production (see Benabou [5] and Cutler

and Glaeser [9] for example). Our assumption is sensible only because housing prices �which we

abstract from �would have most likely capitalized such positive externalities. Many sorting models

would generate the prediction that, ceteris paribus, less-educated households are not willing to pay

as much as highly-educated households to live with more educated neighbors. Thus we consider our

assumption on the preference to live with neighbors of same educational attainment as a convenient

reduced-form, which allows us to place the role of housing prices in the background of the analysis.

The second key ingredient of our model is the notion of critical neighborhood size. In the

model below we capture the notion of critical neighborhood size by assuming that each resident

in a neighborhood has to incur a cost that decreases with its total number of residents. Multiple

interpretations can be given for this formulation. One interpretation is that the average cost of

providing both formal and informal local public goods decreases in neighborhood size, at least over

some range; alternatively, it can capture the idea that a larger population can sustain more local

options in retail, television, restaurants, newspaper, and the internet, both in quantity and quality

(see, e.g., Berry and Waldfogel [4] and reference cited therein).

The third ingredient of our model is some degree of idiosyncratic location preferences, unrelated

11Schelling [27][28] is the �rst to note that individual racial preferences can lead to segregation.12Cornell and Hartmann [8], Farley et al [13], O�Flaherty [21] and Lundberg and Startz [20] provide various

theoretical arguments about why individuals might care about the racial composition of their neighborhoods.13Also see Vigdor [32] and Charles [6][7] for related evidence. It is important to emphasize that such evidence has

to be at best considered as suggestive, as the MCSUI survey questions makes no mentions of neighborhood amenties,

housing prices, or other factors that might in�uence residential choices. Thus such evidence does not necessarily

reveal fundamental racial preferences. King and Mieszkowski [19], Yinger [35] and Galster [14] found evidence of

segregation preferences based on housing prices or rents.

9

to sorting on the basis of education or race, in the households�residential choices. We capture this

heterogeneity by assuming that households have employment locations distributed in space and

would prefer to commute shorter distances. Such assumption is standard in the �spatial mismatch�

literature (see Kain [18], Ross [26] and Weinberg [33]).

Model. Consider a metropolitan area located on a straight line with length 2, represented by

the interval [�1; 1]. The population density in the metropolitan area is given by N > 0, thus

its total population is 2N: There are two racial groups r 2 fb; wg and a proportion �w 2 (0; 1)is white and the remaining proportion �b = 1 � �w is black. Moreover, individuals within eachracial group di¤er in their educational attainment: a fraction �r 2 (0; 1) of race-r individuals arehighly-educated (denoted by type-h) and the remaining fraction 1 � �r are of the low-educationtype (denoted by type-l).14 Cross-race inequality in socioeconomic characteristics is re�ected by

the di¤erence �w � �b: Clearly for all metropolitan areas in the U.S., the relevant case is �w > �b.Thus a narrowing in the racial gap in educational attainment can be represented by an increase in

�b while keeping �w �xed.

For simplicity, we assume that whites�residential locations are �xed: at each endpoint of the

line, there are two communities, one for highly-educated whites (called communities HW and HW�)

and one for less-educated whites (called communities LW and LW�). We focus our analysis on the

residential location choices of black households and the emergence of black neighborhoods.

We model idiosyncratic locational preferences of the blacks by assuming that their job locations

are uniformly distributed on the straight line. Commuters experience a cost of � > 0 per unit

distance between their work and place of residence.

There is a cost of maintaining a community, and the average per-resident cost is given by c(n)

where n is the number of residents in the community.15 Naturally we assume that c decreases in n:

We now describe black households� preferences. Consider a black household with education

e 2 (l; h) whose job location is at point z 2 [�1; 1] on the straight-line. Its utility from living in a

community j 2 J where J is the set of available communities to be determined in equilibrium, isgiven by:

u(j; z; e) = � [pb(j) + 1pw(j)] + � [pe(j) + 2pe0(j)]� �D(j; z)� c (n(j)) ; (1)

where e0 6= e is the other education category; pr(j) is the proportion of residents in community j ofrace r; pe(j) is proportion of residents in j with education attainment e; D(j; z) is the commuting

distance between community j and z�s job location; n(j) is the number of residents in community

14The heterogeneity could equally be in terms of income.15Technically this rules out tiny enclaves of individuals claiming to form a neighborhood of their own.

10

j; � > 0; � > 0; 1 2 (0; 1) ; 2 2 (0; 1) are constants. In utility function (1), the �rst term

�[pb(j) + 1pw(j)] captures the utility from interacting with people of di¤erent races in the same

community, where 1 < 1 captures the same-race preference discussed earlier. The interpretation

of the second term �[pe(j) + 2pe0(j)] is more subtle. As we explained previously, it is meant to

capture, in a reduced form, the idea that highly-educated workers will on net (taking into account

both human capital externality and housing price) prefer to live in more expensive neighborhoods

with many other highly-educated residents, while less-educated workers will prefer on net to live in

cheaper neighborhoods with other less-educated residents.

An equilibrium of this simple model is characterized by the set of neighborhoods J� (including

the existing neighborhoods HW, HW�, LW, LW�) and the residential choices of all black households

such that: (1) given J�; all black households�residential choices is utility-maximizing; (2) there is

a positive measure of residents in all neighborhoods j 2 J�: Note that we do not need to directlyimpose a threshold neighborhood size in our model. The existence of the four white neighborhoods,

together with the assumption that c (n) is decreasing in n; endogenously ensures that small enclaves

of black households will not claim to form their own neighborhood.

The equilibrium set of neighborhoods J� depends on the parameters of the model. We are

particularly interested in how J� is a¤ected by an increase in �b �the fraction of highly-educated

black households in the metropolitan area. Consider an equilibrium in which a single black com-

munity, community B, emerges at point 0. Clearly community B, were it to emerge, will consist

of blacks whose job locations are close to point 0. Thus given J� = {HW, HW�, LW, LW�, B},

black households�optimal residential choices can be characterized by a pair fx�h; x�l g such that allhighly-educated (less-educated, respectively) black households will choose to live in community B

if and only if their job location z satis�es jzj � x�h (jzj � x�l ; respectively). The marginal types

fx�h; x�l g can be determined from the indi¤erence conditions (see Appendix A for details). Figure 3

depicts this type of equilibrium when �b is small.16


Imagine that we increase the fraction of highly-educated blacks �b from an initial low level.

First, note that as �b increases, the proportion of highly-educated blacks in community B, ph (B) ;

will increase even if the thresholds fx�h; x�l g were hypothetically unchanged. As ph (B) increases,

16 If such an equilibrium exists with a su¢ ciently small �b, one can show that xl� > xh�. The reason is simple:

when �b is small, community B is necessarily a predominantly lowly-educated all-black community. Because 2 < 1,

the utility for a lowly-educated black from community B is always higher than that for a highly-educated black at

any job location. Thus lowly-educated blacks are more willing to commute to community B. This is not important

for the analysis but explains the ranking of x�l and x�h in Figure 3.

11

community B becomes more attractive vis-à-vis community HW for highly-educated blacks. As a

result, the marginal type of highly-educated black who commutes to community B, x�h; will increase.

Thus the probability of a highly-educated blacks living in all-black community B with less-educated

blacks will initially increase in �b. The results for less-educated blacks are more ambiguous. On

the one hand, community B becomes more educated. On the other hand, the increase of the total

population in community B drives down the average community cost c. Thus, whether or not

community B becomes more desirable for less-educated blacks is indeterminate. It is possible that

exposure of highly- and less-educated blacks to one another may increase. Moreover, the increase

in exposure of highly-educated blacks to other highly-educated blacks comes at the expense of

exposure to highly-educated whites (see Figure 4b for the graphical illustration).


When �b is su¢ ciently high, however, a point may be reached when it becomes pro�table for

highly-educated blacks in community B to form their own community at point 0, called BH, and

leaving behind a less-educated black community BL (see Figure 4c). The exact point at which the

highly-educated black neighborhood BH emerges is determined by the balance of two forces. First,

by separating from the less-educated blacks living in community BL, highly-educated blacks have

to incur a higher per capita cost c as a result of smaller population size; second, because community

BH consists only of highly-educated blacks, the utility component ph (BH) = 1 > ph (B)+ 2pl (B)

because 2 < 1:17 This is the key insight of our simple model: a highly-educated black community

BH emerges only when the proportion of highly-educated blacks �b is su¢ ciently high. Of course,

the emergence of such communities also depends positively on the population density N , the overall

proportion of blacks in the metropolitan area �b. It also indirectly depends on the commuting cost

� and the community cost function c (n) via their e¤ects on xh�.

It is also worth pointing out that the emergence of community BH is likely to induce an ac-

celerated emigration of highly-educated blacks from community WH and WH�to community BH,

resulting in greater racial segregation in residential locations.

4 Empirical Results

We now present the main empirical analysis of this paper. We begin this section by charac-

terizing the pattern of segregation broken out by race and education in U.S. metropolitan areas.

17We abstract from the coordination problem among highly-educated blacks in their decision to form their own

neighborhood.

12

We then explore how this pattern varies with the sociodemographic composition of the metropol-

itan area, focusing especially on how the segregation of highly-educated blacks (and blacks more

generally) is a¤ected by the fraction of highly-educated blacks in the metropolitan area.

4.1 Segregation Patterns in US Metro Areas

We begin by describing the general pattern of segregation in the United States as a whole. Panel

A of Table 5 shows the average tract-level cross exposure for households in four race-education

categories (black/white; college/non-college educated) for U.S. metropolitan areas. The �rst entry

in Panel A, for example, implies that the average black household without a college degree lives in

a tract where 33.2 percent of the households are black and without a college degree. This compares

to the national average exposure to less-educated blacks of 9.4 percent. Panel B reports the average

cross exposure of households by race-education categories relative to the MSA average. The �rst

row of Panel B states that relative to an average household in the same metropolitan area, blacks

without a college degree are exposed to 19.6 percentage points more blacks without a college degree

and 2.1 percentage points more highly-educated blacks, etc. Table 5 illustrates a clear pattern of

racial segregation in U.S. metropolitan areas for highly-educated blacks as well as those with lower

levels of educational attainment.


Table 6 reports segregation patterns in a manner analogous to Panel B of Table 5, but sepa-

rately for metropolitan areas with above and below the median fraction of college-educated black

households (1.23 percent). Table 6 provides some initial evidence on how segregation patterns

vary with the sociodemographic composition of the metropolitan area. Table 6 shows that, the

relative exposure of blacks in each education category to both highly- and less-educated blacks is

signi�cantly greater in metropolitan areas with above-median fractions of highly-educated black

households. For both highly- and less-educated blacks, the average tract-level exposure to blacks

relative to the fraction of blacks in the MSAs above the median more than doubles that for MSAs

below the median.


4.2 A Regression-Based Approach

To more formally control for the sociodemographic structure of the metropolitan area, Table 7

reports the results of a series of regressions of various tract-level exposure measures on individual

13

and MSA characteristics.18 The dependent variable is listed in the head of each column. For

example, column 1 corresponds to a regression whose dependent variable is individuals�tract-level

exposure rate to highly-educated blacks. Each regression includes a complete set of controls for

individual race-education categories and MSA �xed e¤ects. The inclusion of the MSA �xed e¤ects

ensures that all of the other parameters characterize tract-level exposure relative to the MSA

average for each set of individuals. In addition, the regressions also include individual characteristics

interacted with MSA characteristics.19 The coe¢ cients on the interaction terms characterize how

tract-level exposure for various sets of individuals varies with MSA characteristics.


Due to the various interaction terms in these regressions, coe¢ cient estimates are di¢ cult to

interpret in isolation. In what follows we use these coe¢ cient estimates to present two statistical

tests. The �rst test is about whether an increase in the fraction of highly-educated blacks, holding

constant the fraction of black households, changes the relative tract-level exposure of less- and more-

educated blacks, respectively, to households in the given race-education category. This corresponds

to examining the impact of an increase in the average education level of black population. The

second test is about whether an increase in the fraction of highly-educated blacks, holding constant

the fraction of highly-educated individuals in the metropolitan area, changes the relative tract-level

exposure of less- and more-educated blacks, respectively, to households in the given race-education

category. This corresponds to increasing the fraction of the educated population that is black.


Table 8 summarizes these e¤ects for a 1 percentage point change in the fraction of college-

educated blacks in the MSA population. Panel A shows that when the fraction of college-educated

blacks in a metropolitan area increase by 1 percent at the expense of less-educated blacks, the

relative exposure of highly-educated blacks to other highly-educated blacks increase by 1 percentage

point and it is statistically signi�cant; the relative exposure of less-educated blacks to highly-

educated blacks also increase by 1.1 percentage points. Overall, the relative exposures of both

highly- and less-educated blacks to blacks increase by 4 and 6.1 percentage points respectively. This

empirical �nding is consistent with our model�s prediction when �b is in intermediate range (Figure

18We include individual level characteristics to the extent that they are available in the Census Tract Summary

Files. In practice this is equivalent to running weighted OLS where the weight is given by the number of individuals

in each race/education cell.19 If we do not include these interactions, the coe¢ cients on the individual characteristics in these regressions would

return the estimates reported in Panel B of Table 5.

14

4b), which we think is the plausible scenario for most of the U.S. metropolitan areas in the U.S.

Panel B shows that similar results holds when the fraction of highly-educated blacks is increased

by reducing the fraction of highly-educated whites (i.e., increasing the fraction of the educated

population that is black, Panel B). For both highly- and less-educated blacks, the increased relative

exposure to blacks is driven by increased exposure to blacks in both education categories. These

relative increases are o¤set by a decrease in the exposure to (especially highly-educated) whites. On

net, an increase in the average education of the black population has a slightly negative (although

marginally statistically insigni�cant) e¤ect on the average education level in the neighborhoods

that blacks reside in relative to the metropolitan area average.20


Table 9 reports results analogous to those reported in Table 8 with the exception that the

underlying measure of highly-educated is changed to include those having at least attended college.

With this de�nition, the fraction of households in U.S. metro areas that are highly-educated is

54 percent, the fraction that are both highly-educated and black is 5 percent, and the fraction of

black households that are highly educated is 45 percent. Our primary objective in examining an

alternative de�nition is to explore a de�nition of highly-educated that includes a larger fraction

of households and especially black households. A comparison of the results reported in Table 9

to those reported in Table 8 reveals a qualitatively similar pattern. With the expanded highly-

educated category, the relative increase in exposure of both highly- and less-educated blacks to

other blacks is more evenly split between highly- and less-educated blacks.

Both de�nitions of �highly-educated� reveal a pattern of increased relative exposure of both

highly- and less- educated blacks to blacks in each education category when the fraction of highly-

educated blacks in the metropolitan area increases. This pattern is consistent with the predictions

of our model corresponding to an increase in �b from small to moderate levels. With an increase

in the average education level of the black population, highly-educated blacks move on net into

20To ensure that the results of Table 7 are not driven by the form of the dependent variable that we employ, we also

conducted a series of regressions analogous to those reported in Table 7 de�ning the dependent variable as the fraction

of households in a given category in an individual�s tract divided by the fraction in the metropolitan area as a whole.

In this way, an increase in tract-level exposure to households in a given category from 6 to 12 percent following an

increase in the proportion of these households in the metro area from 3 to 6 percent would not result in an increase

in the dependent variable in this case, while it would result in a 3 percentage point increase in the dependent variable

used in the regressions reported in Table 7. The resulting parameter estimates led to a qualitatively identical set of

conclusion, thereby ensuring that our initial results are not driven by the functional form of the dependent variable.

Throughout the remainder of the paper, we present the results of regressions analogous to those reported in Table 7.

15

more segregated neighborhoods, increasing the average education level in some of the most segre-

gated neighborhoods. In terms of the scatterplots, this pattern is consistent with the formation of

segregated, black neighborhoods with mixed education levels � that is, with shift of the implicit

neighborhood supply constraint that is more upward than outward.

5 Robustness to Alternative Explanations

Our main empirical �nding can be summarized as the discovery of positive relationship at the

MSA level between the educational attainment of blacks and the segregation of highly educated

blacks. We suggested a forceful mechanism that can explain this positive correlation based on

the idea that middle-class black neighborhoods may emerge as the sociodemographics of blacks

improve.

In this section, we present arguments and robustness tests to cast doubt on alternative expla-

nations for our empirical �ndings. The �rst possible story is reverse causality. Cutler and Glaeser

[9], however, cast doubt on this explanation. They studied the impact of segregation at the MSA

level on educational outcomes for blacks aged 20-30, and �nding a large negative e¤ect.21 The

second possible story is statistical discrimination, both in housing market and in mortgage market.

However, statistical discrimination story would predict a negative correlation: as the fraction of

highly educated blacks increases in a metropolitan area, blacks in general would be less likely to

be discriminated against thus leading to less not more segregation. To the extent that statistical

discrimination exists in reality, the actual mechanism that we identify in our paper may in fact be

stronger than what our main estimates would imply. Below we consider in detail the possibility of

omitted variables and selection biases.

Omitted Variable Bias. The concern over omitted variables is that the fraction of the metropol-

itan area that is highly-educated and black might be correlated with other variables that are as-

sociated with di¤erent levels of segregation. This is a di¢ cult problem to address satisfactorily,

especially since we use cross-MSA data. Here we address this problem by adding metropolitan size

and region, two most prominent factors because of historic patterns of black settlement, migration,

and segregation in the U.S., into our regression. Table 10 reports the results of a set of regressions

analogous to those reported in Table 9 with the addition of interactions between each individual�s

race-education category and a measure of metropolitan size and four dummies for Census region

21There are other di¤erences between their paper and ours: for example, they use dissimilarity index as the

segregation measure, and the data set they used is the 1990 Census.

16

(Northeast, Midwest, South, West).22


A comparison of the results presented in Table 10 to those in Table 9 reveals a qualitatively

similar pattern in terms of both the magnitude and the statistical signi�cance of the results. In

particular, with the additional controls, the increase in the relative exposure of both highly- and

less-educated blacks to other blacks declines by 15-20 percent in magnitude, but remains highly

signi�cant. Changes in relative exposure to highly-educated neighbors also decline and remains

insigni�cant in each case. Taken together, these results give us con�dence that the main conclusions

of the paper are not driven by obvious omitted variable biases.

While we added metropolitan size and interactions in the results reported in Table 10, we still

assumed that the e¤ects of the fraction of highly-educated blacks on segregation does not depend

on metropolitan size. The critical mass story implicit in our model implies that not only the

fraction but the number of highly-educated blacks in the metropolitan area may be important for

the formation of more-educated and segregated black neighborhoods. Given the same fraction of

highly-educated black households, highly-educated black neighborhoods might more easily form in

large (in population) rather than small metropolitan areas.


Table 11 estimates separate regressions, including the additional 16 control variables added

in Table 10, for small (0-200k), medium (200-600k), and large (600k+) metropolitan areas. For

brevity, we only report results related to exposure to black households. A clear pattern emerges

in the table � the increased relative exposure of both highly- and less-educated blacks to other

blacks following an increase in average education level of the black population is much greater in

large versus small metropolitan areas. For highly-educated blacks, the magnitude of the e¤ect rises

from 0.002 in small, to 0.025 in medium-sized, and 0.40 in large metro areas. The results tend

to have higher statistical signi�cance in larger metropolitan areas. A similar pattern emerges for

less-educated black households. Interestingly, in large metropolitan areas, the increased exposure

of highly-educated blacks to other blacks is dominated by an increased exposure to other highly-

educated blacks. Thus, for this subsample, an increase in the average education of the black

population might be associated with the formation of predominantly highly-educated, segregated

black neighborhoods �i.e., the prediction of the model corresponding to an increase in the fraction

of highly-educated blacks from a moderate to large number.

22A total of 16 interaction terms are added to the regression.

17

Selection Bias. The concern about selection bias is that highly-educated blacks that select into

MSAs with a higher fraction of educated blacks may have a stronger taste for segregation. To

address this possibility, we make use of an alternative organization of the 2000 Census �the Public

Use Microdata Sample (PUMS). PUMS has the advantage over the Tract Summary Files used in

the above analysis in that observations are at the individual level, but has the disadvantage that

a less detailed level of geographic speci�city is provided. In PUMS individuals are assigned to

PUMAs, which contain greater than 100,000 households. From our perspective the key additional

variable contained in the PUMS data is the metropolitan area in which each individual resided

�ver years ago (i.e., in 1995). This variable allows us to explore whether the pattern of active

selection across metropolitan areas over this �ve-year period is in the direction of causing an over-

or under-statement in the coe¢ cients estimated in our main speci�cations.

Before exploring the selection bias issue with these data, we �rst replicate our main speci�cations

reported in Table 9 for this organization of the Census data. Table 12 shows, as one may expect

as a result of the increased geographic aggregation in this data set, a similar pattern of relative

exposures to those seen in Table 9 but at rates that are smaller in magnitude.



Table 13 explore the likely direction of selection bias in our main speci�cation by estimating the

following speci�cation. Using the metropolitan area in which each individual resided �ve years prior

to the 2000 Census, we decompose the sociodemographic composition of each individual�s current

metropolitan area into two components: the �rst component is called the �lagged measure�, which

is the composition of the metropolitan area in which that person live �ve years ago; the second

component is called the �di¤erenced measure,� which is the di¤erence between the composition

of current metropolitan area the lagged measure. For about 90 percent of the population that

did not move, the di¤erenced measure is zero; while for movers this di¤erence re�ects the change

in metropolitan area sociodemographics associated with their move. We then include distinct

interaction terms with both measures in the same speci�cation. The estimated coe¢ cients on the

lagged versus di¤erenced measures indicate the direction of the selection bias. Table 13 indicates

that the estimated coe¢ cients on the di¤erenced measures are smaller in magnitude than those

on the lagged measures for all three categories of exposure. This indicates that the active across-

metropolitan selection observed over the past �ve years leads to an understatement of the main

coe¢ cients in our main speci�cation. To the extent that selection in previous periods in time was

qualitatively similar to that over the past �ve years, we would generally expect, then, that our

18

main speci�cation understates the impact of the average education of the black population on the

segregation of both highly- and less-educated blacks.

6 Conclusion

This paper explores the relationship between metropolitan level sociodemographic composition

(particularly racial inequality in education) and residential segregation. We present theoretical

argument and empirical evidence that the conventional wisdom �which suggests that residential

segregation will fall when racial di¤erences in education and other sociodemographics decline �may

not be accurate. We show that, middle-class black neighborhoods are in short supply given the cur-

rent black sociodemographics in many U.S. metropolitan areas, forcing wealthy blacks either to live

in white neighborhoods with high levels of neighborhood amenities or in more black neighborhoods

with lower amenity levels. Increases in black sociodemographics in the metropolitan areas will lead

to the emergence of new middle-class black neighborhoods, relieving the prior neighborhood sup-

ply constraint and leading to increases in residential segregation. We present cross-MSA evidence

from the 2000 Census indicating that this mechanism does in fact operate: as the proportion of

highly educated blacks in an MSA increases, the segregation of educated blacks, and all blacks

more generally, goes up. We also present evidence against leading alternative hypotheses.

Our primary empirical analysis examines how changes in the structure of the population within

a metropolitan area a¤ect the way that households sort on the basis of race and education. The

results reveal that relative to other households in the MSA, highly-educated blacks are increasingly

exposed to other blacks as the education level of blacks in the metropolitan area increase. This

change is driven primarily by a large relative increase in exposure to other highly-educated blacks

and is more than completely o¤set by a decrease in exposure to highly-educated whites. At the

same time, highly-educated blacks are also increasingly exposed to less-educated blacks and vice-

versa. This e¤ect is consistent with the predictions of our theoretical model when moving from a

low to a moderate proportion of highly educated blacks. We also show, to the extent possible, that

our results are robust to concerns related to omitted variable and selection biases.

Our results have a number of important implications. First, they imply that racial segregation

is unlikely to disappear with convergence in racial di¤erences in socioeconomic characteristics. The

results also have implications concerning the impact of racial sorting in the housing market on the

long-run convergence of educational attainment across race. In particular, the results indicate that

given the current sociodemographic structure of U.S. metropolitan areas, increases in the average

education level of black households may result in a slight decrease in the relative exposure of both

highly- and less-educated blacks to educated neighbors. A third implication relates to the literature

19

following Wilson (1987), which demonstrates that reductions in institutional discrimination in the

housing market in the middle of the 20th century led to large-scale reductions in the exposure of

less educated to more educated blacks as more educated blacks left the inner city neighborhoods

to which they were formerly restricted. The evidence we present here suggests that this trend may

not have been severe in cities in which the black population was more educated initially and may

partially reverse itself as the black population becomes relatively more educated over time.

References

[1] Bajari, Patrick and Matthew Kahn (2003). �Estimating Housing Demand With an Application

to Explaining Residential Segregation in Cities,�NBER Working Paper 9891.

[2] Bayer, Patrick, Robert McMillan and Kim Rueben (2004). �What Drives Racial Segregation?

New Evidence Using Census Microdata.�Forthcoming, Journal of Urban Economics.

[3] Bayer, Patrick, Robert McMillan, and Kim Rueben (2004). �Residential Segregation in General

Equilibrium,�mimeo, Yale University.

[4] Berry, Steven and Joel Waldfogel (2003). �Product Quality and Market Size.�NBER Working

Paper 9675.

[5] Benabou, Roland (1992). �Working of a City: Location, Education and Production,�Quarterly

Journal of Economics, Vol. 108: 619-652.

[6] Charles, Zubrinsky C. (2000). �Neighborhood Racial-Composition Preferences: Evidence from

a Multi-ethnic Metropolis.�Social Problems, Vol. 47, 379-407.

[7] Charles, Zubrinsky C. (2001). �Processes of Racial Residential Segregation,�in Urban Inequal-

ity: Evidence from Four Cities, edited by Alice O�Connor, Chris Tilly and Lawrence D. Bobo.

New York: Russell Sage Foundation.

[8] Cornell, S. and Hartmann D. (1997). Ethnicity and Race: Making Identities in a Changing

World. Thousand Oaks: Pine Forge Press.

[9] Cutler, David and Edward Glaeser (1997). �Are Ghettos Good or Bad?�Quarterly Journal of

Economics, 112, 826-72.

[10] Cutler, David, Edward Glaeser and Jacob Vigdor (1999). �The Rise and Decline of the Amer-

ican Ghetto.�Journal of Political Economy, 107(3): 455-506.

20

[11] Duncan, O. and B. Duncan (1955). �A Methodological Analysis of Segregation Indices.�Amer-

ican Sociological Review, 20:210-217.

[12] Durlauf, Steven N. (1996). �A Theory of Persistent Income Inequality." Journal of Economic

Growth, Vol. 1, 75-93.

[13] Farley, R., C. Steeh, M. Krysan, T. Jackson and K. Reeves (1994). �Stereotypes and Segrega-

tion: Neighborhoods in the Detroit Area.�American Journal of Sociology, 100: 750-780.

[14] Galster, G. (1982). �Black and White Preferences for Neighborhood Racial Composition.�

AREUEA Journal, 10, 39-66.

[15] Glaeser, Edward L. and Jacob L. Vigdor (2001). �Racial Segregation in the 2000 Census:

Promising News.�Brookings Center on Urban & Metropolitan Policy.

[16] Ihlanfeldt, K. and D. Sjoquist (1990). Job accessibility and racial di¤erences in youth employ-

ment rates, American Economic Review, 80, 267-276.

[17] Ihlanfeldt, Keith and Ben Sca�di (2002). �Black Self-Segregation as a Cause of Housing Seg-

regation: Evidence from the Multi-City Study of Urban Inequality,� Journal of Urban Eco-

nomics, 51, 366-390.

[18] Kain, John (1968). �Housing Segregation, Negro Employment, and Metropolitan Decentral-

ization.�Quarterly Journal of Economics, 82, 175-197.

[19] King, A. Thomas and Peter Mieszkowski (1973). �Racial Discrimination, Segregation, and the

Price of Housing,�Journal of Political Economy, 81, 590-606.

[20] Lundberg, Shelly and Richard Startz (1998). �Race, Information and Segregation.�mimeo,

University of Washington at Seattle.

[21] O�Flaherty, B. (1999). �Troubled Transactions and their Consequences: Race in the United

States.�mimeo, Columbia University.

[22] Massey, Douglas S. and Nancy A. Denton (1993). American Apartheid: Segregation and the

Making of the Underclass. Cambridge, Massachusetts: Harvard University Press.

[23] Mayer, Susan E. (2004). �Income Inequality, Economic Segregation and Children�s Educational

Attainment.�Working Paper, Harris Graduate School of Public Policy, University of Chicago.

[24] Miller, V. and John M. Quigley (1990). �Segregation by Racial and demographic Group:

Evidence from the San Francisco Bay Area,�Urban Studies, 27: 3-21.

21

[25] O�Reagan, Kathy and John Quigley (1996). �Teenage employment and the spatial isolation of

minority and poverty households,�Journal of Human Resources, Vol. 31, 692-702.

[26] Ross, Stephen L. (1998). �Racial Di¤erences in Residential and Job Mobility: Evidence con-

cerning the Spatial Mismatch Hypothesis,�Journal of Urban Economics, 43(1): 112-35.

[27] Schelling, Thomas C. (1969). �Models of Segregation.� American Economic Review, 59(2):

488-93.

[28] Schelling, Thomas C. (1971). �Dynamic Models of Segregation,� Journal of Mathematical

Sociology, 1: 143-186.

[29] Sethi, Rajiv and Rohini Somanathan (2001). �Racial Income Disparities and the Measurement

of Segregation.�mimeo.

[30] Sethi, Rajiv and Rohini Somanathan (2004). �Inequality and Segregation,�forthcoming, Jour-

nal of Political Economy.

[31] Tiebout, Charles M. (1956). �A Pure Theory of Local Expenditures,� Journal of Political

Economy, 64: 416-424.

[32] Vigdor, Jacob L. (2003). �Residential Segregation and Preference Misalignment,�Journal of

Urban Economics, Vol. 54 No. 3, 587-609.

[33] Weinberg, Bruce (2000). �Black Residential Centralization and the Spatial Mismatch Hypoth-

esis,�Journal of Urban Economics, 48(1): 110-34.

[34] Wilson, William Julius (1987). The Truly Disadvantaged: The Inner City, the Underclass, and

Public Policy. Chicago: University of Chicago Press.

[35] Yinger, John (1978). �The Black-White Price Di¤erential in Housing: Some Further Evidence,�

Land Economics, 54, 187-206.

A Appendix

In this appendix, we provide details about how the marginal types of black households fx�h; x�l gare determined. We �rst restriction to equilibrium in which if a highly-educated (less-educated,

respectively) black household is to choose not to live in community B, she will choose commu-

nity HW or HW�(community LW or LW�respectively) depending on proximity. Given a pair of

thresholds fxl; xhg ; the total measure of less- and highly-educated black households in community

22

B are respectively 2N�b(1 � �b)xl and 2N�b�bxh. Thus the total population in community B is2N�b[�bxh + (1� �b)xl]. Moreover, the relevant proportions for community B are

pb(B) = 1; pw(B) = 0; ph(B) =�bxh

�bxh + (1� �b)xl; pl(B) =

(1� �b)xl�bxh + (1� �b)xl

:

Thus, the utilities for a highly- and less-educated black household with job location z 2 [0; 1] fromliving in community B are respectively:

V hB (z;xh; xl) = �+ �

��bxh

�bxh + (1� �b)xl+ 2

(1� �b)xl�bxh + (1� �b)xl

�� z � c (2N�b [�bxh + (1� �b)xl]) ;

V lB(z;xh; xl) = �+ �

�(1� �b)xl

�bxh + (1� �b)xl+ 2

�bxh�bxh + (1� �b)xl

�� z � c (2N�b [�bxh + (1� �b)xl]) :

We can also calculate the utilities from living in communities HW for a highly-educated black with

job location z 2 [0; 1] : Given the postulated threshold xh; the measure of highly-educated blacksin community HW is N�b�b(1�xh): Taking into account of the measure of highly-educated whitesin community HW, we have the following proportions:

pb(HW ) =�b�b(1� xh)

�b�b(1� xh) + �w�w; pw(HW ) =

�w�w�b�b(1� xh) + �w�w

; ph(HW ) = 1; pl(HW ) = 0:

Thus the utility for a highly-educated black from living in community HW is:

V hHW (z;xh) = �

��b�b(1� xh)

�b�b(1� xh) + �w�w+ 1

�w�w�b�b(1� xh) + �w�w

�+��(1�z)�c (N [�b�b(1� xh) + �w�w]) :

Finally, the utility for a less-educated black with job location z 2 [0; 1] from living in community

LW is:

V lLW (z;xl) = �

��b(1� �b)(1� xl)

�b(1� �b)(1� xl) + �w(1� �w)+ 1

�w�w�b(1� �b)(1� xl) + �w(1� �w)

�+� � �(1� z)� c (N [�b(1� �b)(1� xl) + �w(1� �w)]) :

The equilibrium pair of thresholds (x�l ; x�h)must satisfy:

V hB (x�h;x

�h; x

�l ) = V hHW (x

�h;x

�h); (2)

V lB(x�l ;x

�h; x

�l ) = V lLW (x

�l ;x

�l ): (3)

Equation (2) requires that the marginal type for highly-educated blacks xh� is indi¤erent between

living in community B (an all-black mixed-education community) and community HW (a highly-

educated community with a white majority). Equation (3) requires that the marginal type for

less-educated blacks xl� is indi¤erent between living in community B and community LW (a less-

educated community with white majority). We assume that the parameters of the model are such

that equation system (2) and (3) have solutions.

23

Boston

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

Dallas

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

New York

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

St. Louis

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

Figure 1: Neighborhood Choice Sets in Boston, Dallas, New York and St. Louis.

Atlanta

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

Chicago

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

Detroit

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

Washington-Baltimore

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Percent Black

Perc

ent H

ighl

y-Ed

ucat

ed

Figure 2: Neighborhood Choice Sets in Atlanta, Chicago, Detroit and Washington DC-Baltimore.

BLWLW’

HW’ HW

�∗�

�∗�

−�∗�

−�∗�

-¾-¾

--

¾¾

Figure 3: A Graphical Illustration of an Equilibrium.

BLWLW’

HW’ HW

�∗�

�∗�

−�∗�

−�∗�

LWLW’HW’ HW

�∗�

�∗�

−�∗�

−�∗�

BLWLW’

HW’ HW

�∗�

�∗�

−�∗�

−�∗�

(a) small ��

(b) moderate ��

(c) large ��

BLBH

Figure 4: Comparative Statics with Respect to ��.

Table 1: Make-Up of Population Living in US Metropolitan Areas(1) (2)

Percentage of Percentage Race Education Overall Population by RaceBlack Less than HS 0.029 0.258Non-Hispanic HS 0.032 0.291

Some College 0.033 0.297College Degree 0.011 0.102Advanced Degree 0.006 0.052

White Less than HS 0.091 0.132Non-Hispanic HS 0.185 0.266

Some College 0.192 0.277College Degree 0.124 0.178Advanced Degree 0.102 0.147

Note: The first column reports percentages with respect to all households residing in USmetropolitan areas. The second column reports percentages relative to all such households inthe corresponding race category. Race and educational attainment of head of household is usedin the analysis.

Table 2: Number of Tracts in United States in 2000 by Race and Education

0% 20% 40% 60%

Number 49,021 26,351 11,094 3,005Fraction of tracts at least 0% black 100.0% 53.8% 22.6% 6.1%

at least 20% BlackNumber 9,149 2,567 641 59Fraction of tracts at least 20% black 100.0% 28.1% 7.0% 0.6%

at least 40% BlackNumber 5,657 1,164 142 14Fraction of tracts at least 40% black 100.0% 20.6% 2.5% 0.2%

at least 60% BlackNumber 3,921 623 44 5Fraction of tracts at least 60% black 100.0% 15.9% 1.1% 0.1%

at least 80% BlackNumber 2,559 271 21 1Fraction of tracts at least 80% black 100.0% 10.6% 0.8% 0.0%

at least 20% WhiteNumber 43,179 25,178 11,041 2,999Fraction of tracts at least 20% black 100.0% 58.3% 25.6% 6.9%




Percent College Degree or Moreat least

Note: Tracts considered have a minimum of 800 households (the average tract in the US has almost 3,000 households).Analysis is based on households and race and educational attainment of head of household is used.

Panel A: All Tracts

Panel B: Tracts by Percent Black

Panel C: Tracts by Percent White

Table 3: Locations of Tracts with High Fractions of Both Black and College-Educated Households

Percentage black >80% >60% >40% Metro Size Percent % of Black HhldsPercentage with college degree >40% >40% >40% (millions) Black College-Educated

Baltimore-Washington 5 14 33 5.06 0.24 0.21Detroit 5 8 19 3.51 0.19 0.13Chicago 3 16 6.11 0.16 0.15New York 4 15 14.88 0.15 0.17Los Angeles 4 6 10 11.50 0.06 0.18Atlanta 5 5 8 2.65 0.26 0.22Cleveland 1 6 1.96 0.15 0.11Philadelphia 1 5 4.12 0.17 0.13San Francisco-Oakland 5 4.95 0.06 0.19Raleigh-Durham 1 3 0.65 0.12 0.22Indianapolis 3 1.05 0.12 0.14Jackson, MS 1 1 2 0.44 0.25 0.17Houston 1 1 2 3.10 0.15 0.18Columbia, SC 2 0.59 0.17 0.17New Orleans 2 0.85 0.33 0.13

Total 21 44 142 0.11 0.15

Notes: Tracts considered have a minimum of 800 households (the average tract in the US has almost 3,000 households). Analysis is based on households and race and educational attainment of head of household is used.

Average for US Metro Areas

Table 4: Neighborhood Patterns for College-Educated Households in the United States

Panel A: Neighborhood Patterns for College-Educated Black Households Households are first ranked by percent black in Census tract within its MSAMeasures reported by household's corresponding quintile within its MSA

Quintile 1 2 3 4 5 TotalPercent Black 5.7 14.4 28.3 54.6 78.9 32.0Percent Highly Educated 38.0 31.6 26.2 18.4 13.8 27.2Percent Black Low-Ed 4.4 11.1 22.1 46.6 69.0 26.8Percent Black High-Ed 1.3 3.3 6.2 8.0 10.0 5.2Percent White Low-Ed 48.3 44.6 33.7 18.2 7.4 32.9Percent White High-Ed 32.9 24.1 16.1 8.2 2.7 18.8

Panel B: Neighborhood Patterns for College-Educated White Households Households are first ranked by percent white in Census tract within its MSAMeasures reported by household's corresponding quintile within its MSA

Quintile 1 2 3 4 5 TotalPercent White 55.0 77.9 86.6 90.4 94.5 77.4Percent Highly Educated 27.0 36.2 40.7 39.3 39.2 35.3Percent Black Low-Ed 21.1 5.6 2.6 1.8 1.2 8.4Percent Black High-Ed 3.3 1.6 0.9 0.6 0.3 1.6Percent White Low-Ed 34.9 47.5 50.4 54.3 57.1 46.9Percent White High-Ed 20.1 30.4 36.2 36.1 37.4 30.4

Note: The panels of the table summarize the average distribution of neighborhoods in which college-educatedblacks and whites in US metro areas reside, respectively. To construct the numbers in the upper panel, college-educated blacks in each metro area are ranked by the fraction of black households in their tract and assigned to oneof five quintiles . Average neighborhood sociodemographic characteristics are then reported for each quintile,averaging across all metro areas. The lower panel reports analogous figures for college-educated whites, firstranking by their tract-level exposure to whites within each MSA.

Table 5: Average Tract-Level Exposure by Race and Education

Household Black-Low Ed Black - High Ed White - Low Ed White High-Ed High Ed BlackBlack Low-Ed 0.332 0.046 0.315 0.150 0.220 0.378Black High-Ed 0.268 0.052 0.329 0.188 0.272 0.320White Low-Ed 0.075 0.013 0.564 0.222 0.259 0.088White High-Ed 0.084 0.016 0.469 0.304 0.353 0.100

Household Black-Low Ed Black - High Ed White - Low Ed White High-Ed High Ed BlackBlack Low-Ed 0.196 0.021 -0.136 -0.077 -0.063 0.217Black High-Ed 0.133 0.026 -0.103 -0.044 -0.021 0.159White Low-Ed -0.014 -0.003 0.048 -0.003 -0.009 -0.017White High-Ed -0.010 -0.001 0.002 0.043 0.044 -0.011

Panel A: Average Tract-Level Exposure

Panel B: Average Tract-Level Exposure Relative to MSA Average

Note: Table reports average tract characteristics for households in the race-education category shown in row heading. The lower panel reports average tractcharacteristics relative to the average characteristics of the household's metropolitan area.

Table 6: Average Tract-Level Exposure by Race and Education

Panel A: Metropolitan Areas Below Median Fraction of Colleg-Educated Blacks (<1.23 percent)

Individual Black-Low Ed Black - High Ed White - Low Ed White High-Ed High Ed BlackBlack Low-Ed 0.111 0.013 -0.107 -0.049 -0.038 0.124Black High-Ed 0.078 0.016 -0.078 -0.014 0.005 0.094White Low-Ed -0.006 -0.001 0.030 -0.003 -0.004 -0.007White High-Ed -0.007 0.000 0.002 0.042 0.045 -0.007

Panel B: Metropolitan Areas Above Median Fraction of College-Educated Blacks (>1.23 percent)

Individual Black-Low Ed Black - High Ed White - Low Ed White High-Ed High Ed BlackBlack Low-Ed 0.220 0.024 -0.144 -0.085 -0.070 0.244Black High-Ed 0.147 0.028 -0.110 -0.052 -0.027 0.175White Low-Ed -0.024 -0.005 0.067 -0.004 -0.014 -0.029White High-Ed -0.012 -0.002 0.002 0.044 0.043 -0.014

Average Tract-Level Exposure Relative to MSA Average

Average Tract-Level Exposure Relative to MSA Average

Note: This table reports average tract characteristics relative to the average characteristics of the household's metropolitan area. The upper panel reportsaverages for households residing in metro areas in which college-educated black households constitute less than 1.23 percent of the metropolitan area. The lowerpanel reports averages for metro areas in whcih black college-educated households constitute more than 1.23 percent of the population.

Table 7: Segregation and Metropolitan Area Sociodemographics - Average Tract-Level Exposure

Dependent Variable:

% Black % Black % White % White % High Ed % Black& High Ed & Low Ed & High Ed & Low Ed

(1) (2) (3) (4) (5) (6)

I_BlackHighEd* 0.968 3.040 -2.110 -2.492 -1.261 4.008M_BlackHighEd (0.201) (1.575) (0.842) (0.902) (0.861) (1.720)

I_BlackHighEd* -0.065 0.069 0.200 0.111 0.128 0.004M_BlackLowEd (0.062) (0.360) (0.135) (0.264) (0.124) (0.413)

I_BlackHighEd* -0.058 -0.257 0.114 0.239 0.070 -0.315M_WhiteHighEd (0.018) (0.091) (0.043) (0.063) (0.042) (0.104)

I_BlackHighEd* 0.000 0.148 0.055 -0.166 0.066 0.149M_WhiteLowEd (0.024) (0.091) (0.036) (0.088) (0.033) (0.113)

I_BlackLowEd* 1.022 4.911 -3.097 -2.663 -2.340 5.933M_BlackHighEd (0.119) (1.983) (1.150) (0.898) (1.295) (2.030)

I_BlackLowEd* -0.086 -0.062 0.365 0.171 0.319 -0.148M_BlackLowEd (0.040) (0.399) (0.179) (0.255) (0.185) (0.427)

I_BlackLowEd* -0.039 -0.325 0.186 0.301 0.191 -0.364M_WhiteHighEd (0.014) (0.094) (0.053) (0.077) (0.045) (0.105)

I_BlackLowEd* 0.014 0.218 0.026 -0.149 0.079 0.232M_WhiteLowEd (0.017) (0.124) (0.055) (0.105) (0.044) (0.140)

I_BlackHighEd 0.023 0.030 -0.019 0.030 -0.007 0.053(0.015) (0.060) (0.022) (0.046) (0.024) (0.073)

I_BlackLowEd 0.011 0.049 -0.057 -0.025 -0.086 0.059(0.011) (0.078) (0.036) (0.061) (0.029) (0.088)

I_WhiteHighEd -0.004 -0.038 0.100 0.084 0.094 -0.041(0.002) (0.011) (0.013) (0.014) (0.014) (0.011)

I_WhiteLowEd -0.005 -0.037 0.048 0.130 0.035 -0.041(0.002) (0.010) (0.011) (0.015) (0.012) (0.012)

Tract-Level Exposure

Note: All regressions include metropolitan area fixed effects. The I_ prefix indicates an individual's own race-educated category. The M_ prefix indicates the fraction of individuals in the corrresponding race-education categorythat reside in the individual's metro area. Standard errors adjusted for clustering at the metropolitan area level arereported in parentheses; p-values are reported for tests.

Highly-Educated Black 0.010 0.030 -0.023 -0.026 -0.014 0.040

Less-Educated Black 0.011 0.050 -0.035 -0.028 -0.027 0.061



Note: This table summarizes the key coefficient differences for which test statistics were reported at the bottom of Table 7. The upper panel reports theresults of increasing the fraction of highly-educated blacks in the metro area and decreasing the fraction of less-educated blacks by 1 percent,respectively. The lower panel reports the results of increasing the fraction of highly-educated blacks in the metro area and decreasing the fraction ofhighly-educated whites by 1 percent, respectively. In both cases, the corresponding change in the relative exposure of both highly- and less-educatedblacks to each of the racial-education groups shown in the column head is shown. The highly-educated category is defined as having a college-degree ormore. P-values are reported.

Panel A: Increasing Fraction of Highly-Educated Blacks by 1 Percent at Expense of Less-Educated Blacks



Panel B: Increasing Fraction of Highly-Educated Blacks by 1 Percent at Expense of Highly-Educated Whites



Note: This table is analogous to the specification reported in Table 8 with the exception that the underlying regressions (analogous to those reported inTable 7) re-define the highly-educated category is defined as having at least attended college for some time rather than having received a four-yeardegree. P-values are reported in italics.







Note: This table is analogous to the specification reported in Table 9 (highly-educated category is defined as having at least attended college for sometime) with the exception that the underlying regressions also include controls for the interaction of the four individual race-education variables with thetotal population of the metropolitan area as well as four region dummies corresponding to the Census-defined regions (Northeast, Midwest, South, andWest). P-values are reported in italics.

Table 11: Segregation and Metro Area Sociodemographics - By Metropolitan Size

% Black % Black % Black % Black % Black % Black % Black % Black % Black& High Ed & Low Ed & High Ed & Low Ed & High Ed & Low Ed


Highly-Educated Black 0.001 0.001 0.002 0.012 0.013 0.025 0.027 0.012 0.0400.644 0.892 0.810 0.015 0.063 0.022 0.000 0.431 0.067

Less-Educated Black 0.003 0.011 0.014 0.012 0.016 0.028 0.024 0.025 0.0490.311 0.104 0.119 0.001 0.046 0.012 0.001 0.204 0.055




Note: This table is analogous to the specification reported in Table 10 (highly-educated category is defined as having at least attended college for some time; controls for theinteraction of the four individual race-education variables with the total population of the metropolitan area as well as four region dummies) run separately for three categoriesof Metro areas by population size. 147 metro areas have less than 200k individuals; 75 metro areas have between 200-600k individuals; 54 metro areas have more than 600kindividuals; P-values are reported in italics.

Change in Relative Tract-Level ExposureMSA Population (0-200K) MSA Population (200-600K) MSA Population (600K+)

% Black % Black % White % White % High Ed % Black& High Ed & Low Ed & High Ed & Low Ed







Note: This table is analogous to the specification reported in Table 9 (highly-educated category is defined as having at least attended college for sometime) with the exception that the dependent variables in the underlying regressions are measured at the Census PUMA rather than Census tract level. P-values are reported in italics.

Table 13: PUMA-Level Analysis with Lagged and Differenced Metropolitan Area Sociodemographics

Dependent Variable:

Actual Lagged Differenced Actual Lagged Differenced Actual Lagged Differenced







Note: This first column of each panel repeats the corresponding results from Table 12 based on Census PUMA level relative exposure rates. The second and third columns reportthe corresponding coefficients when 2000 MSA sociodemographics are decomposed into a lagged measure based on where the householded lived in 1995 and the differencebetween the 2000 and 1995 measure. For these results, a complete set of interaction terms were included in the underlying regressions for both the lagged and differenced terms. P-values are reported in italics.

Percent Less-Educated Black

Single Regression

PUMA-Level ExposurePUMA-Level Exposure PUMA-Level ExposurePercent Black

Single Regression

Percent Highly-Educated Black

Single Regression

Documents

Separate When Equal? Racial Inequality and Residential ...public.econ.duke.edu/~hf14/WorkingPaper/segregation/seg1_old.pdf · Residential Segregation Patrick Bayer Hanming Fang Robert