This article was downloaded by: [Syracuse University Library]

On: 30 May 2013, At: 07:41

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Housing Policy Debate

Publication details, including instructions for authors and
subscription information:
[Link]

Wrestling Sprawl to the Ground:

Defining and measuring an elusive
concept
a b c
George Galster , Royce Hanson , Michael R. Ratcliffe ,
d e e
Harold Wolman , Stephen Coleman & Jason Freihage
a
Hilberry Professor of Urban Affairs, Wayne State University
b
Visiting Professor in the Policy Sciences Graduate Program,
University of Maryland, Baltimore County
c
Geographer in the Population Distribution Branch, Population
Division, U.S. Bureau of the Census
d
Director of the George Washington Institute of Public Policy
and Professor in the Department of Political Science, George
Washington University
e
Graduate Research Assistant in the Policy Sciences Graduate
Program, University of Maryland, Baltimore County
Published online: 31 Mar 2010.

To cite this article: George Galster , Royce Hanson , Michael R. Ratcliffe , Harold Wolman ,
Stephen Coleman & Jason Freihage (2001): Wrestling Sprawl to the Ground: Defining and
measuring an elusive concept, Housing Policy Debate, 12:4, 681-717

To link to this article: [Link]

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: [Link]

conditions

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any
representation that the contents will be complete or accurate or up to date. The
accuracy of any instructions, formulae, and drug doses should be independently
verified with primary sources. The publisher shall not be liable for any loss, actions,
 claims, proceedings, demand, or costs or damages whatsoever or howsoever caused
arising directly or indirectly in connection with or arising out of the use of this material.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013
 Housing Policy Debate ·
Volume 12, Issue 4
© Fannie Mae Foundation 2001. All Rights Reserved.
681
681

Wrestling Sprawl to the Ground:

Defining and Measuring an Elusive Concept

George Galster
Wayne State University

Royce Hanson
University of Maryland, Baltimore County

Michael R. Ratcliffe
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

U.S. Bureau of the Census

Harold Wolman
George Washington University

Stephen Coleman and Jason Freihage

University of Maryland, Baltimore County

Abstract
The literature on urban sprawl confuses causes, consequences, and conditions. This
article presents a conceptual definition of sprawl based on eight distinct dimensions of
land use patterns: density, continuity, concentration, clustering, centrality, nuclearity,
mixed uses, and proximity. Sprawl is defined as a condition of land use that is repre-
sented by low values on one or more of these dimensions.

Each dimension is operationally defined and tested in 13 urbanized areas. Results for
six dimensions are reported for each area, and an initial comparison of the extent of
sprawl in the 13 areas is provided. The test confirms the utility of the approach and
suggests that a clearer conceptual and operational definition can facilitate research on
the causes and consequences of sprawl.

Keywords: Land use/zoning; Urban environment

A metaphor rich in ambiguity

Urban sprawl is one name for many conditions. It has been attached to
patterns of residential and nonresidential land use, to the process of ex-
tending the reach of urbanized areas (UAs), to the causes of particular
practices of land use, and to the consequences of those practices. Sprawl
has been denounced on aesthetic, efficiency, equity, and environmental
grounds and defended on choice, equality, and economic grounds. Sprawl
has become the metaphor of choice for the shortcomings of the suburbs
and the frustrations of central cities. It explains everything and nothing.
 682 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Much of the confusion about sprawl stems from the conflation of ideolo-
gy, experience, and effects. A term so widely used cannot be easily dis-
missed as too vague for serious discussion. Many policy makers claim to
know it when they see it and make important policy judgments based
on what they see or think they see. As a first step toward developing
policies to deal with the causes or consequences of sprawl, it would help
both critics and apologists if agreement could be reached on what sprawl
is and how to measure it empirically and compare its occurrence across
a large number of urban areas.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Sprawling literature: Lost in a semantic wilderness

Consistent with the findings of others (Burchell et al. 1998), our survey
of the literature yielded no common definition of sprawl and relatively
few attempts to operationally define it in a manner that would lead to
useful comparisons of areas to determine which had experienced greater
or less degrees of sprawl. There are two notable recent exceptions: First,
Torrens and Alberti (2000) offer sophisticated indices for measuring
multiple aspects of sprawl, such as density, scatter, leapfrogging, inter-
spersion, and accessibility, but provide no empirical prototypes. Second,
Malpezzi (1999) has undertaken an ambitious effort to develop some
precise definitions of several dimensions of sprawl such as density and
lack of continuity. Moreover, he has quantified them for a sample of
metropolitan areas and related them statistically to determinants of
sprawl such as the length of average daily commutes.

Our analysis of the social science and planning literature suggests that
definitions of sprawl can be grouped into six general categories:

1. Sprawl is defined by an example, which is seen to embody the char-

acteristics of sprawl, such as Los Angeles.

2. Sprawl is used as an aesthetic judgment about a general urban de-

velopment pattern.

3. Sprawl is a cause of an externality, such as high dependence on the

automobile, isolation of the poor in the inner city, the spatial mis-
match between jobs and housing, or loss of environmental qualities.

4. Sprawl is the consequence or effect of some independent variable,

such as fragmented local government, poor planning, or exclusionary
zoning.

5. Sprawl is defined as one or more existing patterns of development.

Those most frequently mentioned are low density, leapfrogging, dis-
tance to central facilities, dispersion of employment and residential
development, and continuous strip development.
 Wrestling Sprawl to the Ground 683

6. Sprawl is defined as a process of development that occurs over some

period of time as an urban area expands.

Definition by example

In both the popular and scholarly literature, sprawl is frequently de-

fined by one or more examples of scattered or low-density patterns of
urban development. Los Angeles is often given a place of honor in ex-
emplary definitions. Robert Geddes (1997) asserted, “Its key words are
fragmented, incomplete, ad hoc, uncentered” (3). Unfortunately, while it
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

is big and spread over a large area, Los Angeles is more densely settled
than most large areas in the United States. Atlanta has come to replace
it as an example of sprawl, but with the possible exception of Portland,
OR, any area is a potential nominee. The flexibility of definition by ex-
ample makes it possible to include all sorts of development patterns,
from planned communities with clustered housing and mixed uses to
exurban rural estates. A Wells Fargo Bank report, Preserving the Amer-
ican Dream, concluded that sprawl “receives blame for seemingly every
bad aspect of contemporary urban life” (quoted in Myers and Kitsuse
1999, 6).

Aesthetic definition

Ad hoc examples at least imply and often express an aesthetic judgment:

Sprawl is ugly development. In The Language of Cities (1971) Charles
Abrams defined sprawl as follows: “The awkward spreading out of limbs
of either a man or a community. The first is a product of bad manners,
the second of bad planning” (293–94). Even so careful a land economist
as Marion Clawson (1962) could not resist judgmental adjectives and
adverbs in his definition: “[the] rapid spread of suburbs across the pre-
viously rural landscape, tendency to discontinuity, large closely settled
areas intermingled haphazardly with unused areas” (94).

The cause of an unwanted externality

Traffic congestion (Black 1996; Downs 1999; Vermont Forum on Sprawl

1999), environmental contamination (Sierra Club 1998), income and ra-
cial segregation of neighborhoods (Downs 1998), the mismatch between
jobs and housing (Orfield 1997), local fiscal disparities (Burchell et al.
1998), conversion of farmland to urban uses (U.S. General Accounting
Office [GAO] 1999), and civic alienation (Popenoe 1979), among other
maladies of contemporary urban life, have been attributed to sprawl.
Here the definitions segue from judgments about the appearance of
sprawl to assertions of causal links between sprawling land use patterns
 684 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

and environmental, social, and economic costs. But these definitions

basically describe what sprawl does (or is supposed to do) rather than
what it is.

A consequence

Sprawl is also frequently defined as the consequence of something else.

Downs (1998), among others (Black 1996; Burchell et al. 1998; Mosko-
witz and Lindbloom 1993; Orfield 1997), argues that sprawl occurs as
a consequence of the fragmentation of control over land use in metro-
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

politan areas. It is unclear whether sprawl is an intentional, necessary,

or inadvertent consequence of fragmented governance of growth. In
light of development patterns in areas with unified governments, such
as Houston and Lexington, KY, it seems a doubtful proposition. In any
event, understanding the policies that induce specific development pat-
terns could lead to their correction if there were a clearer specification
of what those patterns are.

Selected patterns of land development

Altshuler and Gomez-Ibanez (1993) edge toward a clearer definition

of sprawl by identifying the types of development patterns associated
with it:

Continuous low density residential development on the metropolitan

fringe, ribbon low density development along major suburban high-
ways, and development that leapfrogs past undeveloped land to leave
a patchwork of developed and undeveloped tracts. (67)

The inconsistency of continuous and leapfrog development aside, this

definition at least characterizes land use conditions, and it is conceiv-
able that continuous development, ribbon development along corridors,
and leapfrog development are different forms of sprawl (Harvey and
Clark 1965). Other development patterns frequently characterized as
sprawl include low density (Lockwood 1999), random (GAO 1999), large-
lot single-family residential (Popenoe 1979), radial discontinuity (Mills
1980), single land use or physical separation of land uses (Burchell et
al. 1998; Cervero 1991), widespread commercial development (Downs
1998), strip commercial (Black 1996; Burchell et al. 1998), and non-
compact (Gordon and Richardson 1997).

A process of development

Finally, some commentators (Ewing 1997; Harvey and Clark 1965) sug-
gest that sprawl represents a stage in the development process rather
 Wrestling Sprawl to the Ground 685

than a static condition. This definition suggests that some parts of an

urban area may pass through a sprawl stage before eventually thicken-
ing and diversifying so they can no longer be characterized as sprawl, at
least by those authors. When used to signify a stage or process of devel-
opment, sprawl is a verb, rather than a noun connoting a condition. But
little is gained by changing the part of speech, for there is little in the
literature to indicate when sprawl metamorphoses into nonsprawl. It
does, however, suggest that sprawl may represent some range on a
development pattern continuum.

These descriptions of sprawl leave one grasping for something more

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

solid. How far does the development frog have to leap and how light and
broad must its footprint be for sprawl to be present? When do land uses
thin sufficiently from being compact, centered, or concentrated before
they degrade into sprawl? An empirical definition is needed if the dis-
cussion is to move from polemics to a common understanding and use-
ful analysis of urban form. Once that is achieved, it should be possible
to conduct a better-informed discussion of the forces and factors that
cause certain patterns of development and to address the consequences
that flow from certain urban forms for different population groups, such
as a region’s poor.

Conceptually, a thing cannot simultaneously be what it is and what

causes it or what it causes. If sprawl is to be a useful concept for describ-
ing something important that occurs in urban areas, it must first be re-
duced to some objective conditions or traits. Some of the characteriza-
tions drawn from the literature are too impressionistic for empirical
measurement. Others identify conditions, dimensions, or attributes of
sprawl that can be operationally defined, among them discontinuous,
widespread, or random development; low-density residential or non-
residential development; continuous low-density or strip development;
spatially segregated land uses; and dispersed employment centers.

A conceptual definition

As a noun, sprawl implies a condition characterizing an urban area, or

part of it, at a particular time. If sprawl is to be measured as a condition
of land use, it must be distinguished from other conditions that may
well be related to it. On the basis of the descriptions of conditions char-
acterizing it in the literature and amplified by observation and experi-
ence, the following conceptual definition is suggested:

Sprawl (n.) is a pattern of land use in a UA that exhibits low levels

of some combination of eight distinct dimensions: density, continu-
ity, concentration, clustering, centrality, nuclearity, mixed uses, and
proximity.
 686 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

This definition suggests the possibility that there can be different types
of sprawl, consisting of different combinations of these dimensions.1 It
also suggests the possibility of defining sprawl as a process of develop-
ment by looking at changes in patterns of land use over time, particu-
larly at the periphery.

Emphasizing the pattern of land use distinguishes the condition from

its causes or effects. The pattern of land use refers to the arrangement
of the built environment for residential and employment activities.
Other uses, such as passive or active open space, agriculture, and pub-
lic facilities and networks, will affect that pattern along some of its
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

dimensions.

The UA is a more appropriate unit of analysis than the metropolitan

area (MA), which is composed of contiguous counties, some of which may
contain large outlying rural areas with population densities far below
the minimum UA threshold of 1,000 people per square mile. Including
such rural areas can result in exaggeration of some dimensions of
sprawl, such as density. Using the census-defined UA alone, however,
may exclude semirural development at the urban fringe that some con-
sider the epitome of sprawl. Whatever its limitations, the UA is a well-
established construct that captures settlements averaging as few as
2.4 units per acre and the vast majority of all development of the MA.
It may ultimately be useful to add to the UA subdivisions that are inte-
grally associated with it, as distinct from established rural homesites
and outlying communities located in metropolitan counties. Using the
UA alone, however, should provide a reasonable test of the ability of the
eight dimensions to characterize the extent of sprawl within it. And if
they can, they can also be applied to a carefully delineated larger area.

Despite its disagreements and contradictions, the literature agrees that

all development is not sprawl; simply because a UA is larger does not
mean that it is more sprawled. Moreover, all sprawl does not have the
same characteristics or dimensions. Excluding terms that cannot be
easily quantified, such as ugly or excessive, there appear to be several
objective dimensions of land use, that, if present at low levels in a UA,
can fairly be called sprawl. Therefore, 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. It would seem fair to characterize
UAs with development patterns that score low on all dimensions as ex-
periencing more sprawl than others. Further, the dimensions may be
present in different degrees and combinations across many urban areas,
making it possible to distinguish different types of sprawl.

1
In our conceptualization that sprawl is a multidimensional phenomenon, we are con-
sistent with the work of Malpezzi (1999) and Torrens and Alberti (2000).
 Wrestling Sprawl to the Ground 687

Research strategy

This article reports the first two steps in a long-range research strategy.
The first step, reported in the next section, identifies and develops con-
ceptual and operational definitions of eight dimensions of sprawl (used
as a noun). Each dimension consists of a continuum with low values
representing more sprawl-like features. The study has not attempted
to develop measures for sprawl as a process, although that is a worthy
subject for further inquiry. In any event, measuring the respective di-
mensions of development patterns for an urban area at different times
will reveal the process (or progress) of sprawl.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

The second step tests the operational definitions by applying them to

13 large UAs from different regions, with different economic structures
and demographic composition, to determine whether they make intu-
itive sense.

Future research contemplates making any necessary adjustments to

the description or measurement of each dimension and applying the ap-
proach to the 100 largest UAs in the United States and ranking them on
each of the eight dimensions, as well as any distinctive factors that
emerge from the analysis. Techniques such as factor analysis can be
used to determine whether distinct patterns or combinations of low
values on these dimensions can be identified as different types of sprawl
variables that will enable serious research on sprawl. The resulting vari-
ables could be used in models either as independent variables (What is
the effect of different dimensions or types of sprawl on X?) or dependent
variables (What causes particular dimensions or types of sprawl?).

Eight dimensions of sprawl

We now offer eight conceptually distinct, objective dimensions of land

use that if present at low values and in some combination, characterize
sprawl.

Density

Density is the average number of residential units per square mile of

developable land in a UA.

Density is the most widely used indicator of sprawl (Black 1996; Bur-
chell and Listokin 1991; Burchell et al. 1998; GAO 1999; Gordon and
Richardson 1997; Lockwood 1999; Malpezzi 1999; Moskowitz and Lind-
bloom 1993; Orfield 1997; Popenoe 1979; Sierra Club 1998; Torrens and
Alberti 2000). It is usually expressed simply, as the ratio of the total pop-
 688 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

ulation of a metropolitan area to its total land area. However, residen-

tial units are a better unit for measuring sprawl as a physical condition
of land use. And developable land—land that has no natural features,
public uses, or regulatory barriers to its development at urban densi-
ties—is a better denominator for calculating density than total land
area. It is also a more useful area for measuring all the other dimen-
sions of land use patterns. Using developable land as a measure makes
it possible to eliminate physical features and other constraints, such
as public open space, that interrupt or preclude development, whether
areas are assessed alone or compared.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Residential density is likely to be a more useful indicator than nonres-

idential development. First, that is the way the term is generally under-
stood and used in the literature. Second, nonresidential uses are more
likely than residential uses to be “lumpy” due to agglomeration econo-
mies and regulations that limit such development to fewer locations.
Thus, their average density is a less reliable indicator of their pattern
of distribution. And unlike the case of housing units, which bear a close
relationship to population, the relationship between the number of busi-
ness establishments and employees varies widely. The average employ-
ment per square mile might be a slightly more significant element of
the density dimension, although it fluctuates with business cycles. Be-
cause we are examining sprawl as a condition of land use, and firms
and employees are both far more likely to be clustered than residences,
it may be more appropriate to examine them in another dimension.

Figure 1 illustrates the density dimension. With the same gross land
area, A has a greater number of residential units and thus higher den-
sity than B.

Continuity

Continuity is the degree to which developable land has been built upon
at urban densities in an unbroken fashion.

The second most cited dimension of development is continuity (Ewing

1997; Harvey and Clark 1965). Continuous development may occur at
any level of density, although the steady outward march of low-density
development in concentric rings from the urban center or core is com-
monly characterized as sprawl (Altshuler and Gomez-Ibanez 1993;
American Society of Civil Engineers 1999; GAO 1999; Harvey and Clark
1965; Lockwood 1999). So is continuous “‘ribbon’ low-density develop-
ment along major suburban highways” (Altshuler and Gomez-Ibanez
1993, 67; see also Ewing 1997 and Harvey and Clark 1965). Some com-
mentators identify “discontinuity” as one of the significant attributes of
sprawl (Clawson 1962), citing development that “leap-frogs past unde-
veloped land to leave a patchwork of developed and undeveloped tracts”
 Wrestling Sprawl to the Ground 689

Figure 1. Density: The Average Number of Residential Units

per Square Mile of Developable Land in a UA

A
High-Density Area
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

B
Low-Density Area

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 units

(Altshuler and Gomez-Ibanez 1993, 67; see also Burchell and Listokin
1991; Burchell et al. 1998; Ewing 1997; Gordon and Richardson 1997;
Harvey and Clark 1965; Mieszkowski and Smith 1991; Mills 1980).

Following these definitions, sprawl can be continuous in some places and

discontinuous in others. Discontinuous development could be character-
ized as sprawl in some cases but as something else in others. Thus, the
development of planned urban centers with moderate to high densities,
separated along a transportation corridor by greenbelts or other open
spaces, might not be characterized by some commentators as sprawl,
 690 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

although development of low-density subdivisions or commercial centers

or industrial/office parks that have passed over developable land might
be so characterized. The literature, however, rarely distinguishes between
the smooth and lumpy spread of development across the landscape.
This is particularly true when density is measured solely by the ratio of
population to land area. More sophisticated approaches have used alter-
native indicators of density gradients to capture both the density and
noncompact aspects of sprawl, such as the density of the census tract
containing the 10th percentile of the MA population (Malpezzi 1999).

As defined here, the continuity dimension is concerned with density only

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

as a means of determining whether a tract of developable land contains

enough housing units or employment to consider it part of a continuous
pattern or skipped over. Thus, the continuity dimension indicates only
the extent of leapfrog development. UAs with discontinuous develop-
ment patterns will have low scores on this dimension. Bodies of water;
protected wetlands, forests, parks, slopes, or soils; and freeways, inter-
changes, or other public reservations and facilities are not considered
interruptions of continuous development patterns.

Figure 2 illustrates the application of the continuity dimension to two

urban areas containing the same amount of development. A has a high
level of continuity, but B has less and exhibits a leapfrog pattern.

Concentration

Concentration is the degree to which development is located dispropor-

tionately in relatively few square miles of the total UA rather than
spread evenly throughout.

An urban area may be continuously developed, but no urban area is ever

evenly developed. The density dimension does not tell us anything about
how residential uses are distributed. The same 100-square-mile area
with 500,000 residences and 200,000 jobs would have an average resi-
dential density of 5,000 units per square mile and an average employ-
ment density of 2,000 jobs per square mile. The homes and jobs could be
distributed in an almost infinite number of arrangements with equal
degrees of continuity.

The concentration dimension distinguishes those urban areas in which

most housing units and employment are located in relatively few places
at relatively high densities from those in which development is more
evenly distributed across the urban landscape, as illustrated by figure 3.
With the same amount of development in each diagram, A is more highly
concentrated than B, where development is more evenly distributed.
With lower concentration, B’s development pattern is more sprawl-like
on this dimension.
 Wrestling Sprawl to the Ground 691

Figure 2. Continuity: The Degree to Which Developable Land Has Been

Built Upon at Urban Densities in an Unbroken Fashion

A
High Continuity
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

B
Low Continuity/
Leapfrog Development

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 units

Clustering

Clustering is the degree to which development has been tightly bunched

to minimize the amount of land in each square mile of developable land
occupied by residential or nonresidential uses.

Sprawl is frequently used as an antonym for development that is stacked

or clustered so that its footprint occupies only a small portion of the
land area associated with it (Gordon and Richardson 1997). We empha-
 692 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Figure 3. Concentration: The Degree to Which Development Is Located in

Relatively Few Square Miles Rather Than Spread Evenly across the UA

A
High Concentration
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

B
Low Concentration

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 units

size that clustering is a distinct dimension of land development. Unlike

density and concentration, which are concerned with development pat-
terns across grids, clustering is concerned with the footprint of develop-
ment patterns within grids. Development may be dense and concentrat-
ed and still not be clustered (because it is uniformly spread within all
grids, high- and low-density ones alike). Conversely, a UA may have low
densities and a low concentration, but clustering may be high if what-
ever urbanized uses there are within a grid are tightly bunched.

Low clustering is a dimension that has been linked to several alleged

effects, such as the creation of impervious surfaces that contribute to
 Wrestling Sprawl to the Ground 693

flooding and erosion. Clustered development may also reduce travel dis-
tances within a development, but not necessarily between home and
work if they are situated in different parts of the area.

In figure 4, the development in A has been clustered so that it occupies

half or less than half of the land area in each of the large squares. In B,
the same amount of development in each large square is more evenly
distributed. Thus B’s development is more sprawl-like overall on this
dimension.

Figure 4. Clustering: The Degree to Which Development Has Been

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Tightly Bunched to Minimize the Amount of Land

in Each Square Mile of Developable Land Occupied by Units
of Residential or Nonresidential Use

A
Clustered Development

B
Unclustered Development

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 units
 694 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Centrality

Centrality is the degree to which residential or nonresidential develop-

ment (or both) is located close to the central business district (CBD) of
an urban area.

Loss of centrality is one of the most common laments about sprawl. Gen-
erally, this refers to the extent to which development has diffused across
the landscape from the historic core or CBD of an urban area. Decen-
tralization of urban areas is often cited as a cause for longer travel dis-
tances and times and inefficiencies in land use. The centrality of a UA
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

increases as the radius from the CBD within which the greater propor-
tion of development is located shortens. Conversely, an area will exhib-
it greater sprawl where greater distances from the center are required
to contain the same proportion of development. Figure 5 depicts a highly
centralized area and another with a low level of centralization and, thus,
more of this dimension of sprawl.

Nuclearity

Nuclearity is the extent to which an urban area is characterized by a

mononuclear (as opposed to a polynuclear) pattern of development.

Centrality is a measure best suited to mononuclear urban areas. In-

creasingly, U.S. urban areas have become polynuclear as historic CBDs
have declined in relative or even absolute terms, outlying centers and
edge cities have grown in scale, and different centers have taken on
more specialized functions, such as financial centers, technology centers,
retail, or manufacturing hubs.

If its CBD is the only locus of intense development, an area will have a
mononuclear structure, and its nuclearity is maximized. If the same ac-
tivities are dispersed over several intensely developed places and each
contains an agglomeration of activities that represent a substantial
proportion of the total of such activities in the region, it is polynuclear.

Nuclearity and concentration need not be closely related. A UA may only

have one nucleus or many nuclei, but if their densities are not signifi-
cantly greater than the average density of the rest of the UA, concentra-
tion will be low. Similar logic leads one to the conclusion that nuclearity
is conceptually distinct from our other dimensions of sprawl as well.

Nuclearity is an important dimension. A polynuclear pattern may re-

duce costs for some people by shortening their journey to work, but it
may increase other costs, such as land values in the vicinity of major
employment nodes.
 Wrestling Sprawl to the Ground 695

Figure 5. Centrality: The Degree to Which Development in a UA

Is Located Close to the CBD

A
Highly Centralized
Area
yyyyyyy
;;;;;;;
;;;;;;;
yyyyyyy
;;;;;;;
yyyyyyy
;;;;;;;
yyyyyyy
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

;;;;;;;
yyyyyyy
;;;;;;;
yyyyyyy

B
Highly Decentralized
Area

yyyyyyyy
;;;;;;;;
= UA border ;;;;;;;;
yyyyyyyy
;;;;;;;;
yyyyyyyy
= MA ;;;;;;;;
yyyyyyyy
yyyyyyyy
;;;;;;;;
= Square mile yyyyyyyy
;;;;;;;;
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 units
yyy
;;;
;;;
yyy
yyy
;;; = CBD
yyy
;;;

In figure 6, A illustrates a mononuclear UA and B represents a polynu-

clear area.

Mixed uses

Mixed uses means the degree to which two different land uses com-
monly exist within the same small area, and this is common across
the UA.
 696 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Figure 6. Nuclearity: The Extent to Which a UA Is Characterized

by a Mononuclear or Polynuclear Pattern of Development

A
Mononuclear Area
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

B
Polynuclear Area

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 units

Another feature attributed to sprawl is the separation of different kinds

of land uses from each other (Vermont Forum on Sprawl 1999) and the
income segregation of suburban residential developments, due primarily
to minimum lot sizes in different zoning categories (Burchell et al. 1998;
Cervero 1991; Downs 1999; GAO 1999; Orfield 1997; Sierra Club 1998).
The complaint is that sprawl either causes or is caused by patterns of
exclusive land use, including separation of homes, workplaces, and con-
veniences, as well as income segregation among residential communi-
ties. As the mixture of uses in a community declines, travel time and dis-
tance for those who live or work there increase. If exclusivity of use in
small areas is typical of an entire UA, one would expect an increase in
 Wrestling Sprawl to the Ground 697

the inconveniences often attributed to sprawl, such as traffic congestion,

trip length, and travel times.

Mixed land uses, or the lack thereof, is often inferred from some mea-
sure of accessibility, such as vehicle miles traveled or time spent in
travel. Such measures, however, do not distinguish between the extent
to which they result from exclusive or mixed land uses or other factors,
such as household behavior. As before, we argue that it is crucial to dis-
tinguish the mixture of land uses within a grid (which can be measured
directly) from the consequences of that pattern.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Figure 7 illustrates such an approach. In A, every square contains an

equal proportion of the UA’s residences and employment. Moreover, this
pattern of a high level of mixed use is typical of the entire area. In B,
each square contains only a single land use and represents the lowest
degree of mixed use; it is therefore more sprawl-like on this dimension.

Proximity

Proximity is the degree to which different land uses are close to each
other across a UA.

The mixed-use dimension of development patterns captures only the

extent to which small parts of a UA are typically devoted exclusively
to a single use. Proximity is the dimension that establishes the typical
distance between different uses. For example, the extent to which jobs
and housing for low-income workers are spatially mismatched affects
economic opportunity. And the average distance workers must travel
for employment, or consumers must travel to shop for convenience or
comparison goods, contributes to many of the externalities attributed
to sprawl.2 While proximity of the same uses to each other is a signifi-
cant feature in the agglomeration of related activities in urban space,
that seems a less significant feature of sprawl than the proximity of
different but complementary uses, such as housing and employment or
consumer goods.

Conceptually, proximity is the average distance people must travel

from any “home” or residential square to every other “target” or employ-
ment square. Those UAs where most people must travel great distances
have lower proximity between uses and, therefore, can be considered
to exhibit more sprawl. In figure 8, A illustrates an urban area with
high proximity of land uses, and B illustrates one with low proximity.

2
It may make sense to measure proximity between various types of nonresidential
uses if workers often travel among them.
 698 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Figure 7. Mixed Uses: The Degree to Which Two Different Land Uses
Exist in the Same Small Area and This Pattern
Is Typical throughout the UA

A
UA Typified by Mixed Uses
in Its Communities
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

B
UA Typified by Single Uses
in Its Communities

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 nonresidential units
= 1,000 residential units

Proposed operationalization of the dimensions of sprawl

The next step in our research plan is to operationalize the foregoing

dimensions of sprawl.

We first divide land use into three types: residential, nonresidential (al-
though in principle nonresidential could be further subdivided), and
nondevelopable. In some of our dimensions, the appropriate operational-
ization considers only developable land; that is, land that is nondevel-
 Wrestling Sprawl to the Ground 699

Figure 8. Proximity: The Degree to Which Different Land Uses

Are Close to Each Other across a UA

A
High Proximity of Uses
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

B
Low Proximity of Uses

= UA border
= MA
= Square mile
= One-quarter of a square mile
= Vacant parcel

= Undevelopable land
= 1,000 nonresidential units
= 1,000 residential units

opable because of natural features, public use, or regulatory barriers is

excluded.3 We superimpose on all land within the UA a series of one-
mile squares and, within these one-mile squares, a set of four one-half-
mile squares (one-fourth of a square mile each). Depending on the di-

3
This inconsistency is justifiable on conceptual grounds. For example, residential den-
sity should be computed using only developable acreage so as not to unfairly charac-
terize as more sprawled UAs with more mountains, floodplains, or parklands. Howev-
er, for proximity or centrality, it is appropriate to measure distances between locations
using all intervening acreage, whether developable or not, since such land must be
traversed regardless.
 700 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

mension, the geographic units of analysis we employ are either these

one-mile-square grids or the one-half-mile-square grids.

We now turn to the operationalization of the eight dimensions. More

detailed nomenclature and formulas are presented in appendix A.

Density

Definition: The average number of residential units or the average

number of employees per square mile of developable land in a UA.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Unit of analysis: One-mile-square grids.

Operationalization: Total number of housing units (or employees) in a

UA/area of a UA.

Continuity

Definition: The degree to which developable land has been developed

in an unbroken fashion throughout the UA.

Unit of analysis: One-half-mile-square grids.

Operationalization: A one-half-mile-square grid is considered developed

if it contains 10 or more housing units or 50 or more employees. The
proportion of all such grids in the UA that are so developed is a mea-
sure of continuity.

Concentration

Definition: The degree to which housing units or jobs are dispropor-

tionately located in a relatively few areas or spread evenly in the UA.

Unit of analysis: One-mile-square grids.

Operationalization: Three potential measures.

1. Very high density grids (with respect to housing units or employees)

as a percentage of all grids with developable land within the UA.
Very high density grids are two standard deviations or more above
the mean of the density of all grids in the 100 largest UAs (or in a
sample of the 100 largest UAs).
 Wrestling Sprawl to the Ground 701

2. The coefficient of variation (standard deviation divided by the mean)

of the density of housing units or employees among the grids of scale
m in a UA.

3. Delta index. This is analogous to the dissimilarity index and can be

interpreted as the share of land use i (e.g., housing units) that would
need to shift areal units of scale m to achieve a uniform distribution
across the UA (Massey and Denton 1988). Higher values of DELTA
indicate more concentration of a use in certain subareas, thus less
sprawl.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Clustering

Definition: The degree to which development within any one-mile-

square area is clustered within one of the four one-half-mile squares
contained within (as opposed to spread evenly throughout).

Units of analysis: One-half-mile- and one-mile-square grids.

Operationalization: The average for all one-mile squares of the standard

deviations of the density of a particular land use (e.g., housing units or
employees) among the four squares of each one-mile grid with develop-
able land, standardized by the average density across m-scale grids.

Centrality

Definition: The degree to which observations of a given land use are

located near the CBD of a UA.

Unit of analysis: One-mile-square grids.

Operationalization: Two measures. In both, the CBD is defined as the

address of city hall.

1. The average distance of a land use (e.g., housing units) from the
CBD. This is measured as the inverse of the average of the sum of
the distance from the center of the CBD grid to the center of each
one-mile-square grid weighted by the number of observations of the
land use (e.g., housing units) in the grid, with the resulting average
standardized by the square root of the area of the UA. Lower values
therefore reflect more sprawl.

2. A centralization index that measures how rapidly a given land use

accumulates relative to land area as one moves progressively out-
ward in concentric rings from the CBD (Massey and Denton 1988).
 702 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

The centralization index is computed in the following way. With Geo-

graphic Information Systems software, one draws a series of concentric
rings (bulls-eye style) from the CBD center (say, using one-mile radii).
Then, one essentially compares how fast population or any land use in
question accumulates, starting at the innermost ring and working pro-
gressively outward. This cumulative distribution is compared with the
corresponding cumulative distribution of urbanized land area as a
baseline.

If virtually all of the observations of a particular land use accumulate

within, say, the innermost two rings but these rings represent only a
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

small fraction of the UA, centrality will register a high value. At the
other extreme, if few uses are located near the center but most are in-
stead near the edge, land area will accumulate faster than the particu-
lar land use moving outward, and centrality will have a low (negative)
value, signifying a greater degree of sprawl on this dimension.

Nuclearity

Definition: The extent to which a UA is characterized by a mononuclear

pattern of development.

Unit of analysis: One-mile-square grids.

Operationalization: Nuclearity involves the identification of nodes or

nuclei by means of the following steps:

1. Identify the highest density (in terms of both housing units and,
separately, employees) per one-mile-square grid in the UA.

2. Add all adjacent grids that are within one standard deviation of the
density of this highest-density grid to the node; also include nodes
adjacent to the added nodes, provided they are within one standard
deviation of the highest-density grid. The result is the central
node, c.

3. Recalculate the density of the newly combined highest-density

nucleus c (per #2).

4. Consider all other one-mile-square grids in the UA that are within

one standard deviation of the recalculated density (per #3) as sepa-
rate nuclei, n, provided that they are not immediately adjacent to an
existing nucleus.

5. Add any grids adjacent to any nucleus identified in #4 that are with-
in one standard deviation of the recalculated highest-density nucleus
c (per #3) to the nucleus.
 Wrestling Sprawl to the Ground 703

Measurement: Two measures.

1. The number of nodes (a measure of the degree of polynuclearism).

2. The number of observations (housing units or employees) in the

central (highest-density) nucleus as a percentage of the number of
observations in all of the nuclei (a measure of mononuclearity).

Mixed uses
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Definition: The degree to which substantial numbers of two different

land uses (e.g., housing units and employees) exist within the same
area and this pattern is typical throughout the UA.

Units of analysis: One-mile-square grids.

Operationalization: To operationalize this concept, we employ a version

of Massey and Denton’s well-known exposure index (1988). The intu-
itive interpretation of this index is the average density of a particular
land use (e.g., housing units) in another land use’s (e.g., nonresidential
or employees) area.

Proximity

Definition: The degree to which a particular land use or pair of land

uses are close to each other across the UA.

Unit of analysis: One-mile-square grids.

Operationalization: This measure can be defined not only for a given

use (average distance between households, between jobs, etc.) but, po-
tentially more interesting, between uses. For example, one can define
proximity between households and jobs as the measure of sprawl most
closely associated with spatial mismatch.

Measurement: The measures we propose are adaptations of indices de-

veloped by White (1986). We first compute the weighted average dis-
tance in the UA between a given land use i and all observations of an-
other use j (including the possibility that i = j). We sequentially take
each distance between a centroid of a given one-mile-square area m and
the centroid of another one-mile-square area k and weight it by the pro-
portion of the land use of interest j in the UA represented by the target
area k. This is done using grid area m’s centroid as the origin and com-
puting the weighted distance to every other area’s centroid until all of
 704 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

the weighted distances are summed to get the average.4 This procedure
is then repeated for all one-mile-square areas as the origin point of
distances; all these observations are weighted by the proportion of the
UA’s share of land use i represented in grid area m.

Testing our definitions:

Measuring housing sprawl in 13 UAs

We are now ready to test the foregoing operationalized definitions of

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

dimensions of sprawl by applying them to a small number of UAs in the

United States. We selected 13 large areas from different regions of the
country for our prototype test. Because of both resource and time con-
straints, our test is confined to residential uses; we thus examine in the
following section only housing sprawl, with housing units as the land
use for our analysis.5 Because of this, we will not be able to test our op-
erationalizations for interuse measures, continuity, and diversity. In ad-
dition, given our constraints, we were unable to separate out develop-
able from nondevelopable land; as a consequence, all land is considered
developable.

Method

Each of the 13 UAs was first divided into one-mile-square grids, includ-
ing those that were only partially in the UA.6 Each grid was then di-
vided into four one-half-mile-square grids. A Geographic Information
System was used to estimate the fraction of each grid within the UA,
and this value, including the latitude and longitude of the grid’s centroid,
was entered into the database. Then, 1990 census block data for hous-
ing units were aggregated to create a count for each grid.7 (See appen-
dix B for a technical explanation.)

We next computed a value for each of the 13 UAs on each of the six
dimensions we included: density, concentration, clustering, centrality,
nuclearity, and proximity. The computation formulas are summarized in

4
Note that when k = m, distance = 0.
5
For an examination of land use patterns for offices, see Lang (2000a).
6
For the clustering measure, only grids that were wholly in the UA were used, thus
eliminating those that were partially in the UA but also included clearly nondevelopable
area such as lakes, rivers, etc. This was mandated because of the artificially high den-
sity variance of such boundary grids. Otherwise, cities on large bodies of water would
register less clustering ante facto than landlocked cities.
7
Blocks split by grid boundaries were assigned to the grid containing the largest share
of the block.
 Wrestling Sprawl to the Ground 705

appendix A. For concentration, we employed the delta index as described

in the Operationalization section (operationalization #3 under Concen-
tration in appendix A). For centrality, we used the weighted average
distance from city hall calculation (operationalization #1 under Central-
ity in appendix A). None of our proposed operationalizations worked
well for nuclearity, possibly because we attempted to measure housing
rather than employment nuclearity. Our proposed measure (including
as nodes all those grids within one standard deviation of the densest
grid or set of grids) yielded almost no nodes other than the central one.
Instead, we adopted a second-best operationalization that gave us a
useful measure of residential mononuclearity: the percentage of all hous-
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

ing units in the 2 percent of the densest grids in the UA that are located
in the central node, with the central node consisting of all grids in the
densest 2 percent of the grids that are contiguous and nearest city hall.
For proximity, we used the intrause measure, since we had only data
for residential land use.

After measuring and comparing the 13 areas on six dimensions of

sprawl, we will see how the results comport with our firsthand knowl-
edge of these areas, as well as the conventional wisdom.

Results

Table 1 reports the raw scores for each of the 13 UAs on each of the
six dimensions.

Inasmuch as our sample of 13 UAs is not large enough to permit factor

analysis, we created a series of “Z scores.” (A Z score is simply the num-
ber of standard deviations a UA is from the mean of the distribution for
Table 1. Indicators of Urban Sprawl

Density Concentration Clustering Centrality Nuclearity Proximity

Mean 1,407.42 0.39 0.44 167.46 62.84 0.28

Standard 389.56 0.06 0.06 25.36 25.71 0.07
deviation
Atlanta 806.25 0.28 0.38 177.86 44.17 0.2569
Boston 1,000.63 0.47 0.44 191.83 88.80 0.2754
Chicago 1,647.09 0.40 0.49 160.30 82.42 0.3392
Dallas 901.57 0.41 0.47 149.40 24.03 0.3311
Denver 1,462.90 0.31 0.41 178.22 67.40 0.1656
Detroit 1,265.60 0.34 0.39 141.51 69.94 0.2610
Houston 989.45 0.40 0.54 183.16 13.44 0.2539
Los Angeles 1,861.84 0.37 0.36 166.97 57.92 0.3800
Miami 1,883.26 0.32 0.41 125.91 43.79 0.1793
New York 1,946.48 0.51 0.51 202.24 96.56 0.4048
Philadelphia 1,483.97 0.41 0.52 192.14 89.98 0.2551
San Francisco 1,639.00 0.43 0.37 124.76 78.42 0.3041
Washington, DC 1,408.43 0.36 0.44 182.74 60.06 0.2384
 706 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

that dimension.) Each UA thus has a Z score for each of the six dimen-
sions. We then added the Z scores for each UA across all six dimensions
to obtain a composite sprawl index. This weights each of the dimen-
sions equally in calculating the index. (Analysts may choose to recom-
pute our aggregate scores using their own weighting schemes based on
the constituent scores presented.) The Z scores for each dimension and
the composite sprawl index are presented in table 2. Since more sprawl-
like conditions are rated low and less sprawl-like conditions high on
each of the dimensions, higher Z scores reflect lower levels of sprawl.

The UAs with the greatest degree of sprawl—i.e., the lowest score on the
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

composite index—were Atlanta, followed by Miami, Detroit, and Denver.

The UAs with the lowest degree of sprawl were all older areas in the
Northeast and Midwest: The New York area had the least sprawl, fol-
lowed by Philadelphia, Chicago, and Boston. This comports with the pre-
20th-century development of their cores. Atlanta and Miami represent
the archetypes of what Lang (2000b) calls “wet south region sprawl,”
where 20th-century growth has been unconstrained by the aridity of
the region. Los Angeles, often cited as the prototype for auto-spawned
sprawl, had the fifth-lowest degree of sprawl on the composite index.
Our interpretation of the relative ranking of Los Angeles is that the
conventional wisdom with respect to the degree of its sprawl relative to
other areas may simply be wrong. Natural attributes like topography
and aridity have apparently constrained land use patterns more than is
usually understood (Lang 2000b).

It is possible that an extreme score on one or two dimensions may be

driving the overall value of the composite index. To test this, we ranked
each UA on each of the six dimensions and then summed across dimen-
sions. (Additional means of weighting to compute the aggregate ranking
are possible; ours is merely illustrative.) The top rank meant the lowest
sprawl-like value on that dimension. Thus, the area with the lowest
summary ranking score (New York) had the least sprawl (see table 3).
Using this method led to very little change in the composite index:
Only one area, Washington, DC, which moved from the sixth-lowest to
the eighth-lowest degree of sprawl, moved more than one place in the
rankings.

Table 3 allows us to more easily observe the extent of variation across

the dimensions. Areas with low sprawl rankings on some dimensions
did not necessarily have consistently low rankings overall. Philadelphia,
for example, with the second-lowest ranking for overall sprawl, also
ranked second in terms of clustering, centrality, and mononuclearity, but
ninth with respect to proximity. Los Angeles, with the fifth-lowest rank-
ing overall, ranked second lowest in sprawl with respect to density and
proximity, but last with respect to clustering. Even Atlanta, which had
the most sprawl in the overall ranking, ranked only seventh on the cen-
trality dimension and eighth on proximity.
 Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Table 2. Urban Sprawl Indicators: Z Scores

Composite
Density Concentration Clustering Centrality Nuclearity Proximity Index Rank

New York 1.38 1.92 1.14 1.37 1.31 1.77 8.90 1

Philadelphia 0.20 0.38 1.31 0.97 1.06 –0.36 3.56 2
Chicago 0.62 0.23 0.81 –0.28 0.76 0.84 2.97 3
Boston –1.04 1.31 –0.01 0.96 1.01 –0.07 2.15 4
Los Angeles 1.17 –0.24 –1.34 –0.02 –0.19 1.42 0.80 5
Washington, DC 0.00 –0.39 –0.01 0.60 –0.11 –0.60 –0.50 6
San Francisco 0.59 0.69 –1.17 –1.68 0.61 0.34 –0.63 7
Houston –1.07 0.23 1.64 0.62 –1.92 –0.38 –0.89 8
Dallas –1.30 0.38 0.48 –0.71 –1.51 0.72 –1.94 9
Denver 0.14 –1.16 –0.51 0.42 0.18 –1.63 –2.56 10
Detroit –0.36 –0.70 –0.84 –1.02 0.28 –0.28 –2.93 11
Miami 1.22 –1.01 –0.51 –1.64 –0.74 –1.44 –4.11 12
Atlanta –1.54 –1.63 –1.00 0.41 –0.73 –0.33 –4.83 13

Table 3. Ranking of Urban Sprawl Indicators

Density Concentration Clustering Centrality Nuclearity Proximity Total Rank

New York 1 1 3 1 1 1 8 1
Philadelphia 6 4 2 2 2 9 25 2
Chicago 4 6 4 9 4 3 30 3
Boston 10 2 6 3 3 6 30 3
Los Angeles 2 8 13 8 9 2 42 5
San Francisco 5 3 12 13 5 5 43 6
Houston 11 6 1 4 13 10 45 7
Washington, DC 8 9 6 5 8 11 47 8
Dallas 12 4 5 10 12 4 47 8
Denver 7 12 8 6 7 13 53 10
Wrestling Sprawl to the Ground

Detroit 9 10 10 11 6 7 53 10
Miami 3 11 8 12 11 12 57 12
Atlanta 13 13 11 7 10 8 62 13
707
 708 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Further evidence that our six operationalizations of the dimensions of

sprawl are indeed related to the same core construct yet are quite inde-
pendent is provided by a correlation analysis.8 Pearsonian correlations
among the six revealed only three that were statistically significantly
different from zero: proximity and concentration (0.65), centrality and
clustering (0.53), and clustering and concentration (0.52). Across all in-
dicators, the correlations averaged 0.29.9

Summary and next steps

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Each dimension of urban land use we have defined is conceptually dis-

tinct from each of the others. Each is a condition of land use that can
be observed. Each is free of judgment about its intrinsic value. And as
an objective condition, each is independent of those policies, practices,
or preconditions that may have caused it or of those things that may
be consequences of its existence. Each dimension can be measured on
a continuum and compared across UAs. Individually and in combina-
tion, low scores on the dimensions correspond to intuitive understand-
ing and general observations about the extent of sprawl in the UAs to
which they were applied. Moreover, the rankings do not appear to be
unreasonable.

This exercise also reveals the need for some refinements. Conceptually,
more work is needed to define an appropriate study area that includes
related sprawl-type development located beyond the UA and functional-
ly attached to it without encompassing large areas that are genuinely
rural in character. A precise definition of undevelopable land would im-
prove the measurement of some dimensions. Simplification of some of
our measures to increase their transparency would be desirable if it can
be done without impairing their effectiveness. And it will be important
to develop measures of nonresidential development.

Our prototype test of operational measures of housing sprawl proved

encouraging. With improvements identified by this effort, a conceptual-
ly clearer and more coherent set of sprawl measures is attainable for a
large set of MAs. This would permit using statistical techniques such as
factor analysis to determine the extent to which dimensions are associ-
ated most closely with others and to develop indices that combine some
or all of the dimensions. The objective of this sort of analysis should be
to specify several different types of sprawl—indicated by low values on
related dimensions of the eight urban development patterns—and to
produce an overall sprawl index for UAs.
8
This analysis is available from the first author on request.
9
There were two negative correlation coefficients: density and compactness (–0.09) and
density and centrality (–0.17). However, both fell far short of the value needed for 0.05
statistical significance for such a sample size: 0.51.
 Wrestling Sprawl to the Ground 709

The resulting typologies and indices can then be used as independent

variables in research that examines the effect of different degrees or
types of sprawl on the spatial distribution of income in MAs, the inci-
dence of poverty, or externalities of urban growth patterns. The respec-
tive dimensions and types of sprawl can be compared to determine
which, if any, have a greater effect on variables that are considered to
be consequences of sprawl, and sprawl can be compared with other vari-
ables to determine their relative power to explain urban conditions. As
dependent variables, specific dimensions can help identify policies and
practices that tend to induce particular land use patterns.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Conceptual clarity about the dimensions of sprawl and the ultimate op-
erationalization of these concepts should thus make it possible to inform
policy far more effectively as officials attempt to affect the causes of the
several dimensions or grapple with their consequences. It should also
be possible to measure sprawl for one or more individual areas, or even
parts of areas such as counties, in a time series that can permit in-
formed inferences about the effectiveness of policies and practices de-
signed to manage urban development and its consequences.

Appendix A

Formal operationalization of the dimensions of sprawl

Nomenclature

Let

i = a particular type of land use or spatially based observation, in

our case, either residential use (for which we use housing units) or
nonresidential use (for which we use employees).

j = a different type of land use from i.

u = the largest spatial scale used in the analysis; the entire UA.

m = the medium spatial scale used in the analysis: one square mile;
1, 2, …, m, …, M such medium-sized squares comprise the UA u.

s = the smallest spatial scale used in the analysis: one-quarter of a

square mile (a square with a half-mile per side); 1, 2, …, s, …, S
such small-sized squares comprise the UA u10
10
Using grids of any size produces some distortion in results because of the arbitrary
location of the grid lines. There is a need, however, for a common metric that can be
applied across all areas. The use of one-mile-square and one-quarter-mile-square grids
provides a scale that can be readily understood and visualized.
 710 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

T(i)u = the total number of observations (population) of land use i

in UA u.

T(i)m = the total number of observations (population) of land use i

in land area m (that is also within u).

T(i)s = the total number of observations (population) of land use i

in land area s (that is also within u).

Pm = proportion of land area of spatial scale m within u.

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Ps = proportion of land area of spatial scale s within u.

M
Au = the total developable land area within UA u; = Σ Pm (Am).
m=1

Am = the total developable land area within a grid of spatial scale

m = Pm.

As = the total developable land area within a grid of spatial scale

s = 0.25*Ps.

D(i)u = the density of land use i over the developable UA = T(i)u/Au.

D(i)m = the density of land use i over the developable area in m =

T(i)m/Am.

D(i)s = the density of land use i over the developable area in s =

T(i)s/As.

F[k,m] = the distance between the centroids of grid k and grid m.

Given this nomenclature, we turn to how each of our dimensions can

be formally operationalized.

Density

M
DENS(i)u = D(i)u = T(i)u/Au = Σ [T(i)m]/Au (1)
m=1

[min = 1,000 per square mile (U.S. Bureau of the Census standard
for a UA); max = unlimited]
 Wrestling Sprawl to the Ground 711

Continuity

S
CONT(i)u Σ [D(i)s > 9 Residences and 49 Employees = 1; (2)
s=1

0 otherwise]/S

[min = 0; max = 1]

Concentration (three alternatives)

Downloaded by [Syracuse University Library] at 07:41 30 May 2013

• Very high density grids (with respect to housing units or employees)

as a percentage of all grids with developable land within the UA.
Very high density grids are grids that are two standard deviations
or more above the mean of the density of all grids in the 100 largest
UAs (or in a sample of the 100 largest UAs) or,

• The coefficient of variation.

M M
COV(i) u = (Σ [D(i)m – D(i)u]2/M)1/2 / [Σ D(i)m/M] (3)
m=1 m=1

or,

• Delta index.

M
DELTA(i)u = (1/2) Σ [T(i)m/T(i)u] – [Am/Au] (4)
m=1

Clustering

M 4 M
CLUS(i)u = [Σ (Σ [D(i)s – D(i)m]2/4)1/2/M] / [Σ D(i)m/M] (5)
m=1 s=1 m=1

Centrality (two alternative measures)

• The average distance of a land use (e.g., housing units) from the
CBD.
 712 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

M
CBDDIST = T(i)u(A1/2)/Σ F(k,m) T(i)m (6)
m=1

• The centralization index:

H H
CEN(j)u = Σ [T(j)h – 1][Ah] – Σ [T(j)h][Ah – 1] (7)
h=1 h=1
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Nuclearity

Nuclearity involves the identification of nodes or nuclei. The identifica-

tion proceeds in the following steps:

1. Identify the highest density (in terms of both housing units and,
separately, employees) per one-mile-square grid in the UA.

2. Add all adjacent grids within one standard deviation of the density
of this highest-density grid to the node, as well as nodes adjacent to
the added nodes, provided they are within one standard deviation of
the highest-density grid. The result is the central node, c.

3. Recalculate the density of the newly combined highest-density

nucleus c (per #2)

4. Consider all other one-mile-square grids in the UA that are within

one standard deviation of the recalculated density (per #3) as sepa-
rate nuclei, n, provided that they are not immediately adjacent to
an existing nucleus.

5. Add any grids adjacent to any nucleus identified in #4 that are with-
in one standard deviation of the recalculated highest-density nucle-
us c (per #3) to the nucleus.

Two alternative measures can be defined now:

NODES = c + Σn = c + N (8)

N
MONONUCLEAR = T(i)c/[T(i)c + Σ T(i)n] (9)
n=1
 Wrestling Sprawl to the Ground 713

Mixed uses

M
MXU (j to i) = Σ (D(i)m × [D(j)m/T(j)u])/D(i)u (10)
m=1

[min = 0; max = max D(i)m observed in any area occupied by j]

Proximity
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

The average distance between any two randomly chosen observations of

different land uses i and j can be expressed as

M M
DIST(i, j)u = Σ Σ F(i, j)mk [T(j)k/T(j)u](T(i)m/T(i)u) (11)
m=1 k=1

[min = 1 mile; max = unlimited]

Analogously, the average distance between any two randomly chosen

observations of the same land use j in the UA can be expressed as

M M
DIST(j, j)u = Σ Σ F(j, j)mk [T(j)k × T(j)m]/(T(j)u)2 (12)
m=1 k=1

It makes sense to standardize these distance measures, inasmuch as

bigger UAs will tautologically have greater average distances between
any pair of land uses. For this standardization, we compute the aver-
age distance between centroids of the M medium-scale grid areas:

M M
DISTu = Σ Σ Fmk/M (13)
m=1 k=1

[min = 1 mile; max = unlimited]

From the above terms, we can express three alternative measures of

proximity: intrause, interuse, and (weighted) average across uses:

PROX(j) = [DISTu/DIST(j, j)] – 1 (14)

PROX(ij) = [DISTu/DIST(i, j)] – 1

PROX(u) = (DISTu [T(i)u + T(j)u])/(T(i)u[DIST(i, i)]

+ T(j)u[DIST(j, j)] ) – 1
 714 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

All three versions of proximity indices have the mathematical property

that they will equal zero if observations of the given land use (or aver-
age of land uses) are separated, on average, as are all parcels of land in
the UA. Positive values of these indices signify that observations of the
given use are more proximate, on average, than are all parcels to each
other; the maximum value is undefined since the intra- or interuse dis-
tance may be very small compared with all parcels. Conversely, negative
values of these indices denote use separations greater than those among
parcels; the indices approach but cannot equal –1 as a minimum.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Appendix B

Grid formulation methodology

The data source for these calculations was U.S. Census Bureau 1995
TIGER/Line-based files from Environmental Systems Research Insti-
tute, Inc. (ESRI). MapInfo Professional v. 5.5 and ArcView v. 3.2 were
the types of Geographic Information Systems software used.

Methodological summary

Data for all 13 UAs (UA boundaries and census blocks for all counties
comprising them) were acquired from ESRI and converted to MapInfo
format for analysis. After importation, only block-level geography with-
in the UAs studied was kept. Block-level housing unit data was then
“joined” to the geography.

Grid cell creation

One set each of one-mile-square and one-half-mile-square cells were

then created from a rectangular grid boundary drawn around each UA
to encompass it in its entirety. Grid tables were then created on the
basis of these bounding rectangles.

UA calculation

Once grid cells were created, they were assigned a value between 0 and
1 based on the percentage of their area that fell within the UA being
studied; for example, if a cell fell wholly within the UA boundary, it was
assigned a value of 1. Those entirely outside the area were assigned a
value of 0. Those straddling the border were assigned the geographical
proportion that fell within the area.
 Wrestling Sprawl to the Ground 715

Housing unit calculation

The methodology for this calculation was somewhat different in that

block-level geography was used in the one-mile and one-half-mile cell
calculations. However, the cell file geography remained unaltered. The
number of housing units was assigned on the basis of the proportionate
sum of that portion of the blocks that partially or wholly fell within the
boundaries of the individual cells. For example, if it was geographically
determined that a particular one-mile cell contained all or part of four
blocks that crossed its borders, then only the geographically based pro-
portion of housing units of each block that fell within that cell were
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

summed to it.

Authors
George Galster is Hilberry Professor of Urban Affairs at Wayne State University. Royce
Hanson is a Visiting Professor in the Policy Sciences Graduate Program at the Univer-
sity of Maryland, Baltimore County. Michael R. Ratcliffe is a Geographer in the Popula-
tion Distribution Branch, Population Division, at the U.S. Bureau of the Census. Harold
Wolman is Director of the George Washington Institute of Public Policy and Professor
in the Department of Political Science at George Washington University. Stephen Cole-
man and Jason Freihage are Graduate Research Assistants in the Policy Sciences Grad-
uate Program at the University of Maryland, Baltimore County.

The authors thank Donald Krueckeberg and Stephen Malpezzi for their helpful
suggestions.

The Center for Urban Studies of the College of Urban, Labor, and Metropolitan Affairs
at Wayne State University did a splendid job of creating the databases and program-
ming the calculations for our sprawl indices. Special thanks go to Geographic Informa-
tion Systems specialist Doug Towns and programmers Patricia Case and Dean Erst-
gaard for their outstanding work.

References
Abrams, Charles. 1971. The Language of Cities. New York: Viking.

Altshuler, Alan, and José A. Gomez-Ibanez. 1993. Regulation for Revenue: The Political
Economy of Land Use Exactions. Washington, DC: Brookings Institution.

American Society of Civil Engineers. 1999. ASCE News “Issue Focus.” World Wide Web
page <[Link] (accessed November 1999).

Black, Thomas J. 1996. The Economics of Sprawl. Urban Land 55(3):6–52.

Burchell, Robert W., and David Listokin. 1991. Fiscal Studies for the Governor’s Commis-
sion on Growth in the Chesapeake Bay Region. Baltimore: Maryland Office of Planning.

Burchell, Robert W., Naveed A. Shad, David Listokin, Hilary Phillips, Anthony Downs,
Samuel Siskin, Judy S. Davis, Terry Moore, David Helton, Michelle Gall, and ECONorth-
west. 1998. Costs of Sprawl—Revisited. Washington, DC: National Academy Press.
 716 G. Galster, R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage

Cervero, Robert. 1991. Land Use and Travel at Suburban Activity Centers. Transporta-
tion Quarterly 45(4):479–91.

Clawson, Marion. 1962. Urban Sprawl and Speculation in Suburban Land. Land Eco-
nomics 38(2):94–111.

Downs, Anthony. 1998. How America’s Cities Are Growing: The Big Picture. Brookings
Review 16(4):8–12.

Downs, Anthony. 1999. Some Realities about Sprawl and Urban Decline. Housing Policy
Debate 10(4):955–74.

Ewing, Reid. 1997. Is Los Angeles–Style Sprawl Desirable? Journal of the American
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Planning Association 63(1):107–26.

Geddes, Robert. 1997. Metropolis Unbound: The Sprawling American City and the
Search for Alternatives. American Prospect 8(35):40. World Wide Web page <[Link]
[Link]/v8/35/[Link]> (accessed September 2001).

Gordon, Peter, and Harry W. Richardson. 1997. Are Compact Cities a Desirable Planning
Goal? Journal of the American Planning Association 63(1):89–106.

Harvey, Robert O., and William A. V. Clark. 1965. The Nature and Economics of
Sprawl. Land Economics 41(1):1–9.

Lang, Robert. 2000a. Office Sprawl: The Evolving Geography of Business. Washington,
DC: Brookings Institution, Center on Urban and Metropolitan Policy.

Lang, Robert. 2000b. Open Spaces, Bounded Places: Do Western Cities Really Sprawl?
Unpublished paper. Fannie Mae Foundation.

Lockwood, Charles. 1999. Sprawl. Hemispheres, September, pp. 82–91.

Malpezzi, Stephen. 1999. Estimates of the Measurement and Determinants of Urban

Sprawl in U.S. Metropolitan Areas. Unpublished paper. University of Wisconsin,
Madison Center for Urban Land Economics Research.

Massey, Douglas S., and Nancy Denton. 1988. The Dimensions of Residential Segrega-
tion. Social Forces 67:281–313.

Mieszkowski, Peter, and Barton Smith. 1991. Analyzing Urban Decentralization.

Regional Science and Urban Economics 21(2):183–99.

Mills, David E. 1980. Growth, Speculation, and Sprawl in a Monocentric City. Journal
of Urban Economics 10:201–26.

Moskowitz, Harvey S., and Carl G. Lindbloom. 1993. The New Illustrated Book of
Development Definitions. New Brunswick, NJ: Rutgers University, Center for Urban
Policy Research.

Myers, Dowell, and Alicia Kitsuse. 1999. The Debate over the Future Density of Devel-
opment. Working paper. Lincoln Institute of Land Policy.

Orfield, Myron. 1997. Metropolitics: A Regional Agenda for Community and Stability.
Washington, DC: Brookings Institution, and Cambridge, MA: Lincoln Institute of Land
Policy.
 Wrestling Sprawl to the Ground 717

Popenoe, David. 1979. Urban Sprawl: Some Neglected Sociology. Sociology and Social
Research 31(2):181–88.

Sierra Club. 1998. Sprawl: The Dark Side of the American Dream. World Wide Web page
<[Link] (accessed September 2001).

Torrens, Paul M., and Marina Alberti. 2000. Measuring Sprawl. Unpublished Paper
No. 27. University College, London. Center for Advanced Spatial Analysis.

U.S. General Accounting Office. 1999. Community Development: Extent of Federal

Influence on “Urban Sprawl” Is Unclear. Letter Report, April 30. GAO/RCED 99–87.
Washington, DC.
Downloaded by [Syracuse University Library] at 07:41 30 May 2013

Vermont Forum on Sprawl. 1999. Sprawl Defined. World Wide Web page <[Link]
[Link]/[Link]> (accessed September 2001).

White, Michael. 1986. Segregation and Diversity: Measurement in Population Distrib-

ution. Population Index 52:198—221.