University of Southern California Dissertations and Theses

Urban spatial structure, commuting, and growth in U.S. metropolitan areas

(USC Thesis Other)

Urban spatial structure, commuting, and growth in U.S. metropolitan areas

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

URBAN SPATIAL STRUCTURE, COMMUTING, AND GROWTH
IN US METROPOLITAN AREAS

by

Bumsoo Lee

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PLANNING)

December 2006

Copyright 2006 Bumsoo Lee

ii
DEDICATION

To Sumi and Cherin, and my parents

iii
TABLE OF CONTENTS

DEDICATION............................................................................................................................. ii

TABLE OF CONTENTS.............................................................................................................. iii

LIST OF TABLES ....................................................................................................................... v

LIST OF FIGURES ................................................................................................................... viii

ABSTRACT............................................................................................................................... ix

CHAPTER 1. INTRODUCTION............................................................................................... 1
1.1. Background.......................................................................................................... 1
1.2. Research Questions and Approach ...................................................................... 4
1.3. Organization of the Thesis................................................................................... 5

CHAPTER 2. QUANTIFYING URBAN SPATIAL STRUCTURE ............................................... 8
2.1. Introduction.......................................................................................................... 8
2.2. Multiple dimensions in urban sptial structure...................................................... 9
2.3. Identifying Urban employment centers ............................................................. 12
2.3.1. Absolute Criteria: Minimum Density (MD) Procedure ............................. 13
2.3.2. Relative Criteria: Geographically Weighted Regression (GWR) Procedure.. 14
2.4. Spatial indices of Centrality and Concentration ................................................ 17
2.5. Estimation Results: Description of Modern Metropolitan Spatial Structure ..... 19
2.5.1. Review of Estimated Spatial Indicators..................................................... 19
2.5.2. Description of Modern Metropolitan Spatial Structure ............................. 23

CHAPTER 3. DETERMINANTS OF METROPOLITAN SPATIAL STRUCTURE ..................... 36
3.1. Introduction........................................................................................................ 36
3.1.1. Economic Theories of Urban Spatial Structure ......................................... 37
3.1.2. Path Dependence in Spatial Evolution....................................................... 39
3.2. Descriptive Analysis of Urban Spatial Structure............................................... 41
3.2.1. Urban Spatial Structure by Metropolitan Size........................................... 41
3.2.2. Urban Spatial Structure by Metropolitan Age ........................................... 44
3.3. Determinants of urban Spatial Structure............................................................ 46
3.3.1. Model Specification................................................................................... 46
3.3.2. Determinants of CBD Size and Employment Centralization..................... 55
3.3.3. Determinants of Metropolitan Polycentricity ............................................ 61
3.3.4. Determinants of Employment Dispersion.................................................. 69
3.4. Discussion.......................................................................................................... 73

iv

CHAPTER 4. TRENDS IN URBAN SPATIAL STRUCTURE ................................................... 75
4.1. Introduction........................................................................................................ 75
4.1.1. Perspectives on Spatial Trends .................................................................. 76
4.1.2. Overview of Study Areas........................................................................... 80
4.2. Trends in Urban Spatial Structure ..................................................................... 83
4.2.1. Spatial Changes towards Dispersion and Decentralization........................ 83
4.2.2. Growth Patterns of Metropolitan Employment Centers ............................ 87
4.2.3. Three Patterns of Spatial Evolution ......................................................... 101
4.3. Discussion........................................................................................................ 105

CHAPTER 5. URBAN SPATIAL STRUCTURE AND COMMUTE ECONOMIES ................... 110
5.1. Introduction...................................................................................................... 110
5.2. Descriptive Analysis: Commute Time by Location Type ............................... 113
5.3. Statistical Analysis: Determinants of Average Commute Time ...................... 123
5.3.1. Model Specification................................................................................. 123
5.3.2. Estimation Results ................................................................................... 125
5.4. Discussion........................................................................................................ 129

CHAPTER 6. URBAN SPATIAL STRUCTURE AND METROPOLITAN GROWTH............... 130
6.1. Introduction...................................................................................................... 130
6.2. Model Specification......................................................................................... 133
6.3. Estimation Results ........................................................................................... 138
6.4. Discussion........................................................................................................ 147

CHAPTER 7. CONCLUSIONS AND DISCUSSION ............................................................... 148

REFERENCES....................................................................................................................... 152

v
LIST OF TABLES
Table 2-1. Applications of minimum density procedures in previous studies....................... 14
Table 2-2. List of centrality and concentration indices.......................................................... 18
Table 2-3. Summary statistics of spatial indicators ............................................................... 21
Table 2-4. Correlation coefficients among spatial indicators ................................................ 22
Table 2-5. Employment shares by location type by the GWR procedure.............................. 25
Table 2-6. Employment shares by location type by the MD procedure ................................ 27
Table 2-7. Centrality and concentration indices .................................................................... 29
Table 2-8. Spearman rank-order correlation coefficients ...................................................... 31
Table 3-1. Center and dispersed employment shares by population size group
(GWR method).................................................................................................... 43
Table 3-2. Center and dispersed employment shares by population size group
(MD method)....................................................................................................... 43
Table 3-3. Centrality and concentration indices by metropolitan population size
class ..................................................................................................................... 43
Table 3-4. Center and dispersed employment shares by metro age (GWR method)............. 45
Table 3-5. Center and dispersed employment shares by metro age (MD method)................ 45
Table 3-6. Centrality and concentration indices by metropolitan age class........................... 45
Table 3-7. Rotated component pattern of five industrial structure factors ............................ 49
Table 3-8. Definition of variables.......................................................................................... 53
Table 3-9. Descriptive statistics of variables......................................................................... 54
Table 3-10. Determinants of CBD (GWR method) employment share: OLS
results .................................................................................................................. 57

vi

Table 3-11. Determinants of main center (MD method) employment share: OLS
results .................................................................................................................. 58
Table 3-12. Determinants of centrality factor score: OLS results ......................................... 59
Table 3-13. Determinants of main center (MD method) employment share: 2 SLS
results .................................................................................................................. 60
Table 3-14. Determinants of number of subcenters (GWR method): NB results.................. 63
Table 3-15. Determinants of number of subcenters (MD method): NB results..................... 64
Table 3-16. Determinants of subcenters’ (GWR method) employment share: OLS
results .................................................................................................................. 65
Table 3-17. Determinants of subcenters’ (MD method) employment share: OLS
results .................................................................................................................. 66
Table 3-18. Determinants of subcenters’ (MD method) employment share: 2SLS
results .................................................................................................................. 67
Table 3-19. Regressions of subcenters’ employment share on CBD/Main Center
share .................................................................................................................... 68
Table 3-20. Determinants of dispersed employment share (GWR method): OLS
results .................................................................................................................. 70
Table 3-21. Determinants of dispersed employment share (MD method): OLS
results .................................................................................................................. 71
Table 3-22. Determinants of dispersion factor score: OLS results........................................ 72
Table 4-1. Characteristics of six metropolitan areas.............................................................. 82
Table 4-2. Changes in centralization and concentration indices............................................ 86
Table 4-3. Centers employment trends in New York, 1990 to 2000 ..................................... 89
Table 4-4. Centers employment trends in Los Angeles, 1990 to 2000.................................. 90
Table 4-5. Centers employment trends in Boston, 1990 to 2000........................................... 91

vii

Table 4-6. Centers employment trends in Portland, 1990 to 2000 ........................................ 92
Table 4-7. Centers employment trends in San Francisco, 1980 to 2000 ............................... 93
Table 4-8. Centers employment trends in Philadelphia, 1980 to 2000.................................. 94
Table 5-1. Commute times by location type of workplace defined by the GWR
procedure........................................................................................................... 117
Table 5-2. Commute times by location type of workplace defined by the MD
procedure........................................................................................................... 119
Table 5-3. Descriptive statistics of variables....................................................................... 124
Table 5-4. Mean metro commute time regression results.................................................... 127
Table 5-5. Mean CBD/Main center commute time regression results................................. 128
Table 6-1. Definition of vairables........................................................................................ 136
Table 6-2. Descriptive statistics of variables....................................................................... 137
Table 6-3. Metropolitan growth models without urban spatial structure variables ............. 141
Table 6-4. Metropolitan population growth models with urban spatial structure
variables (GWR results) .................................................................................... 142
Table 6-5. Metropolitan employment growth models with urban spatial structure
variables (GWR results) .................................................................................... 143
Table 6-6. Metropolitan population growth models with urban spatial structure
variables (MD results) ....................................................................................... 144
Table 6-7. Metropolitan employment growth models with urban spatial structure
variables (MD results) ....................................................................................... 145
Table 6-8. Varying growth effects of spatial structure depending on metropolitan
size..................................................................................................................... 146

viii
LIST OF FIGURES
Figure 2-1. Centrality versus concentration by the GWR procedure..................................... 33
Figure 2-2. Centrality versus concentration by the MD procedure ....................................... 34
Figure 2-3. Centrality versus concentration measured by indices factor scores.................... 35
Figure 4-1. Changes in employment shares by density quintile (decile)............................... 85
Figure 4-2. Employment centers in the New York metropolitan area, 1990 to
2000..................................................................................................................... 95
Figure 4-3. Employment centers in the Los Angeles metropolitan area, 1990 to
2000..................................................................................................................... 96
Figure 4-4. Employment centers in the Boston metropolitan area, 1990 to 2000 ................. 97
Figure 4-5. Employment centers in the Portland metropolitan area, 1990 to 2000 ............... 98
Figure 4-6. Employment centers in the San Francisco metropolitan area, 1980 to
2000..................................................................................................................... 99
Figure 4-7. Employment centers in the Philadelphia metropolitan area, 1980 to
2000................................................................................................................... 100
Figure 4-8. Changes of employment shares for both definitions of centers, 1980
to 2000............................................................................................................... 109
Figure 5-1. Mean commute time by workplace location type versus metro
population size................................................................................................... 121

ix
ABSTRACT

There have been “qualitative changes” in metropolitan spatial structure in recent
decades. While these changes have been widely recognized, much less is known about the
specifics – the forms, causes, and consequences of the spatial changes. In that regard, this
research aims to address several questions: What are the prominent features of emerging
urban forms? Are cities becoming more edgy or more edgeless? What are primary forces
driving the spatial changes? What are the consequences of the spatial changes in daily
commuting and urban economic growth?
To answer these questions, this research attempts to uncover the current stage and
directions of spatial evolution, investigate the driving forces of spatial changes, and probe
the links between urban spatial structure, commuting economies, and economic growth in
the contemporary US metropolitan areas.
In Chapter 2, I defined and estimated sets of spatial structure variables by
identifying employment centers consistently within 79 metropolitan areas with population
over one-half million. These spatial descriptors show that one of the most important features
of the modern metropolis is predominant dispersion. Average dispersed employment share
was 82 percent by the GWR method and 73 percent by the minimum density method.
Surprisingly, the majority of jobs were diffused outside any type of employment centers in
all 79 metropolitan areas without exception. Findings from this Chapter parallel the results

x
of Gordon and Richardson (1996) and Lang (2003) – spatial evolution “beyond
polycentricity”.
Chapter 3 presented a series of statistical analyses testing the determinants of
manifold dimensions of urban spatial structure. The results provided valuable findings that
generally conform to the predictions from urban economic theories and path dependence
perspectives.
Larger metropolitan areas tend to have smaller CBD employment shares and more
decentralized and polycentric structures. However, population size was not significant in
explaining another spatial dimension, employment dispersion. Congestion was a significant
contributor to subcenter formation, but was less significant in employment share models.
Industrial composition was also found to be an important spatial determinant, confirming
that different industries are subject to different agglomeration economies with varying
geographical ambits.
The path dependence in urban spatial structure was indirectly identified in two ways.
Recently developed metropolitan areas have smaller CBD and are more decentralized than
pre-war metros; while metros that reached the half of current population 35 to 60 years ago
had more polycentric structure. Second, metros with a strong agglomeration in the urban
core tended to have fewer subcenters and smaller subcenter employment shares.
Chapter 4 explored spatial changes in six metropolitan areas for the last two decades
to address the question whether they are increasingly edgy or edgeless. Findings paralleled
the results of Gordon and Richardson (1996): Jobs continued to decentralize from the
metropolitan core to the suburbs during the 1980s and 1990s and job dispersion was a more
common phenomenon than subcentering.

xi
Nevertheless, the results showed significant variation in spatial decentralization
trends rather than a uniform linear process from monocentric through polycentric, and to
dispersed structure. New York and Boston, with big and long established CBDs, were less
subject to decentralization. The polycentricity of Los Angeles and San Francisco was
reinforced in the last decade, while job dispersion was predominant in Portland and
Philadelphia. Each metro seems to have developed a unique pattern of decentralization, in
light of their histories and circumstances to limit the growth of commuting times.
These findings suggest two important theoretical implications. First, the
geographical and historical contexts of an individual metropolis strongly affect the path that
it takes in response to global trends such as ever decreasing transportation costs and
information technology (IT) development. Also, it seems that there is a “self reinforcing”
pattern in spatial development as firstly observed in technology adoption and industrial
development. Second, industrial composition and restructuring is an important part of the
path dependent spatial evolution processes.
Chapter 5 presented a study on the commuting impacts of metropolitan level spatial
structure. Descriptive analysis identified large potential for commute time saving by spatial
restructuring towards more polycentric and dispersed form, particularly in large metropolitan
areas. Notwithstanding the potential, however, I found only partial commute time saving
with the spatial adjustment in regression analyses. While employment dispersion helped
reduce commute time, polycentricity was not significant.
The insignificance of polycentric spatial dimension can be interpreted in two ways.
First, CBD/main center employment share may be already too small to affect metro wide

xii
average commute time. Second, polycentrization may have system wide effects such as
increased cross commuting, offsetting potential commute time savings.
Chapter 6 examined how the links between metropolitan spatial structure and
economic growth depend on the size of the metropolis. Consistent with theories of urban
system and evolution, growth effects of employment dispersion were found to be dependent
on metropolitan size. A metropolitan area with more clustered spatial form grows faster
when it is small; whereas more dispersion leads to higher growth rate as it grows large. Just
as a city needs to successfully take on higher order functions and economic activities to
move upward within the national urban system, it also needs to restructure its spatial form in
a way to mitigate congestion or other diseconomies of size for continued growth.
Therefore, attempts to find one particular efficient urban form may not be promising,
just as the efforts to find the optimal city size have not been fruitful. Efficient spatial
structure may depend not only on the city size but also on other urban attributes such as
industrial structure and the shape of transportation networks, which are products of the
historical path of urban development. Insignificant growth effects of polycentric versus
monocentric structure imply that there may exist many plausibly competitive urban forms
and different paths of spatial evolution.

1
CHAPTER 1.
INTRODUCTION

1.1. BACKGROUND
There have been “qualitative changes” in metropolitan spatial structure in recent
decades (Anas, Arnott, and Small 1998; Clark 2000). While it is widely recognized, much
less is known about the specifics – the forms, causes, and consequences of the spatial
changes. What are the prominent features of emerging urban forms? What are primary
forces driving the spatial changes? What are the consequences of the spatial changes in our
daily lives?
Two aspects of the spatial evolution in metropolitan areas have drawn academic
interests. Whereas a large body of urban literature had focused on the transformation from
monocentric to polycentric spatial structures; more recent studies suggest the case for the
“generalized dispersion” of economic activities “beyond polycentricity” (Gordon and
Richardson 1996). The titles of two widely cited books, ‘Edge City’ (Garreau 1991) and
‘Edgeless Cities’ (Lang 2003), well depict the two facets of emerging urban spatial structure.
Edge city is a journalistic interpretation of polycentric spatial structure which is
characterized by multiple nodes of urban activities. In a polycentric metropolis, smaller
business concentrations at urban edges may have location edges avoiding congestions and
high land prices, while diseconomies in the urban core tend to outweigh the benefits of
agglomeration for many business sectors.

2
About a decade after Joel Garreau wrote that the rise of these edge cities signals a
new era of urban development, Robert Lang reported that it is the prevalence of edgeless
cities that is a distinguishing feature of modern metropolitan landscape. This would be the
case if economies of clustering have been diluted or diffused throughout the metropolitan
region as individual mobility and metro-wide accessibility improves dramatically due to the
development of transportation and communication technology.
Nevertheless, we have only limited evidence to answer which urban form is
dominant. Lang’s (2003) study was only on office sector and Gordon and Richardson’s
(1996) study was drawn from a single metropolis, Los Angeles.
What are the commuting consequences of the spatial changes toward polycentric or
dispersed structure? Urban economic theories predict that spatial adjustments in cities
should occur in a way that shortens commute times of residents as a city grows. Individual
households and firms “co-locate” to reduce commute time in response to congestion and this
spatial adjustment can be more easily made in dispersed metropolitan space with many
alternative employment centers and residential location choices (Gordon, Richardson, and
Jun 1991; Levinson and Kumar 1994).
On the other hand, most urban planners believe that dispersion of jobs and
population or sprawl type development causes more frequent and longer travels, more auto
uses, and hence more congestion (Sarzynski et al. 2006). Some authors point out that the co-
location process may not occur properly due to growing two-earner households, location
barriers and restricted residential and job mobility, and increased auto dependency (Cervero
and Landis 1992).

3
The links between urban form and transportation have been heavily debated in
recent years. A number of surveys of empirical studies (Anderson, Kanaroglou, and Miller
1996; Badoe and Miller 2000; Crane 2000; Ewing and Cervero 2001) report that the results
are mixed, with some finding significant effects and others presenting marginal effects. The
different – often conflicting – results are due to many factors, the use of different
geographical scales (Crane 2000; Boarnet and Crane 2001) as well as different data and
methodologies (Badoe and Miller 2000). There are only a limited number of cross-section
studies that use consistently measured spatial variables beyond gross density and central
city/suburbs population or employment shares.
This thesis aims to fill the gap in the literature in four main ways. First, it attempts
to define and quantify manifold dimensions of urban spatial structure, which will provide
various spatial variables. Second, it explores the current stage and direction of metropolitan
spatial evolution using estimated spatial descriptors. It also investigates what are the driving
factors behind spatial changes. Third, it probes into the links between metropolitan spatial
structure and commuting economies and economic growth to address the questions on what
are major consequences of the spatial changes. Finally and ultimately, findings of these
seemingly individual inquiries will suggest profound feedback inputs into theories of urban
spatial evolution. I will discuss major findings with reference to urban economic theories
and the perspectives of path dependence in metropolitan spatial development.

4
1.2. RESEARCH QUESTIONS AND APPROACH
Given the stated objectives, I will address specific research questions as below.

1) On the emerging metropolitan spatial structure:
What are the dominant features of the emerging metropolitan spatial structure?
Are contemporary metropolises currently changing into more polycentric or
dispersed structure? Is the metropolitan spatial evolution a linear process or are
there many patterns of spatial developments?
What are the major factors explaining the variation in spatial structures among
US metropolitan areas? Is commuting cost a major determinant as urban
economic theories predict? Is there any trace of path dependent spatial
developments in a cross-section of metropolitan areas?

2) On the impacts of emerging metropolitan spatial structure on commuting:
To what extent does metropolitan spatial structure explain the variation in
average commute times among metropolitan areas, all else being equal? Do the
current spatial changes contribute to decreased or increased commute time?
Is there a more efficient spatial form that is more amenable to metropolitan
growth? Are the relationships between spatial structure and metropolitan
growth dependent on varying metropolitan population size?

Whereas all these research questions are closely interconnected, they cannot be
addressed simultaneously in a single analysis. Instead, I choose to investigate each facet of

5
the interrelated themes of metropolitan spatial evolution. Results of each analysis will be
interpreted with reference to one another from the perspectives of urban economic theories
and path dependent spatial development.
Although panel data models would better fit the purpose of this study involving
spatial changes over time, spatial structure variables can be consistently estimated only for
2000 across metropolitan areas because the Census Transportation Planning Packages
(CTPP) data for measuring spatial descriptors used different geographies not only between
1990 and 2000 but also across metro areas in 1990. Thus, I will conduct most of the
analyses in cross-section frameworks for 79 US metropolitan areas with population over a
half million. An exception is the analysis of spatial trends for the recent two decades in six
selected metropolitan areas, for which I was able to convert all data into one census year
geography.

1.3. ORGANIZATION OF THE THESIS
This thesis comprises a collection of semi-independent chapters, some of which
have been already submitted to academic journals or presented in academic conferences.
Thus, each chapter includes introduction and discussion that partially overlap.
Chapter 2 defines multiple dimensions of urban spatial structure, seeks
methodologies to operationalize and quantify spatial dimensions consistently across
metropolitan areas, and actually estimate spatial structure of 79 metropolitan areas. Results
are consistent with Gordon and Richardson (1996) and Lang (2003) in that one of the most
important features of the modern metropolis is the predominant dispersion of employment.
Estimated spatial descriptors are used in statistical analyses throughout the thesis.

6
Chapter 3 investigates the determinants of various dimensions of urban spatial
structure. Larger metropolitan areas tend to have smaller CBD employment shares and more
decentralized and polycentric structures, while population size is not significantly associated
with employment dispersion. Congestion level is a significant factor only in models for the
number of subcenters, but less significant in employment share models. Rather, industrial
composition is found to be an important spatial determinant.
Chapter 4 explores spatial changes in six selected metropolitan areas over the recent
two decades in order to address the question whether they are increasingly “edgy” or
“edgeless”. Jobs continued to decentralize from the metropolitan core to the suburbs in the
1980s and 1990s and job dispersion was a more common phenomenon than subcentering.
Nevertheless, the results show significant variation in spatial decentralization trends rather
than a uniform linear process from monocentric through polycentric and to dispersed
structure. The geographical and historical contexts of an individual metropolis strongly
affect the path by which it responds to global trends such as the effects of ever decreasing
transportation costs and IT development.
Chapter 5 presents a study on the commuting impacts of metropolitan level spatial
structure. Whereas descriptive analysis identifies a large potential for commute time saving
from spatial restructuring towards more polycentric and dispersed form, particularly in large
metropolitan areas; two step least squares regression analyses find only partial commute
time saving effects of the spatial adjustment. While employment dispersion helps reduce
commute time, polycentricity effect is not significant.
Chapter 6 examines how the links between metropolitan spatial structure and
economic growth depend on the size of the metropolis. Consistent with theories of urban

7
system and evolution, a metropolitan area with more clustered spatial form grows faster
when it is small; whereas more dispersion leads to higher growth rate as it grows large.
Finally, in chapter 7, I will summarize major findings of this thesis and discuss their
implications on planning practice and theorization of urban spatial evolution.

8
CHAPTER 2.
QUANTIFYING URBAN SPATIAL STRUCTURE

2.1. INTRODUCTION
Recent changes in metropolitan spatial structure have been widely recognized by
urban scholars. Beyond the simple decentralization (suburbanization) accounts, polycentric
structure (Giuliano and Small 1991; Forstall and Greene 1997; Fujii and Hartshorn 1995;
McMillen 2003; Bogart and Ferry 1999) and more recently ‘generalized dispersion’ of urban
activities (Gordon and Richardson 1996; Pfister, Freestone, and Murphy 2000; Lang and
LeFurgy 2003) in modern metropolitan areas are increasingly popular topics in urban
studies.
Whereas many of these case studies provided valuable descriptions of the spatial
form in selected metropolitan areas, there is a critical limitation in generalizing the findings
and uncovering spatial variation among metropolitan areas. There are only a limited number
of empirical studies that attempt to explain the regularity in spatial variation among
metropolitan areas and to identify driving factors behind the variation. Further, most
empirical studies examining the impacts of metro level spatial form on metropolitan growth
or commuting efficiency depended on rudimentary spatial variables such as density and
central city/suburban proportion of metro population or employment.
The paucity of cross-section analysis is largely due to the lack of data and spatial
measures that can be consistently applied to multiple metropolitan areas. This chapter

9
defines multiple dimensions of urban spatial structure, seeks methodologies to operationalize
and quantify spatial dimensions consistently across metropolitan areas, and actually estimate
spatial structure of 79 metropolitan areas in the US with population over a half million. The
results will not only present one of the most comprehensive descriptions of the contemporary
metropolitan spatial structure to date, but also provide a set of spatial variables for further
cross-sectional analyses of many aspects of metropolitan development in following chapters.

2.2. MULTIPLE DIMENSIONS IN URBAN SPTIAL STRUCTURE
The conceptual construct of urban spatial structure can be defined as “an abstract or
generalized description of the distribution of phenomena in [urban] geographic space
(Horton and Reynolds 1971).” The urban phenomena include virtually all human activities
in urban space, as we live, work, shop, recreate, and transport. Thus, the term, urban spatial
structure, may be used to denote different spatial processes from one study to another – for
instance, demographic process in Champion (2001) and economic and societal processes in
Dear and Flusty (1998).
In this research, I define urban spatial structure as the distributions of population and
employment in urban space, with an emphasis given on the latter, following the tradition
found in urban economics and land use literature. The rationale for the focus on
employment distribution is that the location of employment concentration acts as the locus of
other urban activities such as commuting flows. Employment distribution has been used as
the reference of even population distribution and land price profile although the exogeneity
of employment location to population distribution is being challenged (Boarnet 1994).

10
How do we operationalize complex and multidimensional urban spatial structure?
Spatial dimensions in urban spatial structure or urban form are highly dependent on
geographical scale. Schwanen’s survey of the literature identifies eight dimensions of urban
form that are closely tied to geographical levels (Schwanen 2003). For instance, the extent
of land use mix, design, and accessibility are of research interest at the neighborhood or
community level while urban size, density, and ‘mono/polycentrism’ are major spatial
variables measured at a city or metropolitan level. The degree of decentralization
(suburbanization) measured by urban/suburban shares of population and employment or the
parameters of density gradient was also often used at the metro level.
A Dutch study has recently classified metropolitan areas (daily urban system) in The
Netherlands into four structure types – centralized, decentralized, cross commuting, and
exchange commuting – based on commuting flows between central city and suburban
municipalities (Van der Laan 1998). While the typology is an innovation in that it measures
spatial relationships between intra-urban locations beyond the simple density measure, the
designation of metros into discrete categories may conceal the spatial variation among
metropolitan areas. The scheme is particularly problematic when metro areas have mixed
characteristics.
Another study of US urbanized areas suggests eight dimensions of urban sprawl-
density, concentration, clustering, centrality, nuclearity, proximity, continuity, and mixed
land uses (Galster et al. 2001). These schemes present one of the most exhaustive lists of
spatial indicators although the conceptual distinction between some of them is vague – e.g.
the ones between concentration and clustering, and proximity and mixed land uses. Because

11
they seek a definition of urban sprawl, the list includes both metro and micro level measures.
A tradeoff is that the index based on the mixture of indicators is less intuitive.
This study seeks more straightforward measures that capture recent metropolitan
level spatial evolution and correspond well to topics of theoretical and empirical literature. I
conceptualize metropolitan spatial structure in two dimensions, centrality and concentration,
as suggested by Anas, Arnott, and Small (1998). As they state, spatial structure in a
metropolitan area can be centralized versus decentralized and it can also be clustered versus
dispersed. Centrality is the extent to which employment is concentrated with reference to
the CBD whereas concentration measures how disproportionately jobs are clustered in a few
locations (Galster et al. 2001).
The two spatial dimensions may be associated, but are distinctive, as empirically
evidenced via factor analysis by Cutsinger et al. (2005). Polycentric urban structure is the
outcome of combined metro-wide decentralization and local level clustering (Anas, Arnott,
and Small 1998). If deconcentration concurs with decentralization, the metropolitan area
would evolve in a more generally dispersed form without significant subcentering.
At the operational level, I measure each spatial dimension based on relative shares
of metropolitan employment by location types, the region level central business district
(CBD), subcenters, and dispersed location. In addition, I also estimate selected indices for
centrality and concentration and extract two composite indices via principal component
analysis. The extracted indices will be compared with measures constructed by employment
shares and complementarily used in the subsequent analyses in following chapters.
In the remaining sections of this chapter, I will seek appropriate procedures by
which employment centers can be consistently identified across metropolitan areas and will

12
introduce spatial indices selected from the literature. Estimation results will follow,
describing current status of spatial evolution and inter-metropolitan spatial variations.

2.3. IDENTIFYING URBAN EMPLOYMENT CENTERS
Given the definition of spatial dimensions in this research, it is a crucial task to
identify urban employment centers including both CBDs and subcenters.
1
Many authors
have proposed and applied various criteria to define urban employment subcenters. While
these criteria included employment size, office and/or retail space, commute flows, job-
housing ratio and land use mix (Cervero 1989; Giuliano and Small 1991), the recent
literature increasingly relies on employment density in defining subcenters.
Primary qualities of urban centers are significantly higher employment density than
the surrounding areas (McDonald 1987) and their influences on density profiles of nearby
locations (Gordon, Richardson, and Wong 1986; Giuliano and Small 1991; McMillen 2001).
Investigators have developed two types of procedures in applying this working definition of
employment centers: absolute and relative density criteria (Giuliano et al. 2005). I will use
both approaches complementarily throughout this research.

1
Various terms are used in the literature to refer to employment centers outside the CBD
such as ‘suburban downtowns’ (Hartshorn and Muller 1989), ‘edge cities’ (Garreau 1991), and
‘technopoles’ (Scott 1990), according to their functions and industrial composition. They presumably
undertake diverse economic roles and spatial functions. Some centers are more specialized in specific
industries while others perform more general functions (Forstall and Greene 1997; Anderson and
Bogart 2001). To denote all these employment concentrations outside the CBD, I use the generic
term ‘employment subcenter’ as in Giuliano and Small (1991).

13
2.3.1. Absolute Criteria: Minimum Density (MD) Procedure
Center identification procedures based on absolute criteria identify candidate zones
(census tracts or TAZs) of employment centers in a straightforward manner by applying a
single minimum density cutoff. Thus, all zones with higher density than the single criteria
become center candidates. Since (Giuliano and Small 1991) applied the density cutoff of 10
jobs per acre to define employment centers in Los Angeles, the minimum density (MD)
procedure has been widely used in empirical research with some revisions: Gordon and
Richardson (1996) set the cutoff at 0.8 standard deviation above the mean trip generation
density; and others included adjacent low density zones as long as the cluster density
exceeds the minimum density (Bogart and Ferry 1999; Anderson and Bogart 2001).
A primary flaw and difficulty of the procedure is in setting the minimum density
cutoff, which is subject to arbitrariness. Researchers used different density criteria in
different cities as shown in Table 2-1 and the density criteria can only be evaluated ex post
with reference to local knowledge. For instance, Cervero and Wu (1998) used a lower
density cutoff for 1980 than for 1990 to identify the same employment centers in both years
in San Francisco Bay Area. Therefore, further guidelines for setting the minimum density
are needed in a comparison study involving cities of various sizes – I set the density cutoff of
each metropolitan area at the level of the ninetieth-percentile density in 2000.
Another drawback of the minimum density procedure is that one cannot take into
account different spatial contexts throughout different parts of a metropolitan area. As will
be seen in chapter 4, physical attributes and spatial organization of employment subcenters
are remarkably different from the traditional CBD. It is impossible to accommodate these

14
sub-metropolitan contexts within the MD procedure without local knowledge in detail. This
suggests the need of a nonparametric procedure.

Table 2-1. Applications of minimum density procedures in previous studies
Authors Cities
Geographical unit of
analysis and data
Minimum density criteria
Giuliano and
Small (1991)
Los Angeles

TAZ; 1980 UTPP 10 jobs per acre
(Gordon and
Richardson
1996)
Los Angeles

TAZ; 1970, 1980
UTPP and 1990 CTPP
Trip generation density:
z-score > 1.8
(Cervero and Wu
1998)
San Francisco Census Tract; 1980
UTPP and 1990 CTPP
1980: 4.5 jobs per acre
1990: 7 jobs per acre
Anderson and
Bogart (2001)
Indianapolis,
Cleveland,
St. Louis, Portland
TAZ; 1990 CTPP 5,000 jobs/mile
2

(Pfister,
Freestone, and
Murphy 2000)
Sydney, Australia TZ (traffic zone);
1981, 1991, and 1996
Journey to work data
from Census
2,470 jobs/km
2

2.3.2. Relative Criteria: Geographically Weighted Regression (GWR) Procedure
The second approach applies relative density criteria, relying upon employment
density functions of various types, parametric and nonparametric. By estimating
employment density surfaces for each metropolitan area, it takes into account different
spatial contexts both within a metropolitan area and across regions. McDonald and Prather
(1994) identified subcenters in Chicago based on significant residuals from an estimated
monocentric density function. Other research on Los Angeles was based on estimated
polycentric models (Gordon, Richardson, and Wong 1986; Small and Song 1994).
More recent developments involve the estimation of nonparametric density
functions. These include spline density curves (Craig and Ng 2001) and geographically

15
weighted regressions (GWR) (McMillen 2001). While both procedures condition subcenters
identification on both the distance and direction from CBD, the latter provides a more
flexible procedure that can be easily applied in many different regions. GWR estimates a
smoothed employment density surface using only nearby observations for any data point
(census tract), with more weights given to closer observations. The first step of McMillen’s
procedure identifies such zones as center candidates that have significantly higher densities
than the estimated surface.
As will be shown in the next section, a procedure using an absolute density cutoff
may fail to identify emerging job concentrations in outlying areas while the GWR procedure
tends to relatively under-bound the CBD in the metropolitan core. Thus, I apply both
procedures to identify center candidates with some modifications. I added a principle of
setting the density cutoff to Giuliano and Small’s (1991) MD procedure: the density cutoff
of each metropolitan area is set at the level of its ninetieth percentile employment density in
2000.
A major modification to McMillen’s (2001) GWR procedure is that I compare two
estimated density surfaces, one with a smaller window and the other with a larger window,
while he identified the differentials (residuals) between actual density and estimated surface
with a large window (50 percent). The bigger the window size, that is the more observations
used for density estimation for each data point, the more smoothed the surface. I identify
such tracts as center candidates whose density estimates by small window GWR (10
neighboring census tracts) is significantly higher than is predicted by large window GWR
(100 census tracts).

16
The small window estimators are preferably used to identify center candidates
instead of actual density on three grounds. First, as a GWR estimator contains density
information of neighboring zones as well as the estimation point, it provides us with
candidates that are closer to the conceptual center definition discussed above. Second, this
procedure is more likely to generate clusters of center candidates, while comparing the
actual density tends to yield fragmented peaks. Finally, these clusters of high density zones
based on small GWR estimators are expected to be more stable over time than the
fragmented peaks.
Only the different statistics due to the use of two GWR surfaces are briefly
explained here because general descriptions of the GWR procedure are provided in
McMillen (2001). The significance of the differential between two density estimators at
each data point is determined at the 10 percent level: 64 . 1 / ) ( ≥ −
∧ ∧ ∧
i
Li Si
y y σ , where
Si
y
∧
is
the small window GWR estimator and
Li
y
∧
is the large window estimator at point i. The
variance of the differential is the sum of variances of two estimators:
2 2 2
Li Si i
∧ ∧ ∧
+ = σ σ σ . The
variance of small window estimator at point 0 is estimated by the equation (2-1) and one for
large window estimator would be obtained by simply replacing the subscript S with L.
) 0 (
2
s
σ is a heteroscedastic error term that is also estimated from another kernel regression
following McMillen (2001).
The dependent variable of GWR estimations is employment density by census tracts
and the independent variables are latitude and longitude coordinates of tract centroids. W(0)
is a diagonal matrix where diagonal elements are a function of each tract’s distance from the
point 0. A tricube weight function is used in constructing the weight matrix.

17

) 0 ( x ] ) 0 ( ' [ ) 0 ( ' ] ) 0 ( ' [ ' x ) var(
2
0
1 2 1
0 0
2
0
s s
S X W X X W X X W X y σ σ
− −
∧ ∧
= = (2-1)

Once significantly dense census tracts are identified by either procedure, I define
clusters of such candidate tracts as employment centers that comprise an employment
threshold, 10,000 jobs. Zones sharing either edge or point are defined as neighboring one
another following the Queen Contiguity Principle. McMillen’s contiguity matrix algorithm
is utilized to save time in the last step of identifying clusters (McMillen 2003).

2.4. SPATIAL INDICES OF CENTRALITY AND CONCENTRATION
The two spatial dimensions are also quantified by a few indices selected from the
literature, three for centrality and two for concentration. Table 2-2 shows equations for the
indices. Modified Wheaton and area-based centralization indices measure how fast
metropolitan employment accumulates along the way from the CBD to the urban edge. The
former is normalized by the distance from the CBD while the latter by land area. Thus, all
census tracts should be sorted by the distance from the CBD in increasing order before
calibration. Both measures range from -1 to 1, with 1 indicating perfect centralization. Two
concentration indices, the Gini Coefficient and the Delta Index, measure how unevenly
metropolitan employment is distributed. All area census tracts are sorted by employment
density in increasing order for calibrating Gini Coefficient.
All these indices, particularly the centrality indices, are sensitive to the presence of
large unpopulated census tracts in outlying areas due to the well known mismatch of
administrative boundaries and functional areas. For instance, the Los Angeles Consolidated

18
Metropolitan Statistical Area (CMSA) might be misrepresented as very centralized if one
includes its huge unpopulated desert tracts (the Mojave Desert) in San Bernardino County.
A compromise involves using a virtual boundary containing 95 percent of metropolitan
population that exclude mostly unpopulated tracts in outlying locations just as Wheaton
(2004) used the 98 percent population area.

Table 2-2. List of centrality and concentration indices
Centrality indices
Modified Wheaton index
(Wheaton 2004)
* ) (
1
1
1
1
DCBD DCBD E DCBD E MWI
n
i
i i i
n
i
i ∑ ∑
=
−
=
−
− =

Area based centralization index
(Massey and Denton 1988)
∑ ∑
=
−
=
−
− =
n
i
i i i
n
i
i
A E A E ACI
1
1
1
1

Weighted average distance from
CBD (Galster et al. 2001)
∑
=
=
n
i
i i
E DCBD e ADC
1
/
Concentration indices
Gini Coefficient (Gordon,
Richardson, and Wong 1986;
Small and Song 1994)
∑∑
==
− −
− =
n
i
n
i
i i i i
A E A E GINI
11
1 1

Delta index (Massey and Denton
1988; Galster et al. 2001)
∑
=
− =
N
i
i i
A
a
E
e
DELTA
1
2
1

e
i
: number of employment at zone i; E
i
: cumulative proportion of employment at zone i;
E: total metropolitan employment; e
i
/E: share of employment at zone i;
a
i
: land area at zone i; A
i
: cumulative proportion of land area at zone i;
A: total metropolitan land area; a
i
/A: share of land area at zone i;
DCBD
i
: the distance of zone i from CBD; DCBD*: metropolitan radius;
n: number of zones.

19
2.5. ESTIMATION RESULTS: DESCRIPTION OF MODERN METROPOLITAN
SPATIAL STRUCTURE

2.5.1. Review of Estimated Spatial Indicators
Summary statistics and correlation coefficients of spatial measures constructed
based on the two urban employment center identification results and indices of centrality and
concentration are presented in tables 2-3 and 2-4, respectively.
Employment centers identification results provide a set of spatial indicators. For
instance, employment share in the CBD/main center presents a proxy of centrality; dispersed
employment share indicates the extent of employment dispersion. The ratio of subcenters’
share to all centered employment share together with the number of centers may present a
proxy of polycentricity.
As expected in Chapter 3, the minimum density approach overstates the size of the
main center in the urban core whereas the boundary of the core center by the GWR
procedure is closer to that of the CBD in the general sense indicated by skyscrapers. For
instance, the GWR procedure strictly defines New York’s CBD as downtown Manhattan
which accounts for about 0.9 million jobs. But, the main center by the MD procedure covers
a much larger agglomeration ranging from the Wall Street financial district to Columbia
University, which contains almost two million jobs (see maps in Chapter 4). On the other
hand, the GWR method tends to identify more subcenters particularly in the largest
metropolitan group while the minimum density approach applying a single density cutoff
often fails in identifying small density peaks in suburban areas.

20
Therefore, metropolitan areas are described as more decentralized, dispersed, and
polycentric on average by the GWR procedure than by the minimum density method.
However, cross-sectional analyses of 79 metropolitan areas (Chapter 3, Chapters 5 to 6) by
either result will not be affected by the difference. Indeed, correlation coefficients between
employment shares by corresponding location type from the two different results are 0.65,
0.68, and 0.67, respectively (Table 2-2).
It is also important to note that there is more variation in subcenters’ employment
share than in CBD/main center’s share. The coefficient of variation in subcenters’ share is
almost twice the coefficient in CBD/main center’s share. Given the historical fact that most
metropolitan areas initially started in a monocentric structure, they maintain a certain portion
of employment agglomeration in the core while the condition for subcenter formation varies
across metropolitan areas. This inter-metropolitan difference will be studied in the next
chapter.
Employment shares in the CBD/main center and dispersed location are significantly
correlated with indices of centrality and concentration, respectively. The close correlation
indicates that the three sets of spatial measures (each based on GWR results, MD results, and
indices) are consistently estimated. I will use all three sets of measures in the analyses
throughout this research while spatial indicators based on employment shares by location
type is more straightforward and easily understandable.

21
Table 2-3. Summary statistics of spatial indicators
Mean Std. dev. Min Max
Coefficient
of variation
(a) (b) (b/a)
GWR # subcenters 4.4 8.0 0.0 53.0 1.81
Method Emp share (%)
CBD (A) 10.7 4.8 2.8 28.2 0.45
Subcenters (B) 7.7 6.5 0.0 28.8 0.84
Dispersed 81.6 6.0 62.9 95.6 0.07
B/(A+B) 38.1 25.5 0.0 91.0 0.67
Minimum # subcenters 4.0 5.7 0.0 36.0 1.43
density Emp share (%)
method Main Center (A) 17.8 6.8 6.3 46.7 0.38
Subcenters (B) 9.4 7.0 0.0 27.1 0.75
Dispersed 72.8 7.8 53.3 89.1 0.11
B/(A+B) 32.8 20.0 0.0 80.5 0.61
Centralization Wheaton index 0.38 0.12 0.12 0.79 0.31
index Area based index 0.63 0.11 0.35 0.86 0.18
Weighted distance 0.31 0.06 0.11 0.44 0.19
Factor score
1)
0.00 1.00 -2.38 3.13
Concentration Gini Coefficient 0.84 0.05 0.74 0.98 0.05
index Delta index 0.70 0.06 0.59 0.93 0.09
Factor score
1)
0.00 1.00 -1.94 3.46
1) Principal component analysis is used to extract a single factor score for each dimension from
multiple index measures. The factor extraction is done separately for each spatial dimension.

22
Table 2-4. Correlation coefficients among spatial indicators
Emp share (GWR) Emp share (MD) Centralization index Concentration index
1) 2) 3) 4) 5) 6) 7) 8) 9) 10) 11) 12) 13)
CBD 1) 1

Subcenters 2) -0.461 1
***
Dispersed 3) -0.305 -0.705 1
*** ***
Main Center 4) 0.652 -0.215 -0.290 1
*** ***
Subcenters 5) -0.333 0.677 -0.461 -0.366 1
*** *** *** ***
Dispersed 6) -0.269 -0.423 0.669 -0.544 -0.582 1
** *** *** *** ***
MWI 7) 0.603 -0.348 -0.108 0.488 -0.167 -0.276 1
*** *** *** **
ACI 8) 0.480 -0.186 -0.184 0.449 -0.032 -0.364 0.777 1
*** *** *** ***
RDC 9) -0.599 0.339 0.115 -0.491 0.158 0.286 -1.000 -0.778 1
*** *** *** ** *** ***
Factor score 10) 0.592 -0.310 -0.141 0.502 -0.128 -0.323 0.978 0.891 -0.978 1
*** *** *** *** *** *** ***
GINI 11) 0.439 0.060 -0.416 0.404 0.213 -0.546 0.707 0.657 -0.709 0.727 1
*** *** *** *** *** *** *** ***
DELTA 12) 0.465 0.006 -0.378 0.406 0.153 -0.493 0.734 0.662 -0.735 0.749 0.981 1
*** *** *** *** *** *** *** *** ***
Factor score 13) 0.454 0.033 -0.399 0.407 0.184 -0.522 0.724 0.663 -0.725 0.742 0.995 0.995 1
*** *** *** *** *** *** *** *** *** ***
Note: MWI: Modified Wheaton index; ACI: Area based centralization index; RDC: Ratio of weighted distance from the CBD to urban radius;
GINI: Gini coefficient; DELTA: Delta index.
** denotes the significance at 5%; *** denotes the significance at 1%.

23
2.5.2. Description of Modern Metropolitan Spatial Structure
Tables 2-5 to 2-7 contain three sets of estimated spatial measures, which are sorted
by dispersed order within each size group. One of the most important features of the modern
metropolis described in the tables is that workplace location is predominantly dispersed.
Lang made a case for edgeless cities – “a form of sprawling office development that does not
have the density or cohesiveness of edge cities” – by emphasizing that they account for twice
the office space of edge cities in thirteen largest metropolitan areas (Lang 2003). The
current research presents far stronger evidence of employment dispersion. Average
dispersed employment share among 79 metropolitan areas is 82% by the GWR method and
73% by the minimum density method. The majority of jobs are diffused outside any type of
employment centers in all 79 metropolitan areas without exception.
The metropolis with the largest dispersed employment share in the largest metro
group is Philadelphia, followed by Boston in the Northeast region. They are highly ranked
also by dispersion indices factor score. While these metros have bigger CBDs than average,
employment outside the CBD is the least clustered in subcenters among the largest metros.
This finding might be counter to popular perception, but coincides with that of Lang’s office
space study. Los Angeles and San Francisco, (and Detroit to a less degree) have a
contrasting spatial form. These two western polycentric metropolises have relatively smaller
CBDs, but large shares of employment clusters in the suburbs.
Contemporary US metropolitan areas are also remarkably decentralized. The CBD’s
average employment share is no more than 11%. The share is just about 18% (less than one
fifth) even when accounting for jobs in high density areas surrounding the CBD (the main
center by the MD procedure). The notion of job decentralization is not new, but the

24
magnitude of it reported here is. Weighted average distance of workplace locations from the
CBD has moved to the point of almost one third of metropolitan radius by 2000 (mean value
of RDC= 0.31).
Las Vegas is the most centralized metropolis of all by all three measures, with 28%
and 47% of total employment in the CBD and main center respectively. Los Angeles has the
smallest CBD that account for only 2.8% of metro employment. In the largest metro group,
New York has the largest CBD employment share (9.9%) followed by Philadelphia and
Seattle. By the minimum density procedure, Houston, Phoenix, New York, and Seattle have
the largest main center employment shares. These four metros are also presented as the least
decentralized of the size group when measured by the decentralization factor score.
To look at the consistency among the rankings by different sets of spatial measures,
I ran Spearman rank-order correlation analysis. All three rankings in each spatial dimension
were significantly correlated (Table 2-8).

25
Table 2-5. Employment shares by location type by the GWR procedure
Metro name Pop Emp No. of Share of emp (%) Sub/ Rank
Sub- CBD Sub- Dis- All Dis- Decen- Poly-
centers centerspersed centers persed tralized centric
A B B/(A+B) B/(A+B)
3 million and plus 17 7.1 15.0 77.9 64.8
Philadelphia 6,188 2,781 68.6 4.586.9 34.3 13 35 49
Boston 5,829 2,974128.08.084.050.133 26 31
Atlanta 4,112 2,088 68.010.781.357.243 25 19
Chicago 9,158 4,248177.011.981.162.945 17 13
Washington 7,608 3,815 167.4 11.8 80.8 61.3 46 20 16
Phoenix 3,252 1,464 97.112.979.964.451 18 12
Dallas 5,222 2,566104.915.879.376.254 4 5
New York 21,200 9,418 33 9.9 11.2 78.8 53.0 57 46 25
Seattle 3,555 1,745 79.311.978.856.058 43 21
Miami 3,876 1,624 67.515.077.566.863 21 10
Detroit 5,456 2,509225.222.272.681.172 5 2
Houston 4,670 2,076148.020.871.272.375 24 8
San Francisco 7,039 3,513 22 5.9 24.2 70.0 80.5 76 10 3
Los Angeles 16,370 6,717 53 2.8 28.8 68.4 91.0 78 1 1
1 to 3 million 2.6 10.8 7.0 82.2 38.3
Orlando 1,645 834 16.51.492.118.1 2 13 57
Cleveland 2,946 1,417 45.97.187.0 54.7 11 11 23
Buffalo 1,170 527 18.94.286.931.812 39 52
Portland 2,265 1,106 28.84.486.833.414 37 50
Milwaukee 1,690 837 38.3 4.986.8 37.0 15 29 46
Oklahoma City 1,083 510 3 6.8 7.1 86.2 51.1 17 14 28
Hartford 1,184 595 013.90.086.1 0.018 60 67
Sacramento 1,797 800 29.0 5.485.6 37.5 22 40 44
Grand Rapids 1,089 562 2 7.3 7.3 85.5 50.0 24 19 32
Minneapolis 2,969 1,627 38.5 6.185.4 41.5 25 34 41
Providence 1,189 507 112.6 2.385.1 15.1 26 58 60
Rochester 1,098 527 112.22.884.9 18.927 54 56
Charlotte 1,499 793 211.54.084.6 25.728 50 53
Cincinnati 1,979 951 38.3 7.784.0 48.0 31 28 35
Greensboro 1,252 618 25.210.884.0 67.4 32 6 9
Pittsburgh 2,359 1,075 114.51.683.9 9.8 34 64 64
Tampa 2,396 1,058 64.212.083.874.135 2 6
Nashville 1,231 659 016.30.083.7 0.036 68 66
Salt Lake City 1,334 661 2 8.4 8.7 82.9 50.9 39 30 30
Indianapolis 1,607 821 214.6 3.581.9 19.3 40 65 55
Raleigh 1,188 652 46.911.581.662.641 15 14
Jacksonville 1,100 506 115.8 2.781.5 14.5 42 67 61
New Orleans 1,338 582 1 16.7 2.0 81.3 10.6 44 72 63
Kansas City 1,776 904 4 8.1 11.4 80.5 58.3 48 27 18
Denver 2,582 1,363 59.510.080.451.249 44 27
St. Louis 2,626 1,250 6 7.7 12.4 80.0 61.6 50 22 15
Norfolk 1,570 659 45.614.879.672.852 8 7
Columbus 1,540 818 312.09.278.8 43.2 56 53 39
Memphis 1,136 525 210.611.078.3 51.059 47 29
Louisville 1,026 523 316.37.876.0 32.4 67 69 51
Austin 1,250 657 121.54.074.615.668 76 59
San Antonio 1,592 681 2 11.8 15.6 72.7 57.0 71 52 20
San Diego 2,814 1,210 10 5.8 22.7 71.6 79.8 73 9 4
Las Vegas 1,563 686 1 28.2 2.3 69.6 7.4 77 79 65

26
Table 2-5. Continued
Metro name Pop Emp No. of Share of emp (%) Sub/ Rank
Sub- CBDSub- Dis- All Dis- Decen- Poly-
centers centers persed centers persed tralized centric
A B B/(A+B)
half to 1 million 0.9 12.2 5.2 82.6 25.7
Allentown 638 270 04.4 0.095.6 0.0 1 3 72
Fort Wayne 502 262 0 8.7 0.0 91.3 0.0 3 36 79
Springfield 602 271 08.8 0.091.2 0.0 4 38 74
Tucson 844 363 09.00.091.00.0 5 4168
Harrisburg 629 332 011.6 0.088.4 0.0 6 51 73
Greenville 962 471 25.9 5.888.349.4 7 12 33
El Paso 680 238 1 5.5 6.5 88.0 53.9 8 7 24
Albuquerque 713 325 012.4 0.087.6 0.0 9 55 70
McAllen 569 171 012.50.087.50.010 5675
Youngstown 595 241 16.9 6.586.6 48.5 16 16 34
Knoxville 687 337 111.23.085.821.2 19 4954
Toledo 618 309 18.45.885.840.620 3142
Wichita 545 269 014.20.085.80.021 6377
Baton Rouge 603 280 1 9.0 5.4 85.6 37.5 23 42 43
Fresno 923 336 18.57.084.545.029 3338
Colorado Springs 517 243 1 9.9 5.6 84.5 36.3 30 45 47
Columbia 537 276 016.30.083.7 0.0 37 7078
Charleston 549 246 016.8 0.083.2 0.0 38 73 76
Syracuse 732 335 019.30.080.7 0.047 7469
Scranton 625 258 18.512.079.658.653 3217
Albany 876 410 213.17.879.137.155 5945
Richmond 997 497 214.1 7.678.335.0 60 62 48
Bakersfield 662 226 021.7 0.078.3 0.0 61 77 71
Little Rock 584 295 2 12.6 9.4 78.0 42.7 62 57 40
Stockton 564 196 211.011.877.251.764 4826
Mobile 540 221 18.014.977.165.265 2311
Dayton 951 461 120.13.876.115.966 7558
Birmingham 921 420 122.8 2.974.3 11.5 69 78 62
Tulsa 803 388114.011.874.245.770 6137
Sarasota 590 243 215.413.371.346.374 6636
Omaha 717 379 316.420.862.955.979 7122
1) Dispersion ranking is ranked by dispersed employment share; decentralization ranking is by
employment share outside the CBD; and polycentricity ranking is by the ratio of subcenters’
share to all center employment.
2) Metropolitan areas sorted by dispersed employment share within each size population group.

27
Table 2-6. Employment shares by location type by the MD procedure
Metro name Pop Emp No. of Share of emp (%) Sub/ Rank
Sub- Main Sub- Dis- All Dis- Decen- Poly-
centers center centers persed centers persed tralized centric
A B B/(A+B) B/(A+B)
3 million and plus 13 17.6 14.3 68.1 42.8
Philadelphia 6,188 2,781 813.4 6.680.1 33.0 16 21 38
Boston 5,829 2,974 618.33.178.614.520 45 62
Chicago 9,158 4,248 1217.75.676.823.927 43 53
New York 21,200 9,418 24 21.2 7.7 71.1 26.7 41 60 52
Miami 3,876 1,624 618.810.870.436.545 4832
Atlanta 4,112 2,088 618.912.668.440.052 50 30
Detroit 5,456 2,509186.325.867.980.558 1 1
Los Angeles 16,370 6,717 36 12.2 21.2 66.6 63.5 61 15 3
Phoenix 3,252 1,464 722.811.965.234.468 65 35
Washington 7,608 3,815 1718.4 16.7 64.9 47.6 71 46 20
San Francisco 7,039 3,513 15 13.8 22.7 63.6 62.2 72 24 6
Houston 4,670 2,076 924.513.062.534.874 70 34
Dallas 5,222 2,566 819.8 20.5 59.7 50.9 76 54 18
Seattle 3,555 1,7451120.521.857.751.577 5617
1 to 3 million 3.1 17.9 9.9 72.2 34.6
Grand Rapids 1,089 562 2 8.3 5.8 85.9 41.3 4 4 29
Buffalo 1,170 527 112.92.284.914.4 7 18 63
Providence 1,189 507 113.9 2.184.0 13.3 8 25 65
Milwaukee 1,690 837 213.2 2.884.0 17.7 9 19 60
Hartford 1,184 595 117.41.780.9 8.913 39 70
Oklahoma City 1,083 510 2 14.3 5.9 79.8 29.2 17 28 45
Cleveland 2,946 1,417 712.5 9.777.843.6 23 17 25
Kansas City 1,776 904 2 15.4 7.3 77.3 32.3 24 30 39
Cincinnati 1,979 951 218.5 4.577.1 19.4 25 47 59
Greensboro 1,252 618 39.0 15.176.0 62.7 28 6 5
Rochester 1,098 527 117.5 6.675.927.3 29 40 51
Charlotte 1,499 793 222.7 2.774.610.7 32 64 67
Pittsburgh 2,359 1,075 321.9 3.574.513.8 33 61 64
Raleigh 1,188 652 510.914.974.157.835 13 9
Sacramento 1,797 800 220.8 6.372.9 23.2 38 58 55
Norfolk 1,570 659 510.817.571.761.940 12 7
Portland 2,265 1,106 419.410.070.633.943 52 36
Salt Lake City 1,334 661 2 25.8 3.6 70.5 12.3 44 71 66
Louisville 1,026 523 221.1 8.670.328.9 47 59 47
Orlando 1,645 834 317.312.570.242.048 38 27
Denver 2,582 1,363 714.1 16.6 69.3 54.0 49 27 15
Nashville 1,231 659 320.810.468.933.2 50 57 37
Jacksonville 1,100 506 119.5 11.868.7 37.8 51 53 31
Columbus 1,540 818 422.0 9.668.430.4 53 62 44
Tampa 2,396 1,058 816.914.868.346.854 35 22
St. Louis 2,626 1,250 4 13.7 18.1 68.2 56.9 55 23 11
Minneapolis 2,969 1,627 617.2 15.167.7 46.8 59 37 21
New Orleans 1,338 582 2 23.3 10.4 66.3 30.9 63 67 43
San Antonio 1,592 681 2 13.3 20.6 66.0 60.8 64 20 8
Memphis 1,136 525 314.919.465.756.6 67 29 12
Indianapolis 1,607 821 620.5 14.565.1 41.4 69 55 28
Austin 1,250 657 324.011.065.031.470 6940
San Diego 2,814 1,210 6 17.5 20.4 62.0 53.8 75 41 16
Las Vegas 1,563 686 0 46.7 0.0 53.3 0.0 79 79 79

28
Table 2-6. Continued
Metro name Pop Emp No. of Share of emp (%) Sub/ Rank
Sub- Main Sub- Dis- All Dis- Decen- Poly-
centers center centers persed centers persed tralized centric
A B B/(A+B)
half to 1 million 0.9 17.8 6.7 75.5 26.4
Allentown 638 270 17.0 4.089.136.5 1 2 33
Harrisburg 629 332 012.3 0.087.7 0.0 2 16 71
El Paso 680 238 0 13.9 0.0 86.1 0.0 3 26 72
Springfield 602 271 110.2 4.185.628.7 5 9 48
Toledo 618 309 110.34.285.529.1 6 1046
Charleston 549 246 016.2 0.083.8 0.0 10 34 73
Youngstown 595 241 17.3 9.783.0 56.9 11 3 10
Fort Wayne 502 262 1 9.8 9.0 81.2 48.0 12 8 19
Fresno 923 336 110.88.680.644.414 1123
Greenville 962 471 213.7 6.180.230.9 15 22 42
Stockton 564 196 111.59.079.643.918 1424
Scranton 625 258 19.211.978.956.519 713
Knoxville 687 337 116.06.177.927.6 21 3350
Colorado Springs 517 243 1 17.8 4.4 77.8 19.8 22 44 58
McAllen 569 171 023.10.076.90.026 6674
Wichita 545 269 119.05.675.422.830 5156
Dayton 951 461 122.62.375.19.231 6369
Sarasota 590 243 117.78.174.331.334 4241
Albany 876 410 218.87.473.828.236 4949
Bakersfield 662 226 026.4 0.073.6 0.0 37 73 75
Mobile 540 221 027.90.072.10.039 7576
Syracuse 732 335 123.35.970.820.142 6857
Tucson 844 363 117.012.770.442.746 3626
Little Rock 584 295 1 26.9 4.9 68.2 15.5 56 74 61
Columbia 537 276 032.00.068.0 0.0 57 7677
Albuquerque 713 325 033.3 0.066.7 0.0 60 78 78
Baton Rouge 603 280 2 8.3 25.3 66.4 75.3 62 5 2
Richmond 997 497 226.1 8.065.923.6 65 72 54
Tulsa 803 388115.418.865.855.066 3114
Birmingham 921 420 133.0 3.963.1 10.7 73 77 68
Omaha 717 379 116.027.156.962.978 324
1) Dispersion ranking is ranked by dispersed employment share; decentralization ranking is by
employment share outside the CBD; and polycentricity ranking is by the ratio of subcenters’
share to all center employment.
2) Metropolitan areas sorted by dispersed employment share within each population size group.

29
Table 2-7. Centrality and concentration indices
Metro name Pop Emp Centrality Indices Concentration indices
MWIACI RDC FactorRank GINI DELTA FactorRank
score Decen- score Dis-
tralized persed
3 million and plus 0.3 0.6 0.3 -0.6 0.8 0.7 -0.4
Chicago 9,158 4,2480.2650.6290.368-0.731 170.759 0.587 -1.838 2
Boston 5,829 2,9740.2680.4960.366-1.098 100.758 0.593 -1.800 3
Atlanta 4,112 2,0880.2590.5860.371-0.895 160.785 0.617 -1.309 9
Philadelphia 6,188 2,7810.3490.5930.326-0.321 330.792 0.636 -1.073 13
New York 21,200 9,418 0.420 0.6570.290 0.298 490.824 0.654 -0.561 23
Miami 3,876 1,6240.1580.5780.422-1.543 60.823 0.681 -0.346 30
Detroit 5,456 2,5090.2540.5800.373-0.941 130.827 0.684 -0.283 32
Phoenix 3,252 1,4640.4000.6580.301 0.177 460.831 0.690 -0.187 34
Dallas 5,222 2,5660.2700.6250.365-0.712 180.848 0.704 0.122 43
Washington 7,608 3,8150.3190.5680.341-0.577 240.854 0.706 0.208 44
Houston 4,670 2,0760.4270.7330.287 0.558 550.856 0.706 0.233 45
San Francisco 7,039 3,513 0.177 0.4500.412-1.787 20.860 0.713 0.324 49
Los Angeles 16,370 6,717 0.347 0.5990.327-0.317 340.862 0.724 0.437 55
Seattle 3,555 1,7450.3790.6800.311 0.113 420.889 0.749 0.949 67
1 to 3 million 0.4 0.6 0.3 0.0 0.8 0.7 0.0
Hartford 1,184 5950.2760.5360.363-0.938 140.744 0.596 -1.938 1
Cleveland 2,946 1,4170.1890.4890.406-1.601 40.759 0.599 -1.737 4
Providence 1,189 5070.3630.5210.321-0.454 270.765 0.599 -1.679 5
Milwaukee 1,690 8370.3040.5810.349-0.637 210.764 0.619 -1.521 6
Charlotte 1,499 7930.3360.5950.334-0.404 300.776 0.622 -1.363 8
Buffalo 1,170 5270.3220.6320.341-0.383 320.783 0.629 -1.220 12
Tampa 2,396 1,0580.2470.4580.378-1.340 90.793 0.648 -0.958 15
Raleigh 1,188 6520.2310.4520.387-1.465 80.802 0.657 -0.783 17
Orlando 1,645 8340.3890.6650.307 0.123 430.810 0.661 -0.663 18
Pittsburgh 2,359 1,0750.3710.5830.315-0.220 360.817 0.657 -0.612 20
Salt Lake City 1,334 661 0.378 0.4210.312-0.645 200.809 0.670 -0.597 22
San Diego 2,814 1,210 0.288 0.502 0.357-0.963 120.821 0.665 -0.506 26
Grand Rapids 1,089 562 0.260 0.3860.373-1.477 70.813 0.681 -0.452 27
Greensboro 1,252 6180.1170.3770.444-2.376 10.822 0.677 -0.394 29
Rochester 1,098 5270.4760.6590.264 0.637 590.830 0.712 -0.009 37
Cincinnati 1,979 9510.3190.6320.341-0.393 310.843 0.697 0.001 38
Norfolk 1,570 6590.3950.5980.304-0.030 380.844 0.699 0.032 40
Minneapolis 2,969 1,6270.4470.7570.277 0.748 610.844 0.704 0.083 41
Kansas City 1,776 904 0.372 0.7190.315 0.175 450.857 0.715 0.310 46
Indianapolis 1,607 8210.4210.7030.291 0.428 540.853 0.720 0.317 47
Louisville 1,026 5230.4020.7400.301 0.413 530.852 0.722 0.318 48
Nashville 1,231 6590.3290.5890.338-0.466 260.855 0.724 0.363 51
St. Louis 2,626 1,250 0.416 0.6680.293 0.301 500.859 0.724 0.402 53
Jacksonville 1,100 5060.5360.7300.234 1.210 710.862 0.720 0.408 54
Columbus 1,540 8180.4150.6900.294 0.356 520.861 0.727 0.455 57
Memphis 1,136 5250.3460.6990.329-0.042 370.865 0.728 0.516 58
San Antonio 1,592 681 0.497 0.7550.253 1.048 670.870 0.749 0.735 63
Sacramento 1,797 8000.4660.6670.268 0.603 570.875 0.746 0.762 64
Oklahoma City 1,083 510 0.498 0.7560.252 1.056 680.882 0.769 1.036 68
Austin 1,250 6570.4690.7310.267 0.802 630.894 0.755 1.059 69
Denver 2,582 1,3630.5080.7700.247 1.157 700.893 0.771 1.170 70
New Orleans 1,338 582 0.517 0.7490.243 1.153 690.908 0.780 1.412 75
Portland 2,265 1,1060.4820.7150.260 0.843 640.909 0.800 1.600 76
Las Vegas 1,563 686 0.787 0.8610.108 3.130 790.979 0.930 3.455 79

30
Table 2-7. Continued.
Centrality Indices Concentration indices
Metro name Pop Emp MWI ACI RDC Factor Rank GINI DELTA Factor Rank
score Decen- score Dis-
tralized persed
half to 1 million 0.4 0.7 0.3 0.2 0.8 0.7 0.2
Greenville 962 4710.2300.3460.387-1.769 30.771 0.621 -1.425 7
Youngstown 595 2410.3500.5560.328-0.441 280.775 0.633 -1.280 10
Dayton 951 4610.3200.5270.342-0.698 190.783 0.624 -1.275 11
Springfield 602 2710.2800.5370.364-0.928 150.799 0.636 -0.988 14
Allentown 638 2700.4160.5760.295 0.025 400.796 0.653 -0.878 16
El Paso 680 238 0.262 0.688 0.372-0.600 230.808 0.667 -0.635 19
Knoxville 687 3370.2610.5500.373-1.006 110.813 0.664 -0.597 21
Toledo 618 3090.4830.6720.262 0.707 600.817 0.664 -0.554 24
McAllen 569 1710.3220.6210.344-0.439 290.813 0.672 -0.544 25
Fort Wayne 502 262 0.344 0.548 0.332-0.502 250.819 0.674 -0.451 28
Baton Rouge 603 280 0.435 0.650 0.286 0.352 510.828 0.678 -0.324 31
Stockton 564 1960.1860.5140.412-1.575 50.831 0.688 -0.198 33
Columbia 537 2760.4830.7530.262 0.940 650.835 0.694 -0.114 35
Richmond 997 4970.3860.6490.309 0.057 410.842 0.685 -0.103 36
Colorado Spring 517 243 0.358 0.7380.325 0.124 440.830 0.716 0.023 39
Mobile 540 2210.3750.6630.318 0.012 390.843 0.707 0.088 42
Little Rock 584 295 0.362 0.5770.322-0.302 350.861 0.714 0.349 50
Birmingham 921 4200.5560.7740.224 1.455 760.858 0.723 0.396 52
Harrisburg 629 3320.4240.6130.291 0.179 470.857 0.731 0.450 56
Charleston 549 2460.4770.7130.266 0.783 620.870 0.733 0.607 59
Syracuse 732 3350.4630.6700.2710.584 560.866 0.742 0.63860
Sarasota 590 2430.3730.7520.318 0.259 480.868 0.745 0.682 61
Fresno 923 3360.5340.7480.236 1.246 720.866 0.753 0.723 62
Scranton 625 2580.3980.3940.304-0.606 220.883 0.748 0.877 65
Wichita 545 2690.5380.7920.234 1.393 740.878 0.762 0.934 66
Albany 876 4100.5140.6850.245 0.945 660.896 0.771 1.214 71
Tucson 844 3630.5160.8300.2441.370730.895 0.778 1.25872
Omaha 717 3790.4300.7480.2870.613580.895 0.782 1.29673
Bakersfield 662 2260.6730.7380.167 2.060 780.897 0.793 1.405 74
Tulsa 803 3880.5360.7990.2331.414750.916 0.803 1.70277
Albuquerque 713 3250.5780.8400.213 1.779 770.926 0.807 1.844 78
Note: Metropolitan areas sorted by the rank of employment dispersion within each population size
group.

31
Table 2-8. Spearman rank-order correlation coefficients

CBD
emp share
rank
Main center
emp share
rank
Centrality
factor score
rank
Dispersed
emp share
rank
(GWR)
Dispersed
emp share
rank
(MD)
Concentration
factor score
rank
CBD 1
emp share
Main center 0.537 1
emp share ***
Centrality 0.565 0.372 1
factor score *** ***
Dispersed 1
emp share (GWR)
Dispersed 0.661 1
emp share (MD) ***
Concentration -0.392 -0.474 1
factor score *** ***

32
Finally, Figures 2-1, 2-2, and 2-3 present a way to overview the variation in
employment distribution among metropolitan areas. In the first two figures, the X-axis is the
job share in all employment centers indicating the degree of concentration while the Y-axis
is the share in the CBD/main center, representing the extent of centrality. Factor scores
extracted from spatial indices for corresponding spatial dimensions are plotted in the Figure
2-3. The bubble size of each metropolitan area is proportionate to its population size.
Thus, the upper right corner of the chart indicates highly centralized and
concentrated spatial structure, which resembles a nineteenth-century monocentric structure.
Las Vegas, NV, that has a unique location and industrial structure, possesses by far the most
centralized spatial structure by all measures.
While employment location is decentralized in most metropolitan areas, the
decentralization takes two spatial forms. It occurred in fairly dispersed form without
significant clustering in some metros while decentralizing jobs re-concentrated into
subcenters in others. Subcentering is most pronounced in Los Angeles, San Francisco and
San Diego in the West, and Detroit in the Midwest. Other Midwestern metros such as
Chicago and Cleveland, and Boston and Philadelphia in the Northeast, are decentralized in
relatively more dispersed forms.

33
Figure 2-1. Centrality versus concentration by the GWR procedure

34
Figure 2-2. Centrality versus concentration by the MD procedure

35
Figure 2-3. Centrality versus concentration measured by indices factor scores

36
CHAPTER 3.
DETERMINANTS OF METROPOLITAN SPATIAL STRUCTURE:
CROSS-SECTION ANALYSIS

3.1. INTRODUCTION
Whereas urban spatial structure has been one of main themes in urban economics,
and numerous theoretical urban models predicting spatial patterns have been formulated in
the literature, empirical studies that directly examine what factors influence spatial
configurations in contemporary cities are surprisingly rare. To the author’s knowledge, just
a few studies have attempted to predict the number of subcenters, which is only one of many
urban spatial dimensions (Erickson 1986; McMillen and Smith 2003). Other research has
attempted to explore the relationships between seven dimensions of land use patterns (sprawl
indices) and other metro level characteristics such as population, size, growth, and
development constraint (Cutsinger et al. 2005). They found moderate relationships only for
three land use factors.
The paucity of empirical research seems to be largely due to the lack of spatial
measures and data. Recent releases of small geographical level (Census Tract) employment
counts from the CTPP data enabled me to estimate rich descriptors of urban spatial structure
in US metropolitan areas in the previous chapter. Utilizing these various spatial variables
and other metropolitan level data, this chapter investigates the determinants of various
dimensions of urban spatial structure.

37
What forces drive urban spatial structure and reconfigure it? A brief discussion of
urban economic theories and some implications from the notion of path dependent urban
development are introduced below. This will guide model specification in section 3.3. A
simple descriptive analysis will precede the statistical analysis.

3.1.1. Economic Theories of Urban Spatial Structure
Theories of urban spatial structure have been a prominent theme in urban economics,
seeking rigorous economic foundations for urban configurations. The tension between
agglomeration economies and diseconomies plays a key role (Richardson 1995).
Agglomeration economies arising from a variety of sources
2
eventually benefit a firm
located in urban agglomerations in terms of cost-efficiency or innovative competitiveness,
and hence, act as a centripetal force in urban space. However, urban agglomeration is not
without costs. In particular, congestion, high land price, and environmental degradation
push firms away from the urban center in the metropolitan context.
Most theoretical urban economic models, from monocentric to non-monocentric,
suggest locational equilibria where these two forces interplay to configure urban spatial
structure. In Mills- and Muth-type monocentric urban models, the trade-off between
increasing returns to scale in production only available in the CBD and decreasing returns to

2
Sources of agglomeration economies discussed in the literature include: internal scale
economies mainly arise from the division of labor and the existence of indivisibilities (Eberts and
McMillen 1999); localization economies, external to a firm but internal to an industry in the same
location, derive from labor market pooling, technological spillovers, the greater intra-industry
specialization and scale economies of industry specific infrastructure (Richardson 1995); and
urbanization economies, external both to a firm and a specific industry but internalized within an
urban area, are generally ascribed to specialized business services and public infrastructure and less
frequently to the ‘law of large numbers’ (Mills 1994). More recently, technological spill over or
innovative process localized within urban clusters is gaining among researchers (Malmberg 1996;
Porter 2000).

38
scale in transportation, shapes land rent and residential density gradients (Mills 1967; Muth
1975). A comparative static analysis of monocentric models provides some implications on
the intercity variation in the extent of population decentralization that population growth,
increased incomes, and the decreased travel costs lead to city expansion and flatter rent and
density gradients (Wheaton 1974; Brueckner 1987). However, these monocentric and even
multicentric urban models (White 1976; Sullivan 1986) provide limited insight into
employment distributions in cities because they assume job concentrations in the CBD or
pre-specified subcenters exogenous to the models.
Some later versions of multicentric and non-monocentric urban models present
propositions that are more relevant to specifying empirical models of urban spatial structure.
First, two dynamic subcenter formation models imply that the timing and location of
subcenters or edge cities formation are dependent on “historic capacity of the CBD”
(agglomeration economies) and transportation economies (commuting costs) (Helsley and
Sullivan 1991; Henderson and Mitra 1996). While strong externalities from the CBD deter
or delay the formation of secondary employment centers, higher congestion externalities
accelerate it. The same effects of the costs of communication with the CBD and commuting
on subcenters formation are also found in other studies (Sasaki and Mun 1996; Fujita, Thisse,
and Zenou 1997). More testable prediction comes from (Fujita and Ogawa 1982) that the
equilibrium number of subcenters increases as population and commuting costs increase and
this prediction was strongly supported by an empirical study (McMillen and Smith 2003).
Finally, two recent urban models explicitly allowing mixed land use provide insight
into urban dispersion. A computable general equilibrium model by Anas and his colleague
shows that dispersed (mixed) employment and residents is a unique equilibrium solution

39
when congestion is the only externality in the absence of agglomeration economies
(shopping externalities in the model; Anas and Kim 1996). Strong shopping externalities
relative to the congestion costs make the dispersion unstable and result in employment
clustering in the model. Similar propositions are formulated by (Wheaton 2004), that lower
agglomeration economies and increased travel demand (greater density) generate greater
employment dispersion.

3.1.2. Path Dependence in Spatial Evolution
Whereas economic urban models provide some valuable predictions about urban
configurations founded on economic behavior of firms and households, equilibrating forces
and outcomes of the models should be constrained by historical path of spatial development
in an individual metropolis. If the spatial evolution is indeed path dependent, then the
explanatory power of many theoretical models will be substantially weakened.
The notion of ‘path dependence’ was developed to explain self-reinforcing dynamics
in the economy and the process of technological adoption and development, which was
initiated by a series of pioneering papers by W. Brian Arthur in the 1980s (Arthur 1994 is
the collection of his pioneering works). It is now increasingly used by social scientists in the
fields outside economics to analyze social and political processes (e.g. persistence of
institutions) (Pierson 2000). The idea, founded on increasing returns, challenges the
deterministic view of traditional neoclassical economic theory based on diminishing returns,
typically implying a unique equilibrium (Arthur 1990).
Increasing returns process is particularly significant in the adoption of modern
technologies in that the more they are adopted, the more they are innovated, and the more

40
incentive to be further adopted (Arthur 1989, p. 116). This positive feedbacks process may
lead to situations that “lock into inferior technology-development paths (Arthur 1990, p.
97).” The famous examples include the defeat of Betamax format by VHS in the VCR
market and the persistent use of QWERTY layout typewriter keyboard (David 1985). An
important claim here is that even small and chance events in earlier periods may have
significant consequences later via the positive feedbacks process.
Arthur suggests that there are four generic sources of self-reinforcing (increasing
returns) mechanisms (Arthur 1988): large set-up or fixed costs, learning effects, coordination
effects, self-reinforcing expectations. Under these conditions, earlier actions in a particular
path, whether by policy decisions or chance events, foster advantages of further movements
along the same path (Pierson 2000). In particular, the exit from the locked in path is very
costly in the presence of large fixed costs or learning effects.
What are the sources of path dependence in spatial development? The first and the
most obvious source is agglomeration economies. Various advantages of proximity to other
firms have been a central theme in spatial economics and were already discussed in detail in
the previous section. What is emphasized here is that the location of economic activities is a
self-reinforcing process. In other words, the location of initial development, even if it is
from a minor “historical accident”, influences spatial developments and forms in later
periods. This historical path dependence was formulated in the context of urban system
(Arthur 1988) and industrial location patterns (Arthur 1990; Krugman 1991), and was
empirically examined with the case of persistent urban primacy in Vienna (Nitsch 2003).
Secondly and more importantly in intra-urban spatial context, the durability of built
environment and infrastructure is the key source of path dependence (Anas, Arnott, and

41
Small 1998, p. 1456; Giuliano et al. 2005, p. 25). It is proposed and empirically tested that
urban decline is much slower than growth due to durable housing (Glaeser and Gyourko
2005). Also in the context of urban spatial structure, once a spatial layout of a city is
established, it persists for a long period given the longevity of urban structures.
Both the durability of structures and agglomeration economies explain why the
location of CBDs, which were firstly built around the central rail stations in most US older
cities, persists even to date when the role of railway is minimized in transporting people and
freight. While each transportation technology era in history has been associated with a
unique urban form (Muller 2004), it leaves long standing marks in the urban landscape.

3.2. DESCRIPTIVE ANALYSIS OF URBAN SPATIAL STRUCTURE

3.2.1. Urban Spatial Structure by Metropolitan Size
The relationships between urban spatial structure and metropolitan population size
are tabulated in tables 3-1 to 3-3. The analysis of variance (ANOVA) F-test results for each
variable are shown at the bottom of all three tables.
First, as theoretical urban models predict, urban policentricity is clearly a function of
population size. Both the number of subcenters and their employment share increase with
population size in both results. Average employment shares in subcenters, defined by the
GWR method are about 15% in the largest metro group while decreasing to 7 % in medium-
size metropolitan areas. The difference between the two groups is a bit smaller in the
minimum density method result (14.3% and 9.9%). Subcenters’ employment share was only
slightly over 5% on average in small metropolitan areas and no subcenter was identified in

42
twelve out of thirty in the smallest size group by the GWR method and in eight by the
minimum density method.
On the contrary, CBDs’ employment shares appear to be a decreasing function of
metropolitan size. The CBD accounts for only 7.1% of total employment on average in the
largest metro group, with shares ranging from 2.8% in Los Angeles to 9.9% in New York.
The CBD’s average share is about 11% in the medium-size group and about 12% in small
metropolitan areas. However, the employment shares in the main center by the minimum
density method are quite stable across metropolitan size groups.
When we assume that the proportion of metropolitan employment in the urban core
represents the degree of decentralization, the GWR results indicate greater decentralization
in large metropolitan areas, while the results by the minimum density method do not support
the relationship. Two out of three centralization indices and factor scores also significantly
vary across different population size groups.
The proportion of dispersed employment is a decreasing function of metropolitan
size. The larger share of clustered jobs in larger metropolitan areas is due to the extensive
presence of suburban employment centers. This is supported by both employment
identification procedures although the pattern is more clearly articulated in the results from
the minimum density method. Concentration indices show the same pattern, but the
difference across metro size groups is not statistically significant.

43
Table 3-1. Center and dispersed employment shares by population size group (GWR method)
Metro No. of Pop Emp No. of Share of emp (%) Sub/
population size Metros (000) (000) Sub- All CBD Sub- Dis- All
centers centers centers persed centers
C=A+B A B B/C
3 million and plus
14 7395 3396 17 22.1 7.1 15.0 77.9 64.8
1 to 3 million
34 1672 809 3 17.8 10.8 7.0 82.2 38.3
half to 1 million
31 686 309 1 17.4 12.2 5.2 82.6 25.7
F-value 5.95 15.46 3.41 15.57
Significance 0.004 <.0001 0.038 <.0001
*** *** ** ***
* significant at 10%; ** significant at 5%; *** significant at 1%.

Table 3-2. Center and dispersed employment shares by population size group (MD method)
Metro No. of Pop Emp No. of Share of emp (%) Sub/
population size Metros (000) (000) Sub- All Main Sub- Dis- All
centers centers center centers persed centers
C=A+B A B B/C
3 million and plus
14 7395 3396 13 31.9 17.6 14.3 68.1 42.8
1 to 3 million
34 1672 809 3 27.8 17.9 9.9 72.2 34.6
half to 1 million
31 686 309 1 24.5 17.8 6.7 75.5 26.4
F-value 0.01 6.63 4.94 3.72
Significance 0.992 0.002 0.010 0.029
*** ** **
* significant at 10%; ** significant at 5%; *** significant at 1%.

Table 3-3. Centrality and concentration indices by metropolitan population size class
Metro No. of Centrality index Concentration index
population size Metros MWI ACI RDC Factor
score
GINI DELTA Factor
score
3 million and plus
14 0.31 0.60 0.35 -0.56 0.83 0.67 -0.37
1 to 3 million
34 0.39 0.63 0.31 0.01 0.84 0.70 0.00
half to 1 million
31 0.41 0.65 0.30 0.24 0.85 0.71 0.17
F-value
4.461.024.023.23 0.96 1.86 1.37
Significance
0.0150.3660.0220.045 0.386 0.163 0.259
** ** **
1) MWI: Modified Wheaton index; ACI: Area based centralization index; RDC: Ratio of weighted
distance from the CBD to urban radius; GINI: Gini coefficient; DELTA: Delta index.
2) * significant at 10%; ** significant at 5%; *** significant at 1%.

44
3.2.2. Urban Spatial Structure by Metropolitan Age
Metropolitan spatial structure has been largely shaped by the development of
transportation and communication technology (Muller 2004). Given the durability of urban
infrastructure, dominant transportation technology and mode of a certain period greatly
affect not only the spatial layout of the time but also the development patterns in later
periods. Thus, the spatial structure of metropolises mainly developed in the post-war
automobile age are expected to be more decentralized and polycentric than metros
established in streetcar or railroad eras in the beginning of the last century.
On the other hand, assuming that a metropolitan region evolves from monocentric to
polycentric or decentralized structure over time, it may require a strong core agglomeration
in the initial stage of development. Thus, it is also possible that younger metros show up as
more centralized. It is an empirical question as to which pattern is more pronounced.
I measured the age of a metropolis as the years since its population first reached half
of the 2000 population. I used historical decennial census from 1880 and estimated
population for the periods between census years by linear interpolation. Then, I divided 79
metropolitan areas into four age classes according to their distribution.
The results by the minimum density method present the significant variation among
different metro age groups. Surprisingly, pre-war metropolitan areas have more dispersed
employment share and concentration indices show the same pattern. Two younger metro
groups have larger employment shares in both the main center and subcenters than metros
established more than 65 years ago. This may contradict a general expectation that post-war
metropolitan areas would be more decentralized and dispersed. It will be examined in the
next section whether the age effects stand out after controlling other variables (regions).

45
Table 3-4. Center and dispersed employment shares by metro age (GWR method)
Metro age No. of Pop Emp No. of Share of emp (%) Sub/
(years) Metros (000) (000)Sub- All CBD Sub- Dis- All
centers centers centers persed centers
C=A+B A B B/C
35 and less 23 1903 873 4 19.1 10.5 8.6 80.9 42.1
35 to 60 32 2105 998 5 19.6 11.1 8.5 80.4 39.7
60 to 100 21 3175 1469 5 15.8 10.2 5.6 84.2 31.7
more than 100 3 1286 581 1 19.1 12.0 7.1 80.9 35.2
F-value 0.22 1.03 1.85 0.68
Significance 0.882 0.383 0.145 0.570

Table 3-5. Center and dispersed employment shares by metro age (MD method)
Metro age No. of Pop Emp No. of Share of emp (%) Sub/
(years) Metros (000) (000) Sub- All Main Sub- Dis- All
centerscenters center centers persed centers
C=A+B A B B/C
35 and less 23 1903 873 3 29.7 20.4 9.2 70.3 30.4
35 to 60 32 2105 998 4 29.5 17.9 11.6 70.5 37.1
60 to 100 21 3175 1469 4 21.6 15.0 6.6 78.4 29.1
more than 100 3 1286 581 2 24.2 16.6 7.6 75.8 32.9
F-value 2.52 2.37 6.69 0.83
Significance 0.064 0.078 0.001 0.480
* * ***

Table 3-6. Centrality and concentration indices by metropolitan age class
Metro age No. of Centrality index Concentration index
(years) Metros MWI ACI RDC Factor GINI DELTA Factor
score score
35 and less 23 0.39 0.66 0.30 0.14 0.85 0.72 0.26
35 to 60 32 0.39 0.64 0.31 0.05 0.85 0.71 0.24
60 to 100 21 0.36 0.60 0.32 -0.24 0.81 0.66 -0.73
more than 100 3 0.43 0.55 0.29 0.04 0.87 0.73 0.49
F-value 0.46 1.61 0.46 0.56 6.27 5.81 6.09
Significance 0.711 0.194 0.712 0.645 0.001 0.001 0.001
*** *** ***
1) MWI: Modified Wheaton index; ACI: Area based centralization index; RDC: Ratio of weighted
distance from the CBD to urban radius; GINI: Gini coefficient; DELTA: Delta index.
2) * significant at 10%; ** significant at 5%; *** significant at 1%.

46
3.3. DETERMINANTS OF URBAN SPATIAL STRUCTURE

3.3.1. Model Specification
In this section, I report the results of various multiple regression analyses that
investigate what factors explain the variation in urban spatial structure among US
metropolitan areas. I build empirical models primarily based upon urban economic theories
and attempt to test effects of other relevant covariates that are not included in formal urban
models.
A basic modeling strategy is to estimate empirical models explaining each of two
metropolitan spatial dimensions – centrality and concentration. As explained in the previous
chapter, centrality measures how close to the urban center, metropolitan employment is
located, while concentration denotes the extent to which jobs are clustered versus evenly
distributed. In empirical models for each dimension, I use multiple spatial indicators as
dependent variables: employment share in the CBD identified by the GWR procedure, main
center’s (by the MD method) employment share, and the factor score abstracted from three
centralization indices are alternately used in centrality models; dispersed employment shares
by the two methods and concentration indices factor score are used in concentration models.
Urban polycentricity is a combination of decentralization and local clustering. It can
also be understood as the relative size and functional strength between CBD and subcenters.
I estimate polycentricity models in addition to the two spatial dimensions since it has gained
popularity in describing spatial structure of modern metropolises. As in the previous
research (Erickson 1986; McMillen and Smith 2003), the number of subcenters identified by

47
two different procedures are studied. In addition, subcenters’ shares of metropolitan
employment and of all center employment are also used as dependent variables.
However, theories barely guide us to distinguish explanatory variables for the three
groups of empirical models. Indeed, the three spatial dimensions may be different faces of
the same underlying spatial phenomenon and the tension between agglomeration economies
and diseconomies is a driving force behind the spatial processes. Thus, I include almost the
same types of regressors in estimating models for three spatial dimensions.
First, metropolitan population size and unit commuting costs are key variables. As
formulated in theoretical urban models (Fujita and Ogawa 1982; Ogawa and Fujita 1989)
and empirically tested (McMillen and Smith 2003), the number of subcenters increases with
population size and per-unit commuting cost. Thus, more polycentric and decentralized
structure is expected in larger and more congested metropolitan areas. Population size and
congestion level are also expected to induce more firms to relocate into residential areas to
economize on commuting costs and hence labor costs, resulting in more dispersed
metropolitan structure. This is suggested by a recent urban model explicitly allowing mixed
land use (Wheaton 2004).
I use 2000 travel time index reported by the Texas Transportation Institute to proxy
commuting costs, which measures the ratio of peak time travel time to travel time in free
flow conditions. This measure is preferred to average commuting time because the latter is
also affected by commuting distance. However, the congestion level is highly correlated
with metropolitan size (coefficient= 0.802) while it is also affected by other factors such as
transportation infrastructure and mass transit use. Further, urban spatial structure not only is
shaped by commuting costs but also influences congestion levels by changing distributions

48
of employment and residents. Thus, a Hausman endogeneity test is done for each regression
model presented in this chapter and two stage regression analyses are also conducted where
the test rejects the exogeneity hypothesis. Details of the Hausman test is explained later.
The availability of the TTI index further reduces sample of 79 metropolitan areas over a half
million population to 66.
There is an offsetting force in employment decentralization and dispersion,
agglomeration economies. Most urban economic models, monocentric or non-monocentric,
are built on the trade-off between agglomeration economies and commuting costs.
Agglomeration economies are benefits of proximity to other firms or clustered location. It is
very important to note that the geographical scope of agglomeration economies has greatly
expanded in the last century as a result of developments in transportation and
communication economies. This point provides the rationale for the hypothesis of
“dispersed metropolis” (Gordon and Richardson 1996).
Unfortunately, it is very difficult to measure agglomeration economies within intra-
metropolitan context. In particular, it is almost impossible to build a proxy variable in a
cross-section analysis. One indirect way of identifying the effects of different levels of
agglomeration economies across metropolitan areas is to control their industrial structure.
Different industries benefit to different extents from various sorts of agglomeration
economies with different geographical scopes (Rosenthal and Strange 2001). For instance, a
city with more corporate headquarters and higher order business services tends to have larger
CBD while small retails and personal services are more likely to diffuse into residential
areas.

49
I ran a principal component analysis to reduce the number of variables presenting
industrial structure. Employment shares by twelve industrial sectors excluding agriculture
and armed forces are used as input data. Five Eigen values that are greater than unit and
together account for about 77 % of the variation are retained and oblique promax rotation is
applied. Rotated factor pattern (Table 3-7) shows salient loadings of industrial sectors on
each factor and enables us to label them: business services, convention and construction,
TWU (transportation, warehousing, and utilities) and wholesale, public sectors, and retail
and personal services.

Table 3-7. Rotated component pattern of five industrial structure factors
Business Convention TWU and Public Retail
and
services and
construction
wholesale sectors personal
services
Information 0.926 -0.167 -0.068 -0.038 0.065
Professional, scientific,
management, administrative,
and waste management services
0.884 0.081 -0.259 0.007 0.025
Finance, insurance, real estate
and rental and leasing
0.532 0.023 0.318 0.122 -0.285
Arts, entertainment, recreation,
accommodation and food
services
-0.261 0.908 -0.190 0.126 -0.162
Construction 0.117 0.730 -0.018 0.047 0.359
Education, health and social
services
-0.445 -0.574 -0.171 0.298 0.146
Transportation and
warehousing, and utilities
-0.072 0.002 0.888 0.144 -0.086
Wholesale trade -0.095 -0.158 0.834 -0.183 0.154
Public administration -0.073 -0.088 -0.094 0.906 -0.067
Manufacturing -0.140 -0.371 -0.188 -0.749 -0.186
Other services (except public
administration)
0.105 -0.169 0.014 0.237 0.812
Retail trade -0.254 0.261 0.028 -0.244 0.669
Initial Eigenvalue 2.841 1.897 1.817 1.609 1.028
Variance explained after
rotation
2.034 1.881 1.682 1.530 1.396
Five industrial factors are extracted by principal component analysis utilizing promax rotation.

50
Another important factor affecting urban spatial structure is path dependence.
Spatial development in a city is path dependent largely due to the durability of transportation
infrastructure and the built capital stock, and the significance of accessibility in urban
activities. Thus, the location of initial development, although it might once have been a
random event, substantially influences spatial development and forms in later stages.
However, it is difficult to find evidence of path dependent spatial development in a
cross-section analysis. A strategy that I adopted here is to identify the effects of the period
of major urban developments on metropolitan structure on the premise that a city developed
in the era of highway and widespread auto use has significantly different spatial forms from
those cities where major developments were already completed during the periods of earlier
transportation technologies. In particular, I estimated the age of each metropolis since its
population passed half of the current population, using historical Census data. Metropolitan
age is included as dummy variables because the relationship may be non-linear.
A second strategy is to include the CBD’s (main center) employment share as an
independent variable in regression models explaining polycentricity and dispersion. It is
treated as exogenous because most metropolitan areas started in a monocentric form. The
strength of the urban core is expected to influence subcenter formation and employment
dispersion in later development phases. An urban economic model of edge cities
development also suggests that edge cities’ location and capacity largely depend on
historical capacity of the CBD (Henderson and Mitra 1996).
In addition to those key explanatory variables, other covariates are also controlled.
A dummy variable indicating coastal location is included because limited land resources and
topography of a coastal region, especially in the presence of a bay, may lead to more

51
polycentric or decentralized form. Higher income is expected to induce more
decentralization. Government structure in a metropolitan region may also influence spatial
structure. Two variables are included, the percentage of metro population in the core central
city and the number of municipalities with population over 10,000 per 100,000 persons.
Finally, dummy variables indicating Census Divisions are controlled.
A basic structure of empirical models of urban spatial structure is as below:
Urban spatial structure = f (P, C, I, A, X, R) (3-1)
, where P represents population size; C congestion; I vector of industrial structure; A
metropolitan age dummies; X vector of control variables; and R regional dummies.

The basic estimation strategy is Ordinary Least Squares (OLS) while models for
number of subcenters are estimated by negative binomial regression. The dependent
variable in number of subcenters models is a count variable, which takes on nonnegative
integer value. The Poisson regression model is widely used to estimate count data, which
assumes that the count variable of interest follows a Poisson distribution:
i i
i
y
i
i
x where y
y
e
y Y prob
i i
' ln ,..., 1 , 0 ,
!
) ( β λ
λ
λ
= = = =
−
(3-2)
However, the Poisson regression model’s strong assumption, that the conditional
variance equals the mean, is often violated. A common alternative in these overdispersion
cases is the negative binomial (NB) regression model, which allows the variance to differ
from the mean by introducing a disturbance into the conditional mean (Greene 2000). Thus,
i i i
x ε β λ + = ' ln , where exp( ε) follows a gamma distribution with mean 1 and variance α.
I ran both Poisson and negative binomial regressions, but present only negative binomial
regression results with the likelihood ratio test for overdispersion. Pseudo R
2
(McFadden’s

52
R
2
) and adjusted McFadden’s R
2
based on likelihood ratio of all parameters model to a
model with just the intercept. Note that this statistic does not reach one.
) ( ln
) ( ln
1 . ,
) ( ln
) ( ln
1
*
2 2
Intercept
Full
McF
Intercept
Full
McF
M L
K M L
R Adj
M L
M L
R
∧
∧
∧
∧
−
− = − =
(3-3)
, where K* is the number of parameters.

Another estimation issue is the possible endogeneity of congestion levels as
discussed above. When this variable is correlated with the residual of the model, the OLS
estimator could be potentially biased and inconsistent. On the other hand, the OLS
estimators are generally more efficient than the instrument variable (IV) estimators when the
congestion level is exogenous. Hausman’s specification test (Hausman 1978) is commonly
used to test for endogeneity of the regressors. Specifically, I use the Durbin-Wu-Hausman
(DWH) test statistic (Baum, Schaffer, and Stillman 2003), which is provided by a statistical
software, STATA.
) ( )' (
~ ~ ∧
−
∧
− − = β β β β V T H (3-4)
, where )] ( ) ( [
~ ∧ ∧ ∧
− = β β V V V ,
~
β denotes a vector of IV estimator and
∧
β denotes a vector
of OLS estimators. H is has a χ
2
distribution with K degrees of freedom.

For each alternative urban spatial structure OLS model, I test the exogeneity of travel time
index using the DWH statistic and 2SLS estimation results are shown where the null
hypothesis of exogeneity is rejected. Instrumental variables for 2SLS estimations include
population growth in the 1990s, freeway lane-miles per population, percentage transit use,
and the number of vehicles per household. For the negative binomial regression models for
the number of subcenters, the endogeneity test is done by including residuals of the first

53
stage OLS in the second stage NB estimation. The significant test of the residual term (z-
stat.) is equivalent to the DWH test (Wooldridge 2002).

Table 3-8. Definition of variables
Variables Descriptions
Urban spatial structure
(Dependent variables)

Centrality
CBD employment share (GWR)
Main center emp. share (MD)
Centralization factor score

Polycentricity (GWR/MD)
Number of subcenters
Subcenters’ emp. share
Subcenters’ share of all center emp.

Concentration
Dispersed emp. share (GWR/MD)
Concentration index factor score

CBD employment / metro employment * 100
Main center employment / metro employment *100
Factor score (principal component analysis) from
three centralization indices

Number of subcenters by two different procedures
Subcenters emp. / metro emp. * 100
Subcenters emp. / (Subcenters + CBD/Main) *100

Dispersed emp. outside centers / metro emp. * 100
Factor score (principal component analysis) from
two concentration indices
Independent variables
Population size

Travel time index
Metropolitan age dummies

Industrial structure

Population in million for poisson regressions and
log population for all least squares regressions
2000 travel time index by the TTI
Age since metro pop reached the half of current pop
size: 35 year and less; 35 to 60 years; more than
100 years; 60 to 100 years (reference group)
Five factor scores extracted from emp. shares by
twelve industrial sectors by principal component
analysis: Business services; Convention and
construction; TWU and wholesale; Public sectors;
Retail and personal services
Income
Coastal location dummy
Percentage core central city
The number of cities per pop.
Region dummies
Median household income in $10,000
Dummy variable indicating coastal location
Pop in core central city / metro pop. * 100
Number of cities with 10,000 plus / 100,000 pop
9 Census Divisions (Reference= Northeast)
Instrument variables for travel time index
Population growth for the 1990s
Freeway
% transit use
Vehicle ownership
(Pop 2000 – pop 1990) / pop 1990
Freeway lane miles / 1,000 pop
Transit commuters / total commuters * 100
Number of vehicles / number of household

54
Table 3-9. Descriptive statistics of variables
Variable Mean Std. Dev. Minimum Maximum
No. of
metros
CBD employment share (GWR) 10.8 5.0 2.8 28.2
Main center emp. share (MD) 18.4 6.7 6.3 46.7
Centralization factor score 0.094 0.974 -1.787 3.130
Number of subcenters (GWR) 5.1 8.6 0.0 53.0
Number of subcenters (MD) 4.6 6.1 0.0 36.0
Subcenters’ emp. share (GWR) 8.2 6.6 0.0 28.8
Subcenters’ emp. share (MD) 9.7 7.1 0.0 27.1
Subcenters’ share of center emp.(GWR) 39.5 25.0 0.0 91.0
Subcenters’ share of center emp.(MD) 32.6 19.0 0.0 80.5
Dispersed emp. share (GWR) 81.0 6.2 62.9 95.6
Dispersed emp. share (MD) 71.9 7.9 53.3 89.1
Concentration index factor score 0.034 1.044 -1.938 3.455
Population size (millions) 2.619 3.459 0.517 21.200
Travel time index
1)
1.23 0.13 1.05 1.76
Metropolitan age 50.9 24.3 13.0 115.0
35 years and less 0.318 0.469 0.000 1.000 21
35 to 60 years 0.394 0.492 0.000 1.000 26
60 to 100 years 0.258 0.441 0.000 1.000 17
More than 100 years 0.030 0.173 0.000 1.000 2
Business services 0.206 0.949 -1.538 2.612
Convention & const. 0.045 1.061 -1.583 5.815
Wholesale & TWU 0.008 1.049 -1.927 3.594
Public sectors 0.103 0.940 -2.642 3.245
Retail & services -0.091 0.905 -2.526 2.633
Median household income ($10,000) 4.412 0.565 3.105 6.202
Coastal location dummy 0.379 0.489 0.000 1.000 25
Percentage core central city 30.6 19.4 3.9 88.8
The number of cities per pop. 0.948 0.442 0.288 2.252
New England 0.061 0.240 0.000 1.000 4
Middle Atlantic 0.106 0.310 0.000 1.000 7
East North Central 0.152 0.361 0.000 1.000 10
West North Central 0.061 0.240 0.000 1.000 4
South Atlantic 0.197 0.401 0.000 1.000 13
East South Central 0.061 0.240 0.000 1.000 4
West South Central 0.136 0.346 0.000 1.000 9
Mountain 0.106 0.310 0.000 1.000 7
Pacific 0.121 0.329 0.000 1.000 8
Population growth for the 1990s 0.166 0.139 -0.016 0.833
Freeway
1)
0.650 0.239 0.110 1.255
Percentage transit use 3.3 3.6 0.5 24.7
Vehicle ownership 1.688 0.113 1.259 1.968
1) Urbanized area data are used for freeway density and travel time index due to the limits in data availability.

55
3.3.2. Determinants of CBD Size and Employment Centralization
Tables 3-10 to 3-12 present Ordinary Least Squares (OLS) estimation results of
spatial structure models with three centrality measures (CBD and Main Center employment
shares, and centralization index factor score) used as dependent variables for each set of
models. OLS2 models of each table are estimated by a stepwise model selection procedure.
Dummy variables indicating metropolitan ages and regions were not used in the same model
as they are highly correlated given US urbanization history – the shift of urban growth from
frostbelt to sunbelt regions.
Durbin-Wu-Hausman (DWH) chi-square tests presented at the bottom of each table
rejected the null hypothesis of an exogenous congestion index at 5 to 10% significance level
in the main center employment share models whereas it was not rejected in other two sets of
models. Thus, Two Stage Least Squares (2SLS) estimation results are presented only for
main center employment share models (Table 3-13).
Population size was consistently significant in all CBD model specifications: as
expected, the CBD employment share decreases with population size. As the population
multiplies around the sample average (2.6 million), the CBD share of metro employment is
expected to reduce by 2.7 to 4 percentage point, all else being equal. But, the TTI
congestion index was not significant in any estimated model. While the models with the two
variables accounted for only 20% of the variation in CBD employment shares, additional
variables improved explanatory powers with the R-squares ranging from 50 to 60%.
Metro level industrial structure was also a significant determinant of CBD
agglomeration: The concentration of public sectors contributes to the CBD’s employment
share; the larger share of retail trade and personal services that tend to diffuse into residential

56
areas decreases the relative size of the CBD. But, coefficients of business services and
wholesale & TWU were not significant. Convention (arts, entertainment, recreation,
accommodation and food services) and construction sectors were also a strongly significant
contributing factor. I suspected that the effects could be exaggerated by an outlier, Las
Vegas, which has an extremely large share of the sector and extraordinarily centralized
employment. But, this variable was still significant and positive in the regression runs that
exclude Las Vegas although the effects are a little diminished.
Metropolitan areas in the youngest group (that doubled their population size in
recent 35 years) have the significantly larger CBD employment share than pre-war
metropolitan areas. The coefficient of the other group (35 to 60 years old) was also negative
but not significant. This presents an indirect evidence of path dependence in urban spatial
structure. However, regional variation was less significant than expected.
Models with centrality factor score (Table 3-12) present similar results with slightly
improved R-squares. The difference is that the coefficient of population share in the core
central city became significantly positive in the two specifications while the coefficient of
retail and services became insignificant.
However, employment share in the main center by the MD procedure was least
explained by the estimated models. R-squares were low and only two industrial structure
variables were consistently significant; neither population size nor congestion level was
significant. 2SLS (Table 3-13) estimations do not change the results much except that the
TTI travel time index became significant with negative sign. This change is partly due to
multicollinearity because estimated travel time index in the first step is highly correlated
with population size.

57
Table 3-10. Determinants of CBD (GWR method) employment share: OLS results
OLS1 OLS2 (stepwise) OLS3 OLS4
log population -2.671 ** -2.501 *** -3.623 ** -4.056 **
(-2.4) (-4.54) (-2.43) (-2.56)
travel time index 0.145 3.364 0.318
(0.02) (0.4) (0.03)
income ($10k) -0.947 -0.879
(-0.47) (-0.46)
coastal location 0.247 1.682
(0.18) (1.16)
% core central city -0.017 -0.038
(-0.53) (-1.05)
business services 0.886 1.575
(0.8) (1.35)
convention & const. 2.300 *** 2.223 *** 2.742 ***
(4.51) (3.5) (3.87)
wholesale & TWU 0.422 0.130
(0.66) (0.2)
public sectors 1.471 *** 1.389 ** 1.634 **
(2.93) (2.11) (2.38)
retail & services -1.221 ** -1.511 * -1.614 *
(-2.19) (-1.81) (-1.96)
35 yrs and less -2.301 * -4.276 **
(-1.96) (-2.14)
35 to 60 yrs -1.952
(-1.15)
more than 100 yrs 1.236
(0.37)
Middle Atlantic 1.237
(0.48)
East North Central -0.120
(-0.05)
West North Central -1.865
(-0.59)
South Atlantic -1.839 -5.386 *
(-1.47) (-1.76)
East South Central 4.297 ** 3.220
(2.17) (0.95)
West South Central -1.138
(-0.36)
Mountain -4.686
(-1.27)
Pacific -1.574
(-0.54)
constant 48.780 *** 47.000 *** 64.547 *** 73.687 ***
(4.75) (5.94) (3.72) (3.96)
R sq. 0.202 0.533 0.497 0.574
Adj. R sq. 0.177 0.477 0.371 0.411
DWH chi-sq test (df=1) 0.653 3.198 * 1.383
1) The dependent variable of all models is employment share in the CBD by the GWR procedure. The number
of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

58
Table 3-11. Determinants of main center (MD method) employment share: OLS results
OLS1 OLS2 (stepwise) OLS3 OLS4
log population -1.781 -0.372 0.028
(-1.08) (-0.18) (0.01)
travel time index 7.629 -4.045 -13.897
(0.71) (-0.35) (-1.06)
income ($10k) -1.075 -1.083
(-0.39) (-0.41)
coastal location -1.520 0.311
(-0.8) (0.15)
% core central city -0.027 -0.078
(-0.62) (-1.57)
business services 1.152 1.466
(0.75) (0.91)
convention & const. 3.762 *** 3.456 *** 3.153 ***
(5.77) (3.9) (3.19)
wholesale & TWU 0.754 0.897
(0.84) (0.98)
public sectors 2.395 *** 2.245 ** 2.183 **
(3.34) (2.44) (2.29)
retail & services -1.411 -1.884
(-1.21) (-1.64)
35 yrs and less -1.526
(-0.55)
35 to 60 yrs -1.994
(-0.85)
more than 100 yrs -0.199
(-0.04)
Middle Atlantic 1.683
(0.47)
East North Central 1.336
(0.39)
West North Central -1.872
(-0.43)
South Atlantic -3.410 * -0.984
(-1.94) (-0.23)
East South Central 4.799
(1.01)
West South Central 3.748
(0.84)
Mountain 6.480
(1.26)
Pacific 3.383
(0.84)
constant 34.520 ** 18.691 *** 35.386 39.337
(2.27) (25.49) (1.46) (1.52)
R sq. 0.019 0.401 0.451 0.536
Adj. R sq. -0.012 0.372 0.313 0.358
DWH chi-sq test (df=1) 0.707 4.631 ** 6.430 **
1) The dependent variable of all models is employment share in the main center by the MD procedure. The
number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

59
Table 3-12. Determinants of centrality factor score: OLS results
OLS1 OLS2 (stepwise) OLS3 OLS4
log population -0.714 *** -0.300 *** -0.328 -0.586 **
(-3.24) (-2.92) (-1.17) (-2.02)
travel time index 2.414 * 2.073 1.852
(1.68) (1.31) (1.08)
income ($10k) -0.300 -0.306
(-0.8) (-0.88)
coastal location -0.360 0.151
(-1.4) (0.57)
% core central city 0.010 ** 0.013 ** 0.004
(2.1) (2.11) (0.56)
business services -0.034 0.051
(-0.16) (0.24)
convention & const. 0.556 *** 0.405 *** 0.382 ***
(5.92) (3.39) (2.95)
wholesale & TWU 0.009 0.105
(0.07) (0.88)
public sectors 0.396 *** 0.360 *** 0.413 ***
(4.23) (2.9) (3.3)
retail & services -0.066 -0.174
(-0.42) (-1.16)
35 yrs and less -0.372 * -0.781 **
(-1.83) (-2.08)
35 to 60 yrs -0.331
(-1.04)
more than 100 yrs -0.005
(-0.01)
Middle Atlantic 0.526 * 1.077 **
(1.81) (2.31)
East North Central 0.403
(0.89)
West North Central 1.028 *
(1.79)
South Atlantic -0.803 *** -0.307
-3.44 (-0.55)
East South Central 0.365
(0.59)
West South Central 0.522
(0.9)
Mountain 0.581
(0.86)
Pacific 0.615
(1.16)
constant 7.330 *** 4.229 *** 3.642 6.882 **
(3.61) (2.78) (1.12) (2.03)
R sq. 0.170 0.586 0.527 0.623
Adj. R sq. 0.144 0.536 0.408 0.478
DWH chi-sq test (df=1) 0.100 0.012 1.426
1) The dependent variable of all models is a factor score abstracted from three centrality indices. The number
of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

60
Table 3-13. Determinants of main center (MD method) employment share: 2 SLS results
2SLS1 2SLS2 2SLS3
log population -2.975 5.603 7.275
(-1.35) (1.32) (1.52)
travel time index 17.316 -48.633 -70.664 **
(1.08) (-1.65) (-2.02)
income ($10k) 1.522 1.563
(0.43) (0.45)
coastal location -2.302 -0.451
(-1.04) (-0.19)
% core central city 0.006 -0.030
(0.12) (-0.46)
business services 0.152 0.661
(0.08) (0.34)
convention & const. 4.478 *** 4.301 ***
(3.83) (3.23)
wholesale & TWU 0.340 0.250
(0.33) (0.22)
public sectors 1.896 * 1.259
(1.79) (1.01)
retail & services -0.020 -0.496
(-0.01) (-0.32)
35 yrs and less 0.794
(0.23)
35 to 60 yrs -0.314
(-0.11)
more than 100 yrs -2.779
(-0.51)
Middle Atlantic -0.653
(-0.15)
East North Central 0.976
(0.24)
West North Central -2.168
(-0.42)
South Atlantic 0.990
(0.19)
East South Central 8.228
(1.39)
West South Central 4.160
(0.79)
Mountain 9.825
(1.54)
Pacific 8.985
(1.57)
constant 39.682 ** -8.433 -8.427
(2.4) (-0.22) (-0.21)
R sq. 0.029 0.409 0.472
Adj. R sq. -0.002 0.261 0.270
DWH chi-sq test (df=1) 0.707 4.631 ** 6.430 **
1) The dependent variable of all models is a factor score abstracted from three centralization indices. The
number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

61
3.3.3. Determinants of Metropolitan Polycentricity
This section examines the determinants of metropolitan polycentricity. Tables 3-14
and 3-15 present the results of negative binomial regressions for the number of subcenters.
Since Hausman tests do not reject the null hypothesis of exogenous congestion level, this
research did not estimate two-stage regressions. The results confirm the findings of
McMillen and Smith (2003) in that the number of subcenters increases with population size
and congestion level. The two variables were significant in all specifications.
Coastal location was also a significant factor contributing to the formation of more
subcenters in both results. The coefficient of central city’s population share was consistently
significant and negative in the minimum density method results. However, adding industrial
structure and metropolitan age variables did not improve the models (as tested by log
likelihood ratio) and the coefficients of those variables were not significant. If any, the
coefficient of business services was significant and positive in the full model of both results,
and that of public sectors was significant and negative only in GWR results.
To examine the path dependence in subcenters formation in a different fashion, I ran
the regressions with CBD/main center’s employment share assuming that it is exogenous
(NB model 6 in both tables). This may not be a strong assumption given that the CBD came
into being and developed in much earlier periods in most cities. The coefficient turned out
strongly significant and negative in both results. This result supports the proposition of
Henderson and Mitra (1996) that “historic capacity of the CBD” deters or delays the
formation of secondary employment centers.
Beyond the number of subcenters, I also examined whether larger and more
congested metropolitan areas (or those with some other characteristics) have

62
disproportionately more employment shares in subcenters. I used two alternative variables
as dependent variables: subcenters’ share of total metro employment and subcenters’ share
of center employment (subcenters / (subcenters + CBD/main center)).
Again, GWR results show higher explanatory powers. Population size had
significant and positive impact on the subcenters’ share of center employment, but was
significant only in a stepwise regression among total employment share models. Higher
congestion levels led to higher subcenters’ share of total employment in two specifications.
This congestion effect was identified in the two stepwise regressions with the MD results. It
may be that population size and congestion level coefficients pick up each other’s effects
and it is difficult to separate them from one another, given that these two variables are highly
correlated (correlation coefficient= 0.802). This multicollinearity problem becomes more
serious in 2 SLS estimations where the instrumented congestion index is more correlated
with the population size.
Metropolitan areas in the group of 35 to 60 years old had a significantly larger share
of subcenters employment share of both total and center employment than pre-war
metropolitan areas in both results. The dummy variable for the youngest metros was
significant only in one model with the GWR results. One significant industrial sector
contributing to larger subcenters’ employment share was retail and personal services.
As in the number of subcenter models, I ran both OLS and 2SLS regressions with
the employment share in the CBD/main center (Table 3-19) as an independent variable. This
variable turned out highly significant and explains about 40% of the variation in subcenter
share of total employment (GWR), together with size and congestion variables. Subcenters
formation is likely to be deterred in a metropolis where the CBD agglomeration is strong.

63
Table 3-14. Determinants of number of subcenters (GWR method): NB results
NB1 NB2 NB3 NB4 NB5 NB6
Beta z Beta z Beta z Beta z Beta z Beta z
population (million) 0.118 3.66*** 0.082 3.58*** 0.076 2.88*** 0.065 2.60*** 0.102 3.59*** 0.108 4.66***
travel time index 4.228 4.50*** 3.085 4.01*** 2.490 2.87*** 2.934 3.53*** 1.803 1.84 * 2.992 3.56***
CBD emp share -0.094-3.92***
income ($10k) 0.386 2.39 ** 0.1930.68 0.3551.30 0.0770.32
coastal location 0.5172.91***0.6053.14***0.5973.18***0.7633.91***
% core central city 0.001 0.16 0.000-0.06 0.0000.05 -0.011-1.82*
# cities per pop 0.113 0.62 0.023 0.11 -0.028-0.13 -0.555 -2.44 **
business services 0.2571.56 0.1620.94 0.3161.97**
convention & const. -0.064-0.57 -0.073-0.63 -0.065-0.53
wholesale & TWU 0.0150.15 0.0630.66 0.0020.02
public sectors -0.171-1.69* -0.131-1.34 -0.160-1.72*
retail & services 0.0010.01 0.0400.31 -0.051-0.38
35 yrs and less 0.0050.01
35 to 60 yrs -0.299-1.17
more than 100 yrs 0.3230.50
Middle Atlantic -0.175-0.43
East North Central 0.9782.51**
West North Central 1.2422.55**
South Atlantic 0.1800.39
East South Central 0.4670.81
West South Central 0.9321.94*
Mountain 0.5020.84
Pacific 0.2900.65
Constant -4.381-3.89***-4.950-4.62***-3.324-2.30**-4.411-3.00***-1.800-1.19 -1.891-1.68 *

alpha (chi) 0.20626.18*** 0.092 4.32 ** 0.0713.24** 0.0360.67 0.0000.00 0.11716.16***
log likelihood -138.5 -132.2 -129.5 -127.8 -119.9 -130.3
likelihood ratio test chi(4) chi(5) chi(3) chi(9) chi(1)
(compared to) BN1 12.79**BN2 5.38 BN3 3.28 BN3 19.13** BN1 16.44***
pseudo R sq. 0.222 0.258 0.273 0.282 0.327 0.268
adj pseudo R sq. 0.200 0.213 0.200 0.193 0.209 0.240
Hausman test (z)
2)
2.557 1.20 -1.141-0.58 -1.758-0.64 -2.642-0.95 -3.039-1.04 2.847 1.39
1) The dependent variable of all models is the number of subcenters the GWR procedure. The number of
observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Hausman test is done by including residuals of the first stage OLS estimation in NB estimation. This
significant test is equivalent to the DWH chi-square test. Instrument variables include % change population,
freeway lane-miles per population, % transit use, and the number of vehicles per household in addition to all
explanatory variables in the second step regressions.

64
Table 3-15. Determinants of number of subcenters (MD method): NB results
NB1 NB2 NB3 NB4 NB5 NB6
Beta z Beta z Beta z Beta z Beta z Beta z
population (million) 0.088 3.55*** 0.074 4.26*** 0.066 3.06*** 0.065 2.80*** 0.068 2.53** 0.087 3.83***
travel time index 3.808 4.79*** 2.416 3.87*** 1.963 2.74*** 2.060 2.68*** 2.057 2.15** 3.691 4.72***
main center share -0.037-2.66***
income ($10k) 0.292 2.33 ** 0.1390.57 0.2080.81 0.0380.16
coastal location 0.3522.27**0.421 2.46 ** 0.450 2.50 ** 0.453 2.39 **
% core central city -0.009 -1.87 * -0.009-1.97** -0.009-1.80 * -0.017-2.84***
# cities/100k pop 0.083 0.55 0.0490.27 0.0530.28 -0.322-1.45
business services 0.2181.49 0.1821.15 0.2761.73*
convention & const. -0.078-0.78 -0.091-0.81 -0.101-0.85
wholesale & TWU 0.0210.22 0.0380.40 0.0130.13
public sectors -0.110-1.24 -0.103-1.12 -0.089-1.00
retail & services 0.0230.20 0.0240.19 -0.014-0.11
35 yrs and less 0.1030.32
35 to 60 yrs -0.037-0.15
more than 100 yrs 0.4740.91
Middle Atlantic 0.5941.42
East North Central 1.2282.87***
West North Central 0.9901.85*
South Atlantic 0.5611.15
East South Central 0.8161.48
West South Central 1.0492.01**
Mountain 0.7121.17
Pacific 0.5851.22
constant -3.736-3.88***-3.253-3.55***-2.027-1.60 -2.514-1.82 * -1.945-1.27 -2.923-2.98***

alpha (chi (1)) 0.12610.99*** 0.025 0.41 0.0200.31 0.0170.21 0.0000.00 0.10710.80***
log likelihood -136.1 -127.4 -125.0 -124.4 -118.1 -132.4
likelihood ratio test chi(4) chi(5) chi(3) chi(9) chi(1)
(compared to) BN1 17.47***BN2 4.79 BN3 1.26 BN3 13.73 BN1 7.56***
pseudo R sq. 0.214 0.265 0.279 0.282 0.318 0.236
adj pseudo R sq. 0.191 0.219 0.204 0.190 0.197 0.207
Hausman test (z)
2)
3.088 1.62 -0.176-0.10 -0.113-0.05 -0.468-0.17 0.2950.10 2.0891.11
1) The dependent variable of all models is the number of subcenters the MD procedure. The number of
observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Hausman test is done by including residuals of the first stage OLS estimation in NB estimation. This
significant test is equivalent to the DWH chi-square test. Instrument variables include % change population,
freeway lane-miles per population, % transit use, and the number of vehicles per household in addition to all
explanatory variables in the second step regressions.

65
Table 3-16. Determinants of subcenters’ (GWR method) employment share: OLS results
Subcenters’ share of total employment Subcenters’ share of center employment
log population 2.084 2.775 ** 3.061 2.933 14.422*** 21.112*** 20.909*** 22.366***
(1.6) (2.02) (1.57) (1.42) (2.95) (7.71) (3.01) (3.02)
travel time index 18.744 ** 18.331 ** 7.491 5.014 27.490 -32.002 -36.815
(2.21) (2.13) (0.68) (0.41) (0.86) (-0.82) (-0.84)
income ($10k) 1.576 1.726 6.609 6.661
(0.6) (0.7) (0.71) (0.75)
coastal location 2.248 2.035 7.559 3.683
(1.24) (1.08) (1.17) (0.55)
% core central city 0.063 * 0.078*0.018 0.180 0.310** 0.101
(1.88) (1.82) (0.35) (1.51) (2.01) (0.54)
# 10k cities/100k pop 1.644 -2.468 10.785 * -6.169
(0.92) (-1.16) (1.69) (-0.81)
business services 0.178 0.049 -0.409 -2.421
(0.12) (0.03) (-0.08) (-0.45)
convention & const. -0.664 -0.521 -3.469 -4.833
(-0.8) (-0.55) (-1.17) (-1.43)
wholesale & TWU 0.018 0.156 -0.326 0.141
(0.02) (0.18) (-0.11) (0.05)
public sectors -0.974 -0.606 -4.206 -3.604
(-1.12) (-0.68) (-1.35) (-1.12)
retail & services 1.747 ** 1.470 1.597 8.730*** 6.267 6.553*
(2.46) (1.35) (1.49) (3.27) (1.61) (1.7)
35 yrs and less 3.794 22.755 **
(1.42) (2.38)
35 to 60 yrs 3.895 * 15.541 *
(1.72) (1.92)
more than 100 yrs 4.587 9.252
(1.04) (0.59)
Middle Atlantic 0.049 -13.088* -0.453
(0.01) (-1.78) (-0.04)
East North Central 5.477 10.733 27.225 **
(1.52) (1.62) (2.1)
West North Central 5.432** 10.749** 18.679** 42.519**
(2.15) (2.42) (2) (2.67)
South Atlantic 4.180 31.278**
(1.04) (2.16)
East South Central 4.272 19.567
(0.95) (1.21)
West South Central 7.123 34.372 **
(1.66) (2.24)
Mountain -3.955* 3.221 29.492
(-1.85) (0.65) (1.66)
Pacific 7.031* 31.657**
(1.78) (2.24)
constant -44.596***-55.647***-59.164**-50.995**-200.248***-268.162***-284.016***-286.798***
(-3.71) (-4.41) (-2.61) (-2.1) (-4.44) (-6.66) (-3.5) (-3.29)
R sq. 0.368 0.511 0.515 0.593 0.376 0.548 0.565 0.633
Adj. R sq. 0.348 0.461 0.382 0.425 0.356 0.502 0.446 0.481
DWH test (df=1) 0.003 0.063 0.943 2.590 2.235 2.657
1) The number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

66
Table 3-17. Determinants of subcenters’ (MD method) employment share: OLS results
Subcenters’ share of total employment Subcenters’ share of center employment
log population 1.230 3.141 2.895 4.666 9.980 9.901
(0.8) (1.35) (1.2)(1.09) (1.51) (1.41)
travel time index 18.741 * 19.787*** 3.069 -1.533 34.064 64.549*** -0.495 -6.854
(1.87) (2.96) (0.23) (-0.11)(1.23)(4.1) (-0.01) (-0.16)
income ($10k) 0.911 2.194 3.255 6.088
(0.29) (0.76) (0.37) (0.72)
coastal location -0.320 0.258 -1.817 -1.944
(-0.15) (0.12) (-0.3) (-0.3)
% core central city 0.048 -0.010 0.098 0.011
(0.93) (-0.16) (0.67) (0.06)
# 10k cities/100k pop 0.165 -4.842* -0.018 -12.015
(0.08) (-1.95) (0) (-1.66)
business services 1.489 0.909 0.417 0.161 -1.423
(1.6) (0.53) (0.24) (0.03) (-0.28)
convention & const. -0.211 -0.174 -1.825 -2.155
(-0.21) (-0.16) (-0.65) (-0.67)
wholesale & TWU -0.326 -0.186 -2.094 -2.130
(-0.32) (-0.18) (-0.73) (-0.72)
public sectors -1.279 -0.899 -4.891 -4.198
(-1.23) (-0.86) (-1.66) (-1.38)
retail & services 1.402 1.356 3.854* 5.209 5.212
(1.08) (1.08) (1.7) (1.41) (1.43)
35 yrs and less 2.429 5.721
(0.76) (0.63)
35 to 60 yrs 3.039 ** 5.638** 13.255*
(2.04) (2.09) (1.73)
more than 100 yrs 2.663 2.916
(0.51) (0.2)
Middle Atlantic 1.276 5.025
(0.33) (0.44)
East North Central 7.619* 21.403 *
(1.8) (1.74)
West North Central 6.317** 15.828*** 19.684** 40.457**
(2.06) (3.04) (2.3) (2.67)
South Atlantic 6.541 20.820
(1.39) (1.52)
East South Central 10.336* 26.427 *
(1.96) (1.72)
West South Central 9.476* 21.129
(1.89) (1.45)
Mountain 4.124 -13.321** 10.560
(0.71) (-2) (0.62)
Pacific 9.961** 24.088*
(2.16) (1.79)
constant -30.877**-16.478**-47.455*-41.618 -75.888* -46.070** -132.131* -133.073
(-2.17) (-2) (-1.75) (-1.47)(-1.93)(-2.38) (-1.72) (-1.61)
R sq. 0.225 0.375 0.394 0.511 0.178 0.298 0.325 0.426
Adj. R sq. 0.200 0.334 0.228 0.309 0.152 0.252 0.139 0.189
DWH test (df=1) 0.113 1.471 4.170** 0.309 3.224* 4.773**
1) The number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

67
Table 3-18. Determinants of subcenters’ (MD method) employment share: 2SLS results
Subcenters’ share of total employment Subcenters’ share of center employment
log population 0.783 -0.593 -3.532 2.619 -5.709 -10.117
0.38 -0.14 (-0.7) 0.46 -0.44 (-0.68)
travel time index 22.368 30.973 48.760 50.682 116.750 149.805
1.50 1.01 (1.32) 1.22 1.29 (1.37)
income ($10k) -0.703 -0.084 -3.528 -1.005
-0.19 (-0.02) -0.33 (-0.09)
coastal location 0.137 0.939 0.106 0.177
0.06 (0.37) 0.02 (0.02)
% core central city 0.026 -0.044 0.006 -0.096
0.45 (-0.61) 0.03 (-0.44)
# 10k cities/100k pop -0.012 -4.120 -0.762 -9.767
-0.01 (-1.45) -0.12 (-1.17)
35 yrs and less 0.921 -0.616
0.25 -0.06
35 to 60 yrs 4.538 8.631
1.50 0.96
more than 100 yrs 4.214 9.437
0.74 0.56
business services 1.525 1.146 2.750 0.849
0.81 (0.56) 0.49 (0.14)
convention & const. -0.853 -1.116 -4.523 -5.088
-0.71 (-0.8) -1.26 (-1.24)
wholesale & TWU -0.051 0.334 -0.940 -0.508
-0.05 (0.28) -0.29 (-0.14)
public sectors -1.073 -0.070 -4.027 -1.615
-0.98 (-0.05) -1.23 (-0.42)
retail & services 0.539 0.142 1.586 1.428
0.34 (0.09) 0.33 (0.3)
Middle Atlantic 3.365 11.531
(0.73) (0.85)
East North Central 7.373 20.636
(1.55) (1.47)
West North Central 15.496 ** 39.422 **
(2.64) (2.28)
South Atlantic 4.613 14.814
(0.84) (0.92)
East South Central 7.033 16.139
(1.11) (0.86)
West South Central 8.759 18.896
(1.55) (1.13)
Mountain 0.768 0.105
(0.11) (0.01)
Pacific 4.652 7.550
(0.74) (0.41)
constant -28.945* -19.902 -0.175 -67.034 -16.359 -3.982
-1.88 -0.51 (0) -1.57 -0.14 (-0.03)
R sq. 0.213 0.381 0.380 0.177 0.303 0.251
Adj. R sq. 0.188 0.212 0.124 0.151 0.112 -0.058
DWH chi-sq test 0.113 1.471 4.170** 0.309 3.224 * 4.773**
1) The number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

68
Table 3-19. Regressions of subcenters’ employment share on CBD/Main Center share
A) OLS results
Dependent variable:

Subcenters share of
total employment
(GWR)
Subcenters share of
center employment
(GWR)
Subcenters share of
total employment
(MD)
Subcenters share of
center employment
(MD)
Beta t Beta t Beta t Beta t
log population 1.158 0.88 8.1211.87* 0.6220.43 1.916 0.55
travel time index 18.794 2.30 ** 27.8311.03 21.3472.26 ** 45.844 2.03**
emp share in the CBD/
Main Center -0.346 -2.44 ** -2.359 -5.01 *** -0.342 -3.09 *** -1.544 -5.84 ***
Constant -27.695-2.05**-85.174-1.90* -19.088-1.38 -22.584 -0.68
R sq. 0.424 0.556 0.328 0.469
Adj. R sq. 0.396 0.534 0.296 0.444

B) 2SLS results
Dependent variable:

Subcenters share of
total employment
(GWR)
Subcenters share of
center employment
(GWR)
Subcenters share of
total employment
(MD)
Subcenters share of
center employment
(MD)
Beta t Beta t Beta t Beta t
log population 1.243 0.73 2.907 0.50 0.1590.08 -0.449 -0.08
travel time index 18.107 1.50 70.1131.71* 21.6761.47 43.912 1.14
emp share in CBD/
Main Center -0.346 -2.44** -2.361-4.92*** -0.267-1.55 -1.471 -3.27***
Constant -28.062-1.96* -62.557-1.29 -16.300-0.94 0.992 0.02
R sq. 0.407 0.552 0.241 0.295
Adj. R sq. 0.378 0.531 0.204 0.260
DWH chi-sq (df=1) 0.006 2.171 0.076 0.124
1) The number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.

69
3.3.4. Determinants of Employment Dispersion
Tables 3-20 to 3-22 present OLS estimation results of employment dispersion
models. 2SLS estimation results are not shown here because none of the models reject the
null hypothesis of exogeneity of travel time index and the results were basically the same.
In the models predicting dispersed employment share, population size and
congestion level were not significant factors. Although coefficient of travel time index was
significant and negative in the model OLS 1 of Tables 3-20 and 3-21, it is more likely to be
correlation rather than causal relationship. That is, less congestion occurs in dispersed
metropolitan areas (as will be shown in Chapter 6) rather than that lower unit travel costs
trigger more employment dispersion (against the theoretical prediction). Indeed, the
coefficient became insignificant when other variables were controlled.
Industrial structure was found significant in explaining the variation in employment
dispersion. Convention (arts, entertainment, recreation, accommodation and food services)
and construction sectors were consistently associated with less dispersion. Business services
variable was also significant in stepwise regressions and public sectors variable was
significant in the MD results. Some of these sectors were associated with the larger
CBD/main center’s employment share and others with the larger subcenters’ share.
One metropolitan age dummy variable was significant in the stepwise regression
with variables generated by the MD procedures: metropolitan areas of age 35 to 60 were less
dispersed than metropolitan areas that reached half of the present population in the pre-war
period. The coefficient of this age dummy was also significant with the same sign in the
regressions with dispersion factor score (negative value of concentration factor score).

70
Table 3-20. Determinants of dispersed employment share (GWR method): OLS results
OLS 1 OLS 2 (stepwise) OLS 3 OLS 4
Beta t Beta t Beta t Beta t
log population 0.588 0.41 0.562 0.25 1.173 0.51
travel time index -18.888 -2.01 ** -10.903 -0.87 -5.675 -0.41
income ($10k) -0.640 -0.21 -0.898 -0.32
coastal location -2.089 -1.47 -2.460 -1.18 -3.727 -1.77 *
% core central city -0.060 -1.22 0.011 0.19
# cities per pop -1.441 -0.70 1.752 0.74
35 yrs and less 0.549 0.18
35 to 60 yrs -1.885 -0.73
more than 100 yrs -5.754 -1.14
business services -2.466 -3.38 *** -1.054 -0.64 -1.645 -0.97
convention & const. -2.995 -3.91 *** -1.556 -1.63 -2.290 -2.16 **
wholesale & TWU -0.458 -0.47 -0.238 -0.25
public sectors -0.401 -0.40 -1.044 -1.04
retail & services 0.033 0.03 0.011 0.01
Middle Atlantic -1.319 -0.35
East North Central -4.800 -1.19
West North Central -8.297 -1.67
South Atlantic 5.685 2.97 *** 1.395 0.31
East South Central -7.208 -1.43
West South Central -5.634 -1.17
Mountain 7.573 2.92 *** 1.875 0.34
Pacific -5.080 -1.15
constant 95.816 7.21 ***80.540 82.61 *** 94.420 3.62 *** 77.885 2.86 ***
R sq. 0.114 0.322 0.268 0.415
Adj. R sq. 0.086 0.265 0.067 0.173
DWH test (df=1) 0.335 0.959 0.013
1) The dependent variable of all models is dispersed employment share by the GWR procedure. The number
of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

71
Table 3-21. Determinants of dispersed employment share (MD method): OLS results
OLS 1 OLS 2 (stepwise) OLS 3 OLS 4
Beta t Beta t Beta t Beta t
log population 0.551 0.31 -2.769 -1.23 -2.829 -1.25
travel time index -26.370 -2.25 ** 1.113 0.09 14.783 1.10
income ($10k) 0.193 0.06 -1.206 -0.44
coastal location 1.741 0.83 -0.588 -0.29
% core central city -0.025 -0.50 0.073 1.27
# cities per pop -0.736 -0.36 3.493 1.51
35 yrs and less -1.092 -0.35
35 to 60 yrs -2.480 -1.68 * -3.807 -1.45
more than 100 yrs -2.657 -0.52
business services -4.005 -5.26 *** -2.088 -1.25 -1.921 -1.16
convention & const. -4.410 -5.50 *** -3.254 -3.39 *** -3.107 -3.01 ***
wholesale & TWU -0.375 -0.38 -0.620 -0.66
public sectors -1.350 -1.80 * -1.007 -1.00 -1.314 -1.34
retail & services 0.031 0.02 0.516 0.44
Middle Atlantic -3.022 -0.83
East North Central -7.903 -2.00 *
West North Central -12.848 -2.64 **
South Atlantic 6.246 3.12 *** -5.200 -1.18
East South Central -14.600 -2.96 ***
West South Central -12.561 -2.68 **
Mountain 4.480 1.59 -9.832 -1.81 *
Pacific -12.634 -2.93 ***
constant 96.357 5.81 ***72.304 67.70 *** 112.626 4.28 *** 103.369 3.89 ***
R sq. 0.153 0.566 0.541 0.657
Adj. R sq. 0.126 0.522 0.415 0.515
DWH test (df=1) 1.118 0.615 0.056
1) The dependent variable of all models is dispersed employment share by the MD procedure. The number of
observations is 66. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

72
Table 3-22. Determinants of dispersion factor score: OLS results
OLS 1 OLS 2 (stepwise) OLS 3 OLS 4
Beta t Beta t Beta t Beta t
log population 0.898 3.86 *** 0.174 1.73 * 0.177 0.58 0.516 1.95 *
travel time index -5.081 -3.36 *** -2.776 -1.62 -1.496 -0.95
income ($10k) 0.179 0.44 -0.033 -0.10
coastal location 0.401 1.42 -0.213 -0.88
% core central city -0.009 -1.40 0.012 1.76 *
# cities per pop -0.090 -0.32 0.582 2.14 **
35 yrs and less 0.033 0.08
35 to 60 yrs -0.546 -3.29 *** -0.601 -1.70 *
more than 100 yrs -0.733 -1.07
business services 0.121 0.54 0.037 0.19
convention & const. -0.605 -7.62 *** -0.393 -3.04 *** -0.403 -3.33 ***
wholesale & TWU 0.117 0.88 -0.086 -0.78
public sectors -0.384 -4.28 *** -0.258 -1.90 * -0.410 -3.58 ***
retail & services -0.226 -2.41 ** -0.114 -0.67 -0.026 -0.19
Middle Atlantic -1.270 -2.98 ***
East North Central -1.506 -3.26 ***
West North Central -2.397 -4.21 ***
South Atlantic 1.197 5.28 *** -0.405 -0.78
East South Central -1.633 -2.83 ***
West South Central -2.039 -3.72 ***
Mountain -1.712 -2.69 ***
Pacific -0.561 -2.14 ** -2.124 -4.20 ***
constant -6.630 -3.09 *** -2.421 -1.69 * 0.529 0.15 -4.803 -1.54
R sq. 0.194 0.659 0.526 0.732
Adj. R sq. 0.168 0.618 0.396 0.621
DWH test (df=1) 0.573 0.048 0.048
1) The dependent variable of all models is the negative value of factor score abstracted from two concentration
indices. The number of observations is 66. * Significant at 10%; ** significant at 5%; *** significant at
1%.
2) DWH denotes Durbin-Wu-Hausman. The null hypothesis of exogeneity of travel time index is tested using
% change population, freeway lane-miles per population, % transit use, and the number of vehicles per
household as additional instrument variables.

73
3.4. DISCUSSION
Utilizing various spatial measures, this chapter presented a comprehensive attempt
to investigate the determinants of manifold dimensions of urban spatial structure with cross-
section data for 66 US metropolitan areas. I specified statistical models to test propositions
or implications from urban economic theories and path dependence perspectives although
the empirical models were not rigorously driven by a formal theoretical model.
Whereas it turned out more intricate than expected to explain any regularity in urban
spatial structure, the results presented some valuable findings that generally conform to
theoretical predictions.
Larger metropolitan areas tend to have smaller CBD employment shares and more
decentralized and polycentric structures. However, population size was not significantly
associated with employment dispersion. Congestion level was a significant contributor to
the subcenter formation, but was less significant in employment share models. It was
difficult to separate out congestion effects from size effects in the forms of empirical models
as used in this research.
I also found that metro level concentration of some industrial sectors were
associated with certain spatial forms: convention (arts, entertainment, recreation,
accommodation and food services) and public sectors contribute to employment
agglomeration in the urban core; business services sector was associated with more
subcenters, and retail & services was associated with larger subcenter employment share;
and these clustering industries were contributing to less employment dispersion.
The path dependence in urban spatial structure was indirectly identified in two ways:
the most recently developed metropolitan areas have smaller CBD and are more

74
decentralized than pre-war metros; metros that reached the half of 2000 population 35 to 60
years ago had more polycentric structure in terms of employment share; metros with a strong
agglomeration in the urban core tended to have less number of subcenters and smaller
subcenter employment share.
These findings support some theoretical predictions. As a city grows, it undergoes
spatial transition to more decentralized and polycentric structure. A strong CBD
agglomeration deters the formation of secondary employment centers. However, empirical
models in this research more or less failed in identifying the role of commuting costs, except
in its effect on the number of subcenters. It is possible that the TTI congestion index may
not be a good proxy of unit commuting costs.
Empirical models presented limited explanatory powers in explaining the variation
in employment dispersion although they found that the dispersion is a phenomenon more of
industrial and regional effects than of population size. Further, the cross-sectional research
design in this chapter cannot provide insights into the driving forces behind the general trend
toward more dispersed metropolitan structure. It is also difficult to examine the key features
of path dependent development, positive feedback (self-reinforcing) mechanism, in a cross-
section analysis. These issues will be partly addressed in Chapter 4, an analysis of spatial
trends in six case metropolitan areas.

75
CHAPTER 4.
TRENDS IN URBAN SPATIAL STRUCTURE

4.1. INTRODUCTION
This chapter
3
analyzes changes in urban spatial structure of selected metropolitan
areas over time. Whereas the previous chapter provided a comprehensive description of
modern metropolitan structure and attempted to explain inter-metropolitan variation, a cross-
section analysis offers limited insights into the direction of spatial trends. It could also
provide only indirect evidence of ‘path dependence’ in spatial evolution. A trend analysis in
this chapter will supplement the drawbacks of the cross-section analysis. A tradeoff is the
reduced sample size that includes only six metropolitan areas: New York, Los Angeles,
Boston, San Francisco, Philadelphia, and Portland. Spatial trends in the last one or two
decades are analyzed.
This chapter begins by introducing two competing perspectives on the spatial
evolution and a brief overview of study areas. The analysis sections, utilizing spatial
measures developed in chapter 2, reveal an overall trend of employment dispersion on the
one hand; and three distinctive patterns of spatial changes on the other hand. The latter point
implies that metropolitan areas have developed unique paths of job dispersion in light of
their histories and circumstances. Discussion of the findings will follow.

3
This chapter is a part of the paper that I presented at the 44
th
annual meeting of the Western
Regional Science Association, February 2005, in San Diego, California. It is also accepted for the
publication in the Journal of Regional Science.

76
4.1.1. Perspectives on Spatial Trends
Suburbanization has been occurring throughout US urban history (Bruegmann 2005).
In recent years, jobs have, for the most part, followed people into the suburbs, although with
a time lag (Glaeser and Kahn 2001). As recently as in 1960, 63 percent of metropolitan jobs
were still concentrated in central cities, while the majority of residents already lived in the
suburbs (Mieszkowski and Mills 1993). However, jobs became almost as decentralized as
the population by the turn of the last century as a consequence of the ‘second wave of
suburbanization’ (Stanback 1991; Glaeser and Kahn 2001; Wheaton 2004). Accordingly,
monocentric urban models lost much of their explanatory power. Accessibility to the urban
center no longer does a good job in explaining the distributions of population and
employment (Gordon, Richardson, and Wong 1986; Small and Song 1994; McMillen and
McDonald 1998); nor does it explain land value profiles (Heikkila et al. 1989).
Whereas the decentralization of people and jobs is widely recognized, much less is
known with regard to the nature of emerging urban spatial structure. Do modern
metropolises become polycentric or generally dispersed? Do monocentric, polycentric and
dispersed forms describe stages of a sequential spatial evolution? Chapter 2 found the
prevalence of dispersed employment. This chapter addresses more research questions on the
direction of spatial changes.
There are two competing or complementary perspectives with regard to the spatial
change. The first and dominant view holds that the modern metropolis is increasingly
characterized by the presence of multiple nodes. It emphasizes that there are many
concentrations of employment and commercial activities outside the traditional CBD. The
growth of multiple subcenters reorganizes urban fabrics, land use patterns and commuting

77
flows, which used to be oriented towards the CBD in a monocentric urban place (Fujii and
Hartshorn 1995).
The polycentric structure of US metropolitan areas has been a popular subject in the
literature. Beyond the archetypal polycentric regime Los Angeles (Gordon, Richardson, and
Wong 1986; Giuliano and Small 1991; Forstall and Greene 1997), multiple subcenters are
identified in many of the largest metropolitan areas such as Chicago (McMillen and
McDonald 1998; McMillen 2003), San Francisco (Cervero and Wu 1998), Dallas-Fort
Worth (Waddell and Shukla 1993), Atlanta (Fujii and Hartshorn 1995), and Houston (Craig
and Ng 2001). The list has recently been expanded to include medium-size metropolises
such as Cleveland, Indianapolis, Portland, and St. Louis (Bogart and Ferry 1999; Anderson
and Bogart 2001).
As discussed in the previous chapter, it is the ‘tension between agglomeration
economies and diseconomies’ that plays a key role in the transition from monocentric to
polycentric urban structure (Richardson 1995). A firm, by locating in suburban centers,
benefits from agglomeration economies that used to be available within the centers, while
mitigating diseconomies such as congestion and high land prices that the older employment
centers suffer from (Richardson 1995). In other words, polycentric evolution is one way that
a metropolis manages to facilitate growth, overcoming the negative externalities that go with
size.
The ‘generalized dispersion’ of jobs over clustering, however, would be more of a
norm if the benefits from locating in job centers diminish (Gordon and Richardson 1996).
The same change will be expected if even subcenter location becomes too costly as in the
CBD (Fulton 1996). The second perspective emphasizes this change towards dispersion.

78
Gordon and Richardson (1996) found that the share of total employment in ‘activity centers’
was not only far less than what could be described as polycentric but also had dramatically
decreased from twenty to twelve percent over two decades by 1990 in Los Angeles
metropolitan area. They hypothesized that agglomeration economies are increasingly
ubiquitous throughout the metropolitan region by virtue of enhanced automobile access.
Several recent studies present empirical evidence consistent with this view. A
similar study of Sydney, Australia reports the dispersion trend for the 1980s, but a moderate
reversal in the early 1990s (Pfister, Freestone, and Murphy 2000). Glaeser and Kahn (2001)
also document that suburban jobs (of more recent development) are much more diffused than
central city jobs in average US metropolitan areas. A more recent study on office space
distribution found that “edgeless cities, a form of sprawling office development that does not
have the density or cohesiveness of edge cities,” account for two-thirds of non-downtown
office in thirteen largest metropolitan areas (Lang 2003). It also found that more office
space than that in downtown is dispersed throughout metropolitan areas except in New York
and Chicago.
Will metropolitan structure become more clustered or dispersed? Clark and
Kuijpers-Linde (1994) position the urban form debate within the context of competing views
on technological development, ‘deconcentration’ and ‘restructuring’ schools (for a survey of
the literature, see Audirac 2002). From the deconcentration perspective, declining costs of
transporting goods, people, and information due to technological development are primary
factors shaping the metropolitan landscape (Cairncross 1997; Glaeser and Kahn 2003;
Glaeser and Kohlhase 2004). In particular, it emphasizes substitutability over
complementarity between transportation and communication technology. Therefore, the

79
development of modern (especially information) technology in this view ultimately
contributes to enhancing the mobility of households and firms, implying far greater
dispersion of urban activities.
The restructuring perspective, on the other hand, pays more attention to
organizational changes and economic restructuring entailed by information technology (IT)
development. From this view, IT development confers on a firm more organizational and
location flexibility than ever, which leads to the decentralization of production and routine
functions but also reconcentration of higher order activities at the same time (Castells 1989;
Sassen 1991). Some authors emphasize that the suburbs of large metropolitan areas are
being transformed into the home of high technology clusters and nodes in international
information flows and economic networks (Scott 1988; Muller 1997; Freestone and Murphy
1998). The spatial implication is a polycentric structure.
Again, future changes in metropolitan structure crucially depend on how
technological advances will change the nature and geographical scope of agglomeration
economies. To the extent that metropolitan-wide transportation and communication
infrastructure improvements enhance individual mobility or accessibility, location in
employment centers will become much less attractive given the congestion and other
diseconomies of concentration.
On the other hand, a growing body of economic geography literature holds that
proximity still matters. In light of emerging knowledge-based economies, density is
believed to foster localized learning and innovation due to tacit nature of knowledge
(Malmberg and Sölvell 1997; Maskell and Malmberg 1999). In particular, the significance
of formal and informal face-to-face contacts in creation or exchange of ideas that cannot be

80
simply transmitted (Storper and Venables 2004) implies that some sort of agglomeration
economies will still work within a fairly short spatial range, resulting in continued clustering.
It is the purpose of this chapter is to address the question on which trend has
dominated in recent decades. One complicating factor is the path dependence of spatial
development in urban places, as discussed in chapter 3. While the effects of technological
development will show up as a secular trend in the analysis, these changes should be
contingent on the historical path of spatial development that each metropolitan area has
taken in the past. Thus, a general spatial trend might manifest itself in different forms. If
spatial development is indeed an increasing returns process, we could expect to identify
some self-reinforcing inclination in spatial evolution.

4.1.2. Overview of Study Areas
Six metropolitan areas are selected for the trend analysis in this chapter while most
other analyses in this research use a cross-section of 79 metropolitan areas. The number of
study areas is limited by the consistency of geographical units employed by each year’s
CTPP (Census Transportation Planning Packages, Urban Transportation Planning Package
in 1980) data. The CTPP data series, one of very few sources of employment data by place
of work for small geographical units, employed different geographical systems for different
years. Thus, I include in the analysis only those metro areas whose commute data can be
converted onto 1990 census tracts: spatial changes in New York, Los Angeles, Boston, and
Portland are analyzed for the 1990s; while the changes can be traced for the recent two
decades in San Francisco and Philadelphia.

81
Census tract relationship files from the US Census Bureau are utilized for converting
2000 data to 1990 geography. I converted 1980 data based on correspondence tables
obtained from metropolitan planning organizations (MPOs) of San Francisco and
Philadelphia CMSA. Study area boundaries may be slightly smaller than official CMSA or
MSA definition because they are delineated to include zones that are covered by the data of
all the years. The sample of six metropolitan areas is well balanced in terms of population
size and geography, given the constraints, except that metros in the South are missing.
The basic descriptors of the study areas are shown in Table 4-1. These six places
vary in population (in 2000) by a factor of 9.3, they vary in ten-year population growth by a
factor of 6.4, yet they vary in drive-alone one-way commuting time (in 2000) by a factor of
only 1.2. They seem to have developed unique patterns of job dispersion, in light of their
histories and circumstances that limit the growth of mean travel times.
There are two minor revisions from the employment center identification procedures
introduced in Chapter 2. First, while I keep the same level of employment threshold at ten-
thousand jobs for New York and Los Angeles, I lowered it to seven-thousand jobs for San
Francisco, Philadelphia, Boston, and Portland. Second, those centers that were contiguous to
the CBD or main center in any census year are considered as part of the CBD/main center
also in other census years. Thus, there can be multiple clusters within the CBD/main center
category in Tables 4-3 to 4-8. For these reasons, employment shares by location type for
2000 in this Chapter may be slightly different from those reported in Chapter 2 and used in
the other Chapters.

82
Table 4-1. Characteristics of six metropolitan areas
New York Los Angeles Boston Portland
1990 2000 90-00 1990 2000 90-00 1990 2000 90-00 1990 2000 90-00
Metropolitan area
Employment (000)
1)
9,039 9,4094.1% 6,7516,717-0.5%2,189 2,311 5.6% 704 1,106 57.1%
REIS Emp. (000)
1)
9,668 10,2686.2%6,9447,3145.3% 2,913 3,271 12.3% 771 1,028 33.3%
Population (000) 19,502 21,134 8.4% 14,521 16,370 12.7% 4,056 4,307 6.2% 1,793 2,265 26.3%
Mean commute (min)
2)
30.8 34.411.7% 26.4 29.0 9.8% 25.4 28.4 11.8% 22.0 24.5 11.4%
by drive alone 25.3 28.5 12.6% 25.6 27.8 8.6% 24.3 27.1 11.5% 21.0 23.2 10.5%
Area (sq. mile) 9,841 9,841 33,822 33,822 2,743 2,743 6,950 6,950
Number of zones 5,053 5,053 2,546 2,546 867 867 395 395
95% population area
3)

Employment (000) 8,623 8,953 6,552 6,444 2,130 2,247 697 1,073
Population (000) 18,503 20,074 13,899 15,549 3,863 4,090 1,698 2,146
Area (sq. mile) 8,099 8,099 9,003 9,003 2,335 2,335 5,426 5,426
Number of zones 4,810 4,810 2,467 2,467 833 833 379 379
Urban radius (mile) 67.3 67.3 69.2 69.2 32.8 32.8 46.4 46.4
Mean zone size (acre) 1,078 1,078 2,336 2,336 1,794 1,794 9,163 9,163
Region-wide density 1.7 1.7 1.1 1.1 1.4 1.5 0.2 0.3
Mean density 18.4 18.3 6.2 5.4 7.7 8.1 4.3 4.9
Median density 4.1 4.3 3.1 2.9 2.3 2.4 1.2 1.7
90 percentile density 26.8 25.8 12.3 10.7 16.9 15.2 7.4 8.4
CBD peak density 1,668.5 1,577.5 170.8 190.5 350.8 339.6 250.5 274.5
San Francisco Philadelphia
1980 1990 2000 90-00 198019902000 90-00
Metropolitan area
Employment (000)
1)
2,316 3,0513,40511.6% 1,902 2,300 2,440 6.1%
REIS Emp. (000)
1)
3,3433,91917.2% 2,7002,927 8.4%
Population (000) 5,027 6,014 6,784 12.8% 4,8815,1745,387 4.1%
Mean commute (min)
2)
26.230.516.4% 24.527.7 13.1%
by drive alone 24.2 28.4 17.4% 23.1 26.1 13.0%
Area (sq. mile) 6,922 6,922 6,922 3,743 3,743 3,743
Number of zones 1,284 1,284 1,284 1,308 1,308 1,308
95% population area
3)

Employment (000) 2,279 2,956 3,272 1,8532,1902,313
Population (000) 4,836 5,7356,443 4,6934,9465,116
Area (sq. mile) 4,316 4,316 4,316 2,833 2,833 2,833
Number of zones 1,216 1,216 1,216 1,243 1,243 1,243
Urban radius (mile) 49.6 49.649.6 34.434.434.4
Mean zone size (acre) 2,272 2,272 2,272 1,458 1,458 1,458
Region-wide density 0.8 1.1 1.2 1.01.2 1.3
Mean density 7.2 7.8 8.2 6.46.4 5.5
Median density 1.6 2.3 2.3 1.62.1 2.1
90 percentile density 11.9 13.913.6 11.110.4 8.2
CBD peak density 650.9 584.4 686.0 446.0 576.6 590.8
1) Total employment excludes armed forces. The wage, salary, and employment data from the
Regional Economic Information System (REIS) are also presented for comparison. Although the
CTPP data generally underestimate the number of employment, this does not seem to affect the
analysis of spatial distribution of employment.
2) Mean commute times are calibrated by workplace. Thus, they may be slightly longer than
residence-based figures, depending on how many workers a metropolis draws from the outside.
3) The area that houses 95% of the total metropolitan population excludes mostly unpopulated tracts
in outlying locations. All centralization and concentration indices are measured for this area.

83
4.2. TRENDS IN URBAN SPATIAL STRUCTURE

4.2.1. Spatial Changes towards Dispersion and Decentralization
Figure 4-1 presents one of the simplest ways to identify employment dispersion
trends. All census tracts within the 95 percent population area are grouped into five quintiles
by employment density and the densest quintile is further split into two deciles. Then, each
density group’s share of total employment in each year is presented in the bar charts.
Apparently employment deconcentration occurred in all six metropolitan areas.
They became more dispersed in the later period than in any earlier period, with increased job
shares in low density tracts and decreased shares in higher density zones. The most
significant employment gains were at the bottom two quintiles, the lowest density zones.
Employment shifts from higher to lower density areas have long been observed for the last
half century in both inter- and intra-metropolitan contexts (Carlino and Chatterjee 2002).
They attribute this postwar urban development to congestion costs of density in their
equilibrium model.
Nevertheless, there are substantial variations among regions in terms of the extent
and speed of deconcentration. It is surprising that employment concentration is higher in
Los Angeles and San Francisco than in the other metropolitan areas, with the majority of
jobs concentrated in the densest quintile (two densest deciles). These two western
metropolises also experienced less dispersion during the periods studied. It will be shown in
the next section that the slow dispersion and high degree of concentration in the two metros
are due to jobs clustering in the suburbs. Jobs dispersion was much faster in Philadelphia
and Portland where subcentering was less significant.

84
The same trend is also identified by the changes of concentration and centrality
indices presented in Table 4-2. Overall, employment is more centralized and concentrated
than population in all metropolitan areas at any point of time. Yet, jobs are decentralizing
and diffusing much faster than population. People moved further out to less congested areas
in all cases but Portland in the 1990s. There was no exception, however, in the overall trend
towards more decentralized and dispersed employment distribution.
Jobs were diffusing in Portland and Philadelphia faster than in the other
metropolitan areas. All indices for Philadelphia changed by more than ten percent in the
1990s, which is an accelerated continuation of the same patterns in the 1980s. Portland also
underwent fast employment dispersion while experiencing explosive metropolitan job
growth in the 1990s, by 57 percent over the decade. It is notable that little dispersion
occurred in residential patterns for the same period. Perhaps, all planning and policy
schemes to promote compact urban development may have been effective in containing
residential development, but not in checking workplace dispersion (Ozawa 2004).
Similar trends of decentralization and deconcentration, but to a significantly less
extent, are found in New York and Boston. However, spatial transformation in Los Angeles
and San Francisco is distinctive in that deconcentration occurred in much slower fashion
than decentralization. It implies that a significant proportion of decentralizing jobs have
reconcentrated in suburban clusters in the two western metropolises. This result is
confirmed via the analyses of employment centers in the next section.

85
Figure 4-1. Changes in employment shares by density quintile (decile)

Decile 1
Decile 2
Quintile 2
Quintile 3
Quintile 4
Quintile 5
6.9
10.4
8.8
8.9
13.0
11.2
18.5
18.9
15.5
14.2
37.4
36.4
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Los Angeles 1990 Los Angeles 2000
Employment Share
6.0
10.7
13.8
15.2
18.2
16.8
18.9
17.2
8.7
7.7
34.3
32.4
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
New York 1990 New York 2000
Employment Share
3.6
5.4
7.9
6.5
9.5
8.7
12.7
12.6
14.0
18.2
18.3
17.1
19.8
21.0 16.3
39.1
33.2
36.1
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
San Francisco
1980
San Francisco
1 990
San Francisco
2000
Employment Share
2.4
5.1
10.8
10.1
10.9
12.5
14.9
18.4
17.5
22.4
19.5
21.1
13.8
13.5
11.0
36.3
32.5
27.1
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Philadelphia
1980
Philadelphia
1 990
Philadelphia
2000
Employment Share
Decile 1
Decile 2
Quintile 2
Quintile 3
Quintile 4
Quintile 5
Decile 1
Decile 2
Quintile 2
Quintile 3
Quintile 4
Quintile 5
9.5
13.2
15.0
13.2
17.8
17.7
18.9
18.2
12.0
11.2
26.8 26.6
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Boston 1990 Boston 2000
Employment Share
2.3
10.3
9.8
12.0
13.3
10.5
20.7
19.8
15.4
17.5
38.6
29.9
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Portland 1990 Portland 2000
Employment Share

86
Table 4-2. Changes in centralization and concentration indices
New York Los Angeles Boston Portland
%Ch. %Ch. %Ch. %Ch.
1990 2000 90s 1990 2000 90s 1990 2000 90s 1990 2000 90s
Centralization
MWI Emp 0.46 0.43 -7.8 0.39 0.35 -11.5 0.25 0.23 -8.9 0.57 0.49 -15.1
MWI Pop 0.37 0.37 -0.7 0.33 0.3 -8.9 0.13 0.12 -12.3 0.39 0.38 -2.7
ACI Emp 0.69 0.65 -5.6 0.64 0.6 -6.5 0.53 0.5 -5.2 0.76 0.72 -5.9
ACI Pop 0.61 0.61 -0.4 0.58 0.55 -4.7 0.43 0.42 -4.2 0.66 0.66 0
ADC Emp 18.07 19.29 6.721.0922.65 7.4 12.35 12.71 2.9 9.92 11.94 20.4
ADC Pop 21.24 21.33 0.423.18 24.2 4.4 14.25 14.52 1.9 14.11 14.36 1.7
Concentration
GINI Emp 0.86 0.82 -5.3 0.88 0.85 -2.8 0.73 0.71 -3.8 0.95 0.9 -5.3
GINI Pop 0.78 0.77 -0.9 0.81 0.8 -1.6 0.62 0.6 -3.5 0.84 0.83 -0.2
DELTA Emp 0.7 0.65 -7 0.75 0.71 -4.6 0.57 0.54 -4 0.88 0.79 -10.2
DELTA Pop 0.62 0.61 -1.4 0.68 0.67 -2.5 0.48 0.46 -4.1 0.71 0.71 0.3
San Francisco Philadelphia
%Ch. %Ch. %Ch. %Ch. %Ch. %Ch.
1980 1990 2000 80s 90s 80-00 1980 1990 2000 80s 90s 80-00
Centralization
MWI Emp 0.2 0.12 0.09 -38.1 -26.2 -54.3 0.35 0.3 0.23 -15.4 -23.8 -35.5
MWI Pop 0.07 0.02 0 -65.6 -78.6 -92.7 0.27 0.22 0.19 -15.5 -13.9 -27.3
ACI Emp 0.49 0.42 0.39-13.2 -6.9-19.2 0.59 0.55 0.49 -7.3 -11.3 -17.8
ACI Pop 0.38 0.34 0.32-11.6 -5.3-16.3 0.53 0.49 0.46 -7 -5.9-12.6
ADC Emp 19.94 21.79 22.58 9.3 3.6 13.211.1312.0713.29 8.4 10.2 19.4
ADC Pop 23.14 24.23 24.67 4.7 1.8 6.612.6513.3613.89 5.6 4 9.9
Concentration
GINI Emp 0.9 0.87 0.85 -3.2 -3.2 -6.3 0.85 0.81 0.72 -5.2 -10.9-15.5
GINI Pop 0.83 0.81 0.8 -2.4 -1.4 -3.8 0.75 0.69 0.66 -7.3 -4.6-11.6
DELTA Emp 0.77 0.73 0.69 -5.1 -5.3 -10.1 0.69 0.64 0.56 -7.4 -13.5 -19.8
DELTA Pop 0.7 0.67 0.66 -3.5 -2.1 -5.5 0.6 0.54 0.5 -9.9 -6.4 -15.6
1) MWI: Modified Wheaton index; ACI: Area-based centralization index; ADC: Weighted average
distance from CBD; GINI: Gini coefficient; DELTA: delta index.
2) Extremely low modified Wheaton index in San Francisco and its fast decline are due to the
presence of San Francisco Bay. Note this index is normalized by the distance from the CBD.
3) Comparison of indices across metropolitan areas should be done with caution because the
difference may due to the presence of unpopulated large tracts rather than built-up settlement
variations.

87
4.2.2. Growth Patterns of Metropolitan Employment Centers
Figures 4-2 to 4-7 chart identified employment centers by both GWR and minimum
density procedures and Tables 4-3 to 4-8 present changes of employment shares in these
centers over time. These tables are an extended version of the table in Gordon and
Richardson (1996, p.291), in which they tested the ‘beyond polycentricity’ hypothesis.
Rows in the tables indicate each census year’s employment centers defined by job
distribution in the corresponding year while columns show each year’s number of jobs and
shares of total employment.
Of main interest is the change by location type along the main diagonal (in bold
font). For example, with reference to New York (Table 4), center employment share
identified by the GWR method decreased from 22.8 percent to 21.0 percent in the 1990s.
We can also examine employment growth or decline within the fixed centers boundaries by
moving along each row. Referring to New York again, employment in zones identified as
centers by GWR procedure as of 1990 decreased by 163,544 (7.9 percent) while 2000
centers gained 35,851 jobs (1.9 percent).
What type of location has gained jobs in the recent periods? There are three
important findings from the trend analysis. First, jobs continued to decentralize from the
metropolitan core to suburbs in the 1980s and 1990s. The employment share in the core,
whether defined as the CBD or the main center, shrank in all six metropolitan areas for any
studied period. In particular, the CBDs of New York, Los Angeles, and Philadelphia
experienced absolute job losses. By 2000, the CBDs’ employment shares had decreased to 3
percent in Los Angeles and 12.6 percent in New York.

88
Secondly, jobs dispersion was a more common phenomenon than subcentering.
Dispersed job locations performed better than employment centers in almost all cases, with
the only exception being the 1990s in San Francisco. Dynamic employment subcentering
was typical of the two western metropolises, Los Angeles and San Francisco, rather than
being a norm. New clustering of jobs in surburban areas nearly offsets the jobs loss from
older centers in the two polycentric regimes. In other metropolitan areas, employment
growth in subcenters neither kept pace with the metropolitan average nor compensated for
employment share losses in the core.
Finally and most importantly, the trend analysis of six metropolitan forms shows
that spatial structures did not evolve in one direction. Overall, three different patterns of
spatial transformation were identified when examining decentralization and clustering
patterns. Each type consists of a pair of cases. The three spatial evolution patterns will be
explored in detail in the next section.

89
Table 4-3. Centers employment trends in New York, 1990 to 2000
a) GWR method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 2,061,564 22.8% 1,898,020 20.2% -163,544 -7.9% -44.2%
CBD
1)
3 1,268,19614.0%1,207,51812.8%-60,678 -4.8%-16.4%
Subcenters 31 793,368 8.8% 690,502 7.3% -102,866 -13.0% -27.8%
Dispersed 6,977,68077.2%7,511,43979.8%533,759 7.6%144.2%
2000 Centers 1,935,694 21.4% 1,971,545 21.0% 35,851 1.9% 9.7%
CBD
1)
2 1,194,99113.2%1,190,02512.6%-4,966 -0.4%-1.3%
Subcenters 33 740,703 8.2% 781,520 8.3% 40,817 5.5% 11.0%
Dispersed 7,103,55078.6%7,437,91479.0%334,364 4.7%90.3%
Total 9,039,244 100% 9,409,459 100% 370,215 4.1% 100%

b) Minimum density method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 2,741,243 30.3% 2,675,354 28.4% -65,889 -2.4% -17.8%
Main center 1 1,973,355 21.8% 1,984,192 21.1% 10,837 0.5% 2.9%
Subcenters 29 767,888 8.5% 691,162 7.3% -76,726 -10.0% -20.7%
Dispersed 6,298,00169.7%6,734,10571.6%436,104 6.9%117.8%
2000 Centers 2,595,091 28.7% 2,701,316 28.7% 106,225 4.1% 28.7%
Main center 1 1,975,972 21.9% 1,996,657 21.2% 20,685 1.0% 5.6%
Subcenters 26 619,119 6.8% 704,659 7.5% 85,540 13.8% 23.1%
Dispersed 6,444,15371.3%6,708,14371.3%263,990 4.1%71.3%
Total 9,039,244100%9,409,459100%370,215 4.1%100%
1) All employment centers identified south of Central Park in Manhattan are defined as CBD.

90
Table 4-4. Centers employment trends in Los Angeles, 1990 to 2000
a) GWR method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 1,797,308 26.6% 1,583,703 23.6% -213,605 -11.9% 629.3%
CBD 1 219,948 3.3% 171,566 2.6% -48,382 -22.0% 142.5%
Subcenters 43 1,577,360 23.4% 1,412,137 21.0% -165,223 -10.5% 486.8%
Dispersed 4,953,40073.4%5,133,06476.4%179,664 3.6%-529.3%
2000 Centers 1,987,947 29.4% 1,957,555 29.1% -30,392 -1.5% 89.5%
CBD 1 230,893 3.4% 196,695 2.9% -34,198 -14.8% 100.8%
Subcenters 41 1,757,054 26.0% 1,760,860 26.2% 3,806 0.2% -11.2%
Dispersed 4,762,76170.6%4,759,21270.9% -3,549 -0.1%10.5%
Total 6,750,708 100% 6,716,767 100% -33,941 -0.5% 100%

b) Minimum density method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 2,540,357 37.6% 2,241,297 33.4% -299,060 -11.8% 881.1%
Main center 2 1,032,457 15.3% 916,374 13.6% -116,083 -11.2% 342.0%
Subcenters 40 1,507,900 22.3% 1,324,923 19.7% -182,977 -12.1% 539.1%
Dispersed 4,210,35162.4%4,475,47066.6%265,119 6.3%-781.1%
2000 Centers 2,276,391 33.7% 2,169,966 32.3% -106,425 -4.7% 313.6%
Main center 2 948,608 14.1% 875,531 13.0% -73,077 -7.7% 215.3%
Subcenters 34 1,327,783 19.7% 1,294,435 19.3% -33,348 -2.5% 98.3%
Dispersed 4,474,31766.3%4,546,80167.7%72,484 1.6%-213.6%
Total 6,750,708100%6,716,767100%-33,941 -0.5%100%
1) Centers which were contiguous to the CBD or main center in any single census year are
considered as parts of the CBD or main center. This is why the entry in number of centers for
CBD or main center is sometimes larger than one.

91
Table 4-5. Centers employment trends in Boston, 1990 to 2000
a) GWR method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 423,626 19.4% 412,571 17.8% -11,055 -2.6% -9.0%
CBD 1 267,616 12.2% 280,837 12.2% 13,221 4.9% 10.8%
Subcenters 9 156,010 7.1% 131,734 5.7% -24,276 -15.6% -19.8%
Dispersed 1,764,97280.6%1,898,81582.2%133,843 7.6%109.0%
2000 Centers 341,885 15.6% 356,366 15.4% 14,481 4.2%11.8%
CBD 1 220,556 10.1% 238,101 10.3% 17,545 8.0% 14.3%
Subcenters 7 121,329 5.5% 118,265 5.1% -3,064 -2.5% -2.5%
Dispersed 1,846,71384.4%1,955,02084.6%108,307 5.9%88.2%
Total 2,188,598 100% 2,311,386 100% 122,788 5.6% 100%

b) Minimum density method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 591,727 27.0% 570,591 24.7% -21,136 -3.6%-17.2%
Main center 2 480,894 22.0% 502,857 21.8% 21,963 4.6% 17.9%
Subcenters 8 110,833 5.1% 67,734 2.9% -43,099 -38.9% -35.1%
Dispersed 1,596,87173.0%1,740,79575.3%143,924 9.0%117.2%
2000 Centers 483,972 22.1% 549,006 23.8% 65,034 13.4% 53.0%
Main center 2 453,186 20.7% 502,915 21.8% 49,729 11.0% 40.5%
Subcenters 5 30,786 1.4% 46,091 2.0% 15,305 49.7% 12.5%
Dispersed 1,704,62677.9%1,762,38076.2%57,754 3.4%47.0%
Total 2,188,598100%2,311,386100%122,788 5.6%100%
1) Centers which were contiguous to the CBD or main center in any single census year are
considered as parts of the CBD or main center. This is why the entry in number of centers for
CBD or main center is sometimes larger than one.

92
Table 4-6. Centers employment trends in Portland, 1990 to 2000
a) GWR method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 134,785 19.2% 143,557 13.0% 8,772 6.5% 2.2%
CBD 1 90,887 12.9% 97,750 8.8% 6,863 7.6% 1.7%
Subcenters 2 43,898 6.2% 45,807 4.1% 1,909 4.3% 0.5%
Dispersed 569,04580.8% 962,21887.0%393,173 69.1%97.8%
2000 Centers 121,127 17.2% 129,385 11.7% 8,258 6.8% 2.1%
CBD 1 81,021 11.5% 87,310 7.9% 6,289 7.8% 1.6%
Subcenters 2 40,106 5.7% 42,075 3.8% 1,969 4.9% 0.5%
Dispersed 582,70382.8% 976,39088.3%393,687 67.6%97.9%
Total 703,830 100% 1,105,775 100% 401,945 57.1% 100%

b) Minimum density method
# 1990 Employment 2000 Employment Job % Growth
centers Share Share Growth Growth Share
1990 Centers 243,573 34.6% 257,182 23.3% 13,609 5.6% 3.4%
Main center 1 188,426 26.8% 208,926 18.9% 20,500 10.9% 5.1%
Subcenters 3 55,147 7.8% 48,256 4.4% -6,891 -12.5% -1.7%
Dispersed 460,25765.4% 848,59376.7%388,336 84.4%96.6%
2000 Centers 266,045 37.8% 315,947 28.6% 49,902 18.8% 12.4%
Main center 1 191,210 27.2% 212,732 19.2% 21,522 11.3% 5.4%
Subcenters 6 74,835 10.6% 103,215 9.3% 28,380 37.9% 7.1%
Dispersed 437,78562.2% 789,82871.4%352,043 80.4%87.6%
Total 703,830100%1,105,775100%401,945 57.1%100%

93
Table 4-7. Centers employment trends in San Francisco, 1980 to 2000
a) GWR method
1980 Employment 1990 Employment 2000 Employment % Growth
# Share Share Share Growth Share
1980 Centers 624,676 27.0% 632,654 20.7% 645,409 19.0% 3.3% 1.9%
CBD 2 186,175 8.0% 172,081 5.6% 187,102 5.5% 0.5% 0.1%
Subcenters 17 438,501 18.9% 460,573 15.1% 458,307 13.5% 4.5% 1.8%
Dispersed 1,691,56273.0%2,418,19679.3%2,759,73881.0% 63.1% 98.1%
1990 Centers 653,741 28.2% 807,396 26.5% 771,235 22.6% 18.0% 10.8%
CBD 1 217,139 9.4% 204,524 6.7% 224,535 6.6% 3.4% 0.7%
Subcenters 21 436,602 18.8% 602,872 19.8% 546,700 16.1% 25.2% 10.1%
Dispersed 1,662,49771.8%2,243,45473.5%2,633,91277.4% 58.4% 89.2%
2000 Centers
2)
484,99520.9% 558,33618.3% 652,661 19.2% 34.6% 15.4%
CBD 2 204,169 8.8% 188,960 6.2% 211,195 6.2% 3.4% 0.6%
Subcenters 16 280,826 12.1% 369,376 12.1% 441,466 13.0% 57.2% 14.8%
Dispersed 1,831,24379.1%2,492,51481.7%2,752,48680.8% 50.3% 84.6%
Total Jobs 2,316,238 100% 3,050,850 100% 3,405,147 100% 47% 100%

b) Minimum density method
1980 Employment 1990 Employment 2000 Employment % Growth
# Share Share Share Growth Share
1980 Centers 749,987 32.4% 767,373 25.2% 763,643 22.4% 1.8% 1.3%
Main center 1 400,154 17.3% 407,344 13.4% 443,454 13.0% 10.8% 4.0%
Subcenters 13 349,833 15.1% 360,029 11.8% 320,189 9.4% -8.5% -2.7%
Dispersed 1,566,25167.6%2,283,47774.8%2,641,50477.6% 68.7% 98.7%
1990 Centers 801,095 34.6% 926,021 30.4% 886,695 26.0% 10.7% 7.9%
Main center 1 417,952 18.0% 447,956 14.7% 482,996 14.2% 15.6% 6.0%
Subcenters 18 383,143 16.5% 478,065 15.7% 403,699 11.9% 5.4% 1.9%
Dispersed 1,515,14365.4%2,124,82969.6%2,518,45274.0% 66.2% 92.1%
2000 Centers 809,09534.9% 946,21031.0% 1,151,744 33.8% 42.3% 31.5%
Main center 1 416,317 18.0% 444,962 14.6% 482,246 14.2% 15.8% 6.1%
Subcenters 20 392,778 17.0% 501,248 16.4% 669,498 19.7% 70.5% 25.4%
Dispersed 1,507,14365.1%2,104,64069.0%2,253,40366.2% 49.5% 68.5%
Total Jobs 2,316,238 100% 3,050,850 100% 3,405,147 100% 47% 100%
1) Centers which were contiguous to the CBD or main center in any single census year are
considered as parts of the CBD or main center. This is why the entry in number of centers for
CBD or main center is sometimes larger than one.
2) When using 2000 data by 2000 census tracts without converting it to 1990 census tracts, centers
employment share by the GWR method is 30.5% (1,070,799), combining shares in CBD and
subcenters, 6.3% (220,528) and 24.2% (850,271), respectively.

94
Table 4-8. Centers employment trends in Philadelphia, 1980 to 2000
a) GWR method
1980 Employment 1990 Employment 2000 Employment % Growth
# Share Share Share Growth Share
1980 Centers 400,215 21.0% 403,319 17.5% 359,736 14.7% -10.1% -7.5%
CBD 1 261,514 13.8% 260,959 11.3% 242,980 10.0% -7.1% -3.4%
Subcenters 12 138,701 7.3% 142,360 6.2% 116,756 4.8% -15.8% -4.1%
Dispersed 1,501,42979.0%1,897,13782.5%2,080,61985.3% 38.6% 107.5%
1990 Centers 370,028 19.5% 400,809 17.4% 331,451 13.6% -10.4% -7.2%
CBD 1 239,804 12.6% 247,803 10.8% 228,425 9.4% -4.7% -2.1%
Subcenters 13 130,224 6.8% 153,006 6.7% 103,026 4.2% -20.9% -5.0%
Dispersed 1,531,61680.5%1,899,64782.6%2,108,90486.4% 37.7% 107.2%
2000 Centers 359,41518.9% 374,04616.3% 360,556 14.8% 0.3% 0.2%
CBD 1 239,804 12.6% 247,803 10.8% 228,425 9.4% -4.7% -2.1%
Subcenters 10 119,611 6.3% 126,243 5.5% 132,131 5.4% 10.5% 2.3%
Dispersed 1,542,22981.1%1,926,41083.7%2,079,79985.2% 34.9% 99.8%
Total Jobs 1,901,644 100% 2,300,456 100% 2,440,355 100% 28% 100%

b) Minimum density method
1980 Employment 1990 Employment 2000 Employment % Growth
# Share Share Share Growth Share
1980 Centers 681,348 35.8% 646,320 28.1% 547,278 22.4% -19.7% -24.9%
Main center 2 497,467 26.2% 477,390 20.8% 419,952 17.2% -15.6% -14.4%
Subcenters 13 183,881 9.7% 168,930 7.3% 127,326 5.2% -30.8% -10.5%
Dispersed 1,220,29664.2%1,654,13671.9%1,893,07777.6% 55.1% 124.9%
1990 Centers 648,009 34.1% 726,650 31.6% 582,820 23.9% -10.1% -12.1%
Main center 3 461,648 24.3% 480,752 20.9% 417,574 17.1% -9.5% -8.2%
Subcenters 18 186,361 9.8% 245,898 10.7% 165,246 6.8% -11.3% -3.9%
Dispersed 1,253,63565.9%1,573,80668.4%1,857,53576.1% 48.2% 112.1%
2000 Centers 535,52628.2% 584,62325.4% 553,168 22.7% 3.3% 3.3%
Main center 4 397,500 20.9% 421,275 18.3% 389,134 15.9% -2.1% -1.6%
Subcenters 12 138,026 7.3% 163,348 7.1% 164,034 6.7% 18.8% 4.8%
Dispersed 1,366,11871.8%1,715,83374.6%1,887,18777.3% 38.1% 96.7%
Total Jobs 1,901,644 100% 2,300,456 100% 2,440,355 100% 28% 100%
1) Centers which were contiguous to the CBD or main center in any single census year are
considered as parts of the CBD or main center. This is why the entry in number of centers for
CBD or main center is sometimes larger than one.

95
Major Roads
1990 Centers (minimum density)
Main Center
Subcenters
County boundary
20 0 20 40 Miles
Major Roads
2000 Centers (minimum density)
Main Center
Subcenters
County boundary
20 0 20 40 Miles
Major Roads
1990 Centers (GWR)
CBD
Subcenters
County boundary
20 0 20 40 Miles
Major Roads
2000 Centers (GWR)
CBD
Subcenters
County boundary
20 0 20 40 Miles
Figure 4-2. Employment centers in the New York metropolitan area, 1990 to 2000
a) 1990 centers by GWR method b) 2000 centers by GWR method

c) 1990 centers by MD method d) 2000 centers by MD method

96
Major Roads
1990 Centers (GWR)
CBD
Subcenters
County boundary
20 0 20 40 Miles
Major Roads
2000 Centers (GWR)
CBD
Subcenters
County boundary
20 0 20 40 Miles
Major Roads
1990 Centers (minimum density)
Main center
Subcenters
County boundary
20 0 20 40 Miles
Major Roads
2000 Centers (minimum density)
Main center
Subcenters
County boundary
20 0 20 40 Miles
Figure 4-3. Employment centers in the Los Angeles metropolitan area, 1990 to 2000
a) 1990 centers by GWR method b) 2000 centers by GWR method

c) 1990 centers by MD method d) 2000 centers by MD method

97
10 0 10 20 Miles
Major Roads
1990 Centers (GWR)
CBD
Subcenters
County boundary
10 0 10 20 Miles
Major Roads
2000 Centers (GWR)
CBD
Subcenters
County boundary
10 0 10 20 Miles
Major Roads
1990 Centers (minimum density)
Main Center
Subcenters
County boundary
10 0 10 20 Miles
Major Roads
2000 Centers (minimum density)
Main Center
Subcenters
County boundary
Figure 4-4. Employment centers in the Boston metropolitan area, 1990 to 2000
a) 1990 centers by GWR method b) 2000 centers by GWR method

c) 1990 centers by MD method d) 2000 centers by MD method

98
808 16Miles
Major Roads
1990 Centers (GWR)
CBD
Subcenters
County boundary
808 16Miles
Major Roads
2000 Centers (GWR)
CBD
Subcenters
County boundary
808 16Miles
Major Roads
1990 Centers (minimum density)
Main Center
Subcenters
County boundary
808 16Miles
Major Roads
2000 Centers (minimum density)
Main Center
Subcenters
County boundary
Figure 4-5. Employment centers in the Portland metropolitan area, 1990 to 2000
a) 1990 centers by GWR method b) 2000 centers by GWR method

c) 1990 centers by MD method d) 2000 centers by MD method

99
20 0 20 40 Miles
Major Roads
1980 Centers (minimum density)
Main Center
Subcenters
County boundary
20020 40Miles
Major Roads
2000 Centers (minimum density)
Main Center
Subcenters
County boundary
20020 40Miles
Major Roads
1980 Centers (GWR)
CBD
Subcenters
County boundary
20020 40Miles
Major Roads
2000 Centers (GWR)
CBD
Subcenters
County boundary
Figure 4-6. Employment centers in the San Francisco metropolitan area, 1980 to 2000
a) 1980 centers by GWR method b) 2000 centers by GWR method

c) 1980 centers by MD method d) 2000 centers by MD method

100
Major Roads
1980 Centers (GWR)
CBD
Subcenters
County boundary
10 0 10 20 Miles
Major Roads
2000 Centers (GWR)
CBD
Subcenters
County boundary
10 0 10 20 Miles
Major Roads
1980 Centers (minimum density)
Main Center
Subcenters
County boundary
10 0 10 20 Mile
Major Roads
2000 Centers (minimum density)
Main Center
Subcenters
County boundary
10 0 10 20 Mile
Figure 4-7. Employment centers in the Philadelphia metropolitan area, 1980 to 2000
a) 1980 centers by GWR method b) 2000 centers by GWR method

c) 1980 centers by MD method d) 2000 centers by MD method

101
4.2.3. Three Patterns of Spatial Evolution
Portland and Philadelphia represent the first type of spatial change where jobs
dispersion was predominant without significant suburban clustering. Both decentralization
and deconcentration occurred to the greatest extent. The employment share in the urban core
shrank relatively quickly, but the subcenters were not strong enough to be a magnet for the
decentralizing jobs. As a result, center employment shares defined by any measure fell
substantially.
The CBD of Philadelphia and its surrounding areas underwent remarkable job losses
in absolute terms and hence the employment density fell in the central location. The size of
the main center that passes a minimum density threshold shrank substantially, with its
employment share decreased from 26.2 percent in 1980 to 15.9 percent in 2000. The
subcenters also experienced job losses in the 1990s, while the employment shares in the
subcenters combined were stable in the 1980s. Virtually all metropolitan employment
growth for the recent two decades occurred outside employment centers. Philadelphia
became, as Lang (2003) describes, “the edgeless metropolis of the north.”
In Portland, the proportion of dispersed jobs exploded. The Portland metropolitan
area added more than four-hundred thousand jobs in the 1990s, which is a 57 percent
increase from 1990. Thus, most areas within the region benefited from the fast employment
growth, but with a disproportionate growth share directed to lower density zones. Fully 88
percent to 98 percent of metropolitan employment growth, depending upon center definition,
occurred at dispersed locations. Accordingly, the center employment share diminished
substantially. The employment share of the CBD dropped from 12.9 percent to 7.9 percent
and from 26.8 percent to 19.2 percent in the main center. With regard to subcenter growth,

102
the two different center identification procedures show mixed outlooks. Densification in the
suburbs especially along State Highway 217, about ten miles southwest of Portland’s
downtown, resulted in formation of new subcenters when identified via the minimum density
method. Yet, the GWR method fails in identifying these peaks. In sum, the Portland
metropolitan area became denser but flatter in the 1990s.
Contrary to the first pattern, the urban cores performed better than the suburban
centers and remained strong employment agglomerations throughout the 1990s in Boston
and New York. Even the small job loss in the CBDs was mostly offset by the growth in
adjacent areas. Thus, the main centers’ share remained stable. The spatial process in the
two northeastern metropolises can be summarized as little decentralization and moderate
deconcentration. Overall loss of center employment share was smaller than in the first two
metropolises.
The centralized structure of Boston did not diminish in the last decade. In Table 4-5,
the employment share in the CBD defined by the GWR method appears to have fallen from
12.2 to 10.3 percent. It fell because job centers in Cambridge that were parts of the 1990
CBD became disqualified as centers in 2000. But, this was not due to job losses in
Cambridge but because of densification of the surrounding areas. To put it more technically,
the small window GWR surface in Cambridge area is not significantly higher than the large
window GWR surface in 2000 not because the former fell but because the latter arose.
Vitality of the CBD can also be confirmed by employment growth rates in the CBDs by each
year’s definition, 4.9 percent and 8 percent, which are similar to or higher than the
metropolitan average. The more broadly defined main center also maintained its

103
employment share at around 22 percent. On the contrary, job concentrations in surburban
Boston were trivial in 1990 and shrank even further by 2000.
Manhattan, the largest employment agglomeration in the country also maintained its
predominance throughout the 1990s, containing about two million jobs. Although the
downtown in lower Manhattan experienced some job loss, it was replenished in the lower
density parts of the island. Thus, employment share in the main center was stable at above
21 percent throughout the period. Unlike in Boston, the suburban employment centers
particularly in New Jersey and Long Island also performed well. As a result, there was only
a minor loss of center employment share in the New York metropolitan area in the last
decade.
Two polycentric regimes in the west, Los Angeles and San Francisco, have taken a
quite different path from the two previous development patterns; call it decentralized
concentration. Whereas employment agglomeration in the regional core shrank absolutely in
Los Angeles and relatively in San Francisco, a significant proportion of decentralizing jobs
reconcentrated in suburban centers. The share of clustered employment remained the most
stable in this polycentric structure for the last decade.
The dynamics of subcentering in the two metros call for lengthy explanation because
the two center identification procedures provide different results. In Los Angeles, suburban
employment centers defined by the GWR method added about two hundred thousand new
jobs in all, while the minimum density method captures the employment loss of a similar
size. In other words, the more flexible nonparametric method captures the rise of new
clusters in the outer ring suburbs. Ten new subcenters emerged while seven disappeared and
five merged into others in the metropolitan region. Most new clustering occurred around

104
border areas of Los Angeles and Orange and Riverside counties, while many of inner ring
subcenters in Los Angeles County disappeared perhaps as a result of the industrial
restructuring in the region. The net effect was about a three percentage point increase in the
subcenters’ employment share. Yet, those emerging small peaks in outer ring suburbs were
not dense enough to pass the minimum density test. Thus, the minimum density method
accounts for only 34 subcenters in 2000 and a decreased employment share by three
percentage point from 1990.
On the contrary, in San Francisco, the nonparametric method appears to represent
less clustering in the 1990s while minimum density method identifies more subcenters in
2000 than in 1990. The GWR results suggest that subcenters’ employment share fell from
20 to 13 percent in the 1990s. This extraordinary drop, however, is a statistical construct due
to the imperfect data conversion between census years. Whereas I converted all three year’s
data onto 1990 census tracts as mentioned above, the 1990 tract boundaries do not reflect
well the new developments in the 1990s. This mismatch problem is especially notable in
relatively new and fast growing areas such as Silicon Valley, where the older tracts were
typically very large and have been split in the later surveys. Thus, there was substantial
noise in converting 2000 data back onto 1990 census tracts and this resulted in the failure to
identify the densification of high tech jobs in Silicon Valley by the GWR method.
This reasoning can be confirmed by the fact that a huge Silicon Valley subcenter is
found when using 2000 data and 2000 census tracts without the data conversion. This
subcenter amalgamates high tech clusters from Mountain View, CA in the westerly direction
to Milpitas, CA to the east, containing 283,850 jobs. When using 2000 census tracts, the
total centers employment share by the GWR method is as large as 30.5 percent (1,070,799

105
jobs), combining shares in the CBD and subcenters, 6.3 percent (220,528) and 24.2 percent
(850,271), respectively. The minimum density method also offers the substantial expansion
of the Silicon Valley subcenter and the overall growth of clustered jobs. In sum, the results
of both procedures present about a four percentage point increase in subcenters employment
share in the San Francisco metropolitan area in the 1990s.
It was the clustering of high technology firms that led to the rise and growth of
employment subcenters in the both polycentric regimes. The world-renowned clusters of
semiconductor and IT firms in Silicon Valley, and the computer and biotechnology complex
in Irvine/Santa Ana/Costa Mesa have grown to be even larger regional employment centers
than each region’s CBD by the recent turn of the century. These new employment centers
are very different from either traditional downtowns or old industrial space in their
functions, infrastructure, and urban form, bearing different policy implications. For
instance, they have much lower density, less congestion, and higher amenities often in the
form of industrial and office parks. They are better accessed by car than by public transit.
The impacts and policy implications of these spatial transformations should provide good
research opportunities in the future.

4.3. DISCUSSION
This chapter explored spatial changes in six US metropolitan areas in order to
address the question whether emerging urban forms can be characterized as increasingly
edgy or edgeless. Findings parallel the results of Gordon and Richardson (1996) and Lang
(2003). Jobs continued to decentralize from the metropolitan core to the suburbs in the

106
1980s and 1990s, and jobs dispersion was a more common phenomenon than subcentering
for the periods studied.
Nevertheless, there was a remarkable variation in spatial trends among the six
metropolitan areas. Three patterns of spatial development were distinguished (Figure 4-2).
Jobs dispersion was the predominant spatial process in Portland and Philadelphia, where
rapid decentralization and deconcentration occurred. In New York and Boston, the main
centers in the core remained strong agglomerations while subcenters’ employment share
further diminished. In contrast, progressive employment subcentering occurred in two
polycentric regions, Los Angeles and San Francisco.
To the extent that the results for six metropolitan areas can be generalized, the
results imply that metropolitan spatial evolution may not be a linear process from
monocentric through polycentric and to dispersed structure. A more plausible scenario is
that some metropolitan structures are undergoing the transition to a polycentric structure
while others are more apt to diffuse. In other words, agglomeration economies are realized
differently in different regions. Although more thorough examinations are needed to explain
the sources of the different spatial manifestations, the results of this research provide
important clues for future research.
First, the geographical and historical contexts of an individual metropolis strongly
affect the path by which it responds to global trends such as the effects of ever decreasing
transportation costs and IT development. For instance, the decentralized structure of Los
Angeles has often been associated with its substantial land resources and highway and
boulevard networks whereas San Francisco’s polycentricity is largely configured by the
topography of the region including the presence of the bay (Lang 2003).

107
Spatial development patterns in urban places are also path dependent (Giuliano et al.
2005) as are technology adoption and industrial development (Nelson and Winter 1982;
Arthur 1994). “Positive feedbacks” or “Self reinforcing” is the key notion in the path
dependence literature, as discussed in Chapter 3. This tendency in spatial development was
identified in this chapter: New York and Boston with big and long established CBDs were
less subject to decentralization; while polycentricity of Los Angeles and San Francisco was
reinforced in the last decade.
Second, economic restructuring is also an important factor in metropolitan spatial
development. Different industrial sectors benefit from different sources of agglomeration
economies with varied geographical ambits and distance decay functions (Dekle and Eaton
1999; Rosenthal and Strange 2001). For instance, the benefits of CBD location are greater
for the finance sectors than any other industrial sectors. Thus, the strong and relatively
stable agglomerations in the CBDs of New York and Boston can be associated with their
strong industrial shares of jobs in the finance and business services sectors.
On the other hand, research and development, or production in high technology
sectors has a tendency to cluster in the suburbs of large metropolitan areas. The advantage
of local proximity in these sectors involves the intensive creation and exchange of tacit
knowledge, ultimately contributing to innovation and growth of firms in the clusters. While
the internal dynamics of the two high-tech clusters in Santa Clara and Orange Counties are
discussed in the economic geography literature (Scott 1988; Scott 1990; Saxenian 1994), the
current research shows that this spatial logic works as an important agglomerating force at
the sub-metropolitan scale, creating and fostering new employment subcenters.

108
Other aspects of economic restructuring, however, contribute to the dispersion of
employment locations. Proximity to consumers is a more important location factor in
personal services and retail industries. Thus, the continuing tertiarization of metropolitan
economies will result in further employment dispersion given the extensive suburbanization
of population in US metropolitan areas.
To summarize, the dominant trend in recent decades involved jobs dispersion.
However, there was significant variation in spatial decentralization trends among the six
metropolitan areas studied. They seem to have developed unique patterns of
decentralization, in light of their histories and circumstances to limit the growth of
commuting times. Policy makers have, for the most part, avoided peak-load pricing of road
use. Yet, it appears that land markets allow unique land use pattern adaptations that limit the
effects that metropolitan growth or size have on commuting cost increases.

109
b) MD procedure results
0%
5%
10 %
15 %
20 %
25 %
30 %
0% 5 % 10 % 1 5% 20% 25% 30 %
S ubcenters
CB D
19 80
19 90
20 00
Los Angeles
Philadelphia
Portland
New York
San Francisco
Boston
Note: Dotted line for SF indicates the result when using 2000 census tracts without data conversion.
a) GWR procedure results
0%
5%
10 %
15 %
20 %
25 %
30 %
0 % 5 % 10% 15% 2 0 % 2 5% 30 %
S ubcen te rs
Main C enter
1980
1990
2000
Los Angeles
Philadelphia
Portland
New York
San Francisco
Boston
Figure 4-8. Changes of employment shares for both definitions of centers, 1980 to 2000

110
CHAPTER 5.
URBAN SPATIAL STRUCTURE AND COMMUTE ECONOMIES

5.1. INTRODUCTION
This chapter
4
presents an analysis of the impacts on commuting time of metropolitan
level spatial structure. The links between urban form or land use and transportation have
been one of the most debated topics in recent years. A number of surveys of the literature
are already published (Anderson, Kanaroglou, and Miller 1996; Badoe and Miller 2000;
Crane 2000; Ewing and Cervero 2001). Overall, empirical analysis results were mixed, with
some finding significant effects and others presenting marginal effects.
The gap in the literature is due to many factors. The use of different geographical
scales (Crane 2000; Boarnet and Crane 2001) as well as different data and methodologies
(Badoe and Miller 2000) is one of the reasons leading to varying (often contradicting)
conclusions. Schwanen (2003) shows that eight different spatial dimensions –
mono/polycentrism, urban size, density, distance to activity centers, proximity to transport
infrastructure, land use mix, urban design, accessibility – are tied to different geographical
scales from individual structure level through neighborhood, and to metropolitan areas.
I focus on the urban form effects at the metropolitan scale. More specifically, I
examine whether metro level spatial restructuring towards more polycentric and dispersed

4
Earlier version of this chapter was presented at the 45
th
annual meeting of the Western
Regional Science Association, February 2006, in Santa Fe, New Mexico.

111
forms have led to reduced or increased average commute time. While most of recent
research efforts have centered on neighborhood level studies, metro level analysis results
will also have significant implications given that neighborhood scale impacts may be
contingent on higher level spatial structure.
5
In addition, findings will constitute a part of the
discussion on metropolitan spatial evolution which is a main theme throughout this
monograph.
Urban economists and planners hold very different views on the commuting impacts
of recent metro level spatial evolution. As discussed in chapter 3, commuting economies are
one of building blocks in theoretical urban models. Thus, spatial adjustments in cities
should occur in a way that shortens commute times of residents as a city grows. For
instance, it is implied that job dispersal leads to lower commuting costs and distances in
Wheaton’s (2004) urban model allowing land use mix.
Gordon and Richardson actually found that polycentric or dispersed spatial
structures were associated with shorter commute times in early 1980s (Gordon and Wong
1985; Gordon, Kumar, and Richardson 1989). A seeming paradox, constant average
commute time in spite of increased congestion and commuting distance, led them to suggest
that individual households and firms “co-locate” to reduce commute time and that this
spatial adjustment can be more easily made in dispersed metropolitan space with many
alternative employment centers and residential location choices (Gordon, Richardson, and
Jun 1991; Levinson and Kumar 1994). A more recent study using a panel data also found
that jobs decentralization contributes to shorter average commute distance (Crane and

5
“Transit neighborhoods had decidedly higher rates of bus commuting only in the Bay Area.
Islands of transit-oriented neighborhoods in a sea of freeway-oriented suburbs seems to have
negligible effects on transit commuting (Cervero and Gorham 1995, p.210).”

112
Chatman 2003). However, these studies depended on limited spatial variables, density and
the proportion of employment in central cities versus suburbs.
On contrary, most urban planners believe that dispersion of jobs and population or
sprawl type development causes more frequent and longer travels, more auto uses, and hence
more congestion (Sarzynski et al. 2006). Cervero and Landis (1992) argue that the co-
location process may not work properly to reduce commutes due to growing two-earner
households, location barriers and restricted residential and job mobility, and increased auto
dependency.
Cervero and Wu (1998) show that both commute time and distance increased with
jobs subcentering and decentralization in the San Francisco Bay Area in the 1980s. But, this
case study does not control effects from other factors such as increased wealth. More recent
studies employed a cross section analysis utilizing improved land use measures. A study on
the links between several sprawl indicators and transportation outcomes in 83 US
metropolitan areas found significant impacts of compactness on more non-auto uses, but not
on shortened average commute time and less congestion delay (Ewing, Pendall, and Chen
2003).
Another study on the sprawl impacts with a sample of 50 large urban areas found
mixed results (Sarzynski et al. 2006). Controlling for congestion level in 1990 and other
demographic and transportation supply changes, density/contiguity and housing centrality
were associated with more congestion while housing-job proximity were related with less
congestion in 2000. Whereas this study significantly improved regression models
accounting for simultaneous relationships and time lags, the mixture of metro level spatial

113
structure and micro scale land use factors makes it complicated to interpret the impacts of
recent evolution of metropolitan spatial structure.
This chapter attempts to associate the variation in commute time with more
straightforward variables measuring metropolitan spatial restructuring towards dispersed and
polycentric forms. One Dutch study took a similar approach, providing a partial evidence of
the co-location hypothesis. This study used four dummy variables – centralized/
decentralized/exchange commuting/cross-commuting – each indicating a spatial structure
type classified by Van der Lann (1998) based on commuting flows between the central city
and suburban municipalities. I use continuous rather than dichotomous spatial variables
because metropolitan areas exist in a wide spectrum with respect to spatial structure as seen
in chapter 2.

5.2. DESCRIPTIVE ANALYSIS: COMMUTE TIME BY LOCATION TYPE
Tables 5-1 and 5-2 present commute times by location type of workplace defined by
the two different procedures. Metropolitan average commute time in the tables is slightly
different from those that are usually reported on residential location basis because some
workers commute from and to outside metropolitan boundary. The former is slightly longer
than the latter in most cases since a metropolis usually an importer of commute flows and
commutes from outside are often longer than metropolitan residents.
The discussion focuses on commute time by drive alone mode. The reasons are two
fold. First, drive alone mode accounts for the most of commute, more than 90% of total
commute volume in most metropolitan areas except just a few largest metropolitan areas
such as New York, Chicago, Washington DC, and San Francisco. Second, commutes by

114
other modes such as transit or carpool are usually slow in speed and are subject to many
factors other than spatial structure such as transit infrastructure and local and individual
situations.
As expected, workers in larger metropolitan areas spend more time on journey-to-
work trips. This is the case in both monocentric and polycentric contexts. Average
commute time by drive alone mode in largest metro group is 27.8 minutes, while it is 24.1
and 22.3 minutes in medium and small size metropolitan areas, respectively.
Turning to the average commute times by location type, CBD workers take
substantially longer commute in time than other metropolitan workers. They spend about
ten more minutes than those who work outside CBD in largest metropolitan areas. There are
at least two factors for the longer commute time of CBD workers. First, firms located in
CBD should draw employees from a longer distance because employment is more
centralized than residence distribution (Wheaton 2004). Secondly, CBD is apparently the
most congested area within a metropolitan area.
Thus, the bigger size of CBD should be associated with larger extra commute time
for CBD workers. For instance, CBD workers’ commute time is about twice longer than
metropolitan average in New York where CBD accounts for almost a million employments.
Also in Philadelphia, Chicago, and Boston that also have a relatively big CBD, CBD
workers spend about 40 to 53% extra commute times, while the extra commute times are
only about 15 to 38% in Los Angeles, San Francisco, Detroit, and Dallas with smaller CBD
employment shares. However, the extra commute time to CBD is trivial (4.4%) in small
metropolitan areas with population under one million, where we find only one or two, or no

115
subcenter at all. The average extra commute time of CBD workers is about 12% in medium
size metro group.
The duration of commute to subcenters is only slightly longer than commutes to
dispersed destinations on average. Average extra commute for subcenter workers compared
with those with dispersed workplace location was only about 5% even in largest metro
group. The extra commute is larger than 10% only in four metropolitan areas, Chicago,
Cincinnati, Las Vegas, and Rochester. Subcenter workers’ commute time is about the same
as that of dispersed location workers in both medium and small metropolitan areas.
The commuting disadvantage of CBD workers is more or less a function of
metropolitan population size. This can be also verified by simply plotting commute times by
location type against log population size (Figure 5-1). While each commute time is a
function of log population size, the slope is significantly steeper for CBD workers.
Approximately doubling metro population (unit log point increase) increases the commute
time of CBD workers by 6 minutes. The commute time increases are only 3 and 2 minutes
for subcenters and dispersed location workers, respectively.
Thus, subcenter location on average appears to have the edge in commuting
economies. This can be realized to the extent that it provides agglomeration economies
matching those in CBD. However, authors observe that some subcenters or edge cities
already grew out of initial advantages and suffer similar diseconomies of CBD (Fulton 1996;
Lang 2003). A closer look at commute times by individual subcenter in New York and Los
Angeles reveals that commute time to some subcenters is almost as long as that to CBD.
Overall, the larger and more centrally located a subcenter is, the longer the commute time is.

116
On the other hand, urban economic theory predicts that the longer commute of CBD
workers should be compensated by higher wages (Timothy and Wheaton 2001). Thus, firms
in CBD should take advantage of agglomeration to pay off the extra labor costs or CBD
location should provide other unique amenities of density such as cultural and recreational
opportunities. A relatively stable employment concentration in Manhattan in spite of
doubled commute time than dispersed location presents a good case as shown in chapter 4.
It remains to be seen how information technology will change the landscape.

117
Table 5-1. Commute times by location type of workplace defined by the GWR procedure
MSA Population All modes Drive alone mode
name size Metro CBD Sub- Dis- Metro CBD Sub- Dis-
centers persed centers persed
3 millions and plus 29.4 38.8 29.9 28.4 27.8 37.1 28.5 27.2
New York 21199865 34.3 51.1 38.6 31.6 28.5 55.6 30.2 27.8
Los Angeles 16369949 29.0 39.0 30.0 28.1 27.8 36.6 28.9 27.0
Chicago 915754031.346.433.3 29.7 28.9 41.8 32.1 28.0
Washington 760807032.142.032.2 31.2 30.3 40.2 30.2 29.8
San Francisco 7039362 30.4 40.9 30.7 29.4 28.4 39.3 29.3 27.8
Philadelphia 618846327.738.826.4 26.6 26.1 36.6 26.1 25.7
Boston 582867228.342.326.5 27.2 27.1 41.6 25.9 26.7
Detroit 545642826.632.027.7 25.9 26.2 31.0 27.7 25.4
Dallas 522180128.133.328.5 27.6 27.4 31.5 28.0 27.1
Houston 466957129.235.830.0 28.2 28.1 32.9 28.9 27.3
Atlanta 411219831.937.832.4 31.3 30.9 36.0 31.4 30.3
Miami 387638028.935.829.6 28.0 27.9 33.8 28.9 27.1
Seattle 355476027.935.127.5 27.1 26.2 30.7 26.3 25.8
Phoenix 325187626.232.225.6 25.7 25.4 31.1 24.7 25.0
1 to 3 millions 24.8 28.0 23.9 24.4 24.1 26.9 23.4 23.8
Minneapolis 296880624.429.325.3 23.8 23.7 27.5 24.5 23.4
Cleveland 294583124.533.123.8 23.9 23.8 31.1 23.6 23.5
San Diego 2813833 25.3 27.7 26.1 24.9 24.5 25.0 25.1 24.2
St. Louis 2626411 26.4 31.2 27.8 25.7 25.5 29.8 27.2 24.9
Denver 258150626.132.124.3 25.6 25.2 29.8 24.0 24.9
Tampa 239599725.530.626.0 25.1 25.2 29.9 25.8 24.8
Pittsburgh 235869525.833.021.5 24.6 24.9 31.7 21.5 24.2
Portland 226522324.428.621.8 24.1 23.2 26.2 21.1 23.1
Cincinnati 197920224.928.926.6 24.3 24.3 27.5 26.4 23.8
Sacramento 179685725.429.423.025.224.827.8 22.624.7
Kansas City 1776062 24.0 28.7 23.5 23.5 23.4 27.9 23.1 23.0
Milwaukee 168957222.525.521.9 22.2 21.9 24.4 21.7 21.7
Orlando 164456128.129.225.2 28.0 27.5 28.1 25.2 27.5
Indianapolis 160748624.828.3 26.7 24.1 24.4 27.9 26.0 23.7
San Antonio 1592383 24.9 26.9 25.2 24.5 24.0 25.9 24.1 23.7
Norfolk 156954124.025.224.2 23.8 23.6 24.9 22.8 23.7
Las Vegas 1563282 24.1 25.8 26.0 23.3 22.6 24.3 25.9 21.8
Columbus 154015725.027.024.6 24.8 24.4 26.1 24.7 24.2
Charlotte 149929327.430.323.327.226.629.5 22.626.3
New Orleans 1337726 28.6 32.1 25.5 28.0 27.0 30.3 24.5 26.5
Salt Lake City 1333914 23.0 28.6 23.4 22.4 21.9 26.2 22.4 21.5
Greensboro 125150922.922.722.6 23.0 22.4 22.5 22.2 22.5
Austin 124976326.427.625.5 26.1 25.5 26.8 24.5 25.2
Nashville 123131127.630.1 27.126.929.5 26.4
Providence 118861320.922.918.8 20.6 20.8 22.9 19.1 20.5
Raleigh 118794126.428.324.4 26.6 25.9 27.5 25.0 25.9
Hartford 118380323.727.8 23.023.527.0 22.9
Buffalo 117011121.425.322.7 21.0 21.0 24.3 22.6 20.6
Memphis 113561425.827.525.1 25.7 25.0 27.0 23.9 24.9
Jacksonville 110049127.029.522.5 26.6 26.5 28.8 21.5 26.2
Rochester 109820121.623.223.8 21.3 21.4 22.5 23.4 21.1
Grand Rapids 1088514 22.1 23.2 19.0 22.3 21.7 22.8 19.0 21.9
Oklahoma City 1083346 22.6 26.4 19.8 22.6 22.3 26.1 19.9 22.2
Louisville 102559824.425.924.6 24.1 23.7 25.3 23.6 23.4

118
Table 5-1. Continued
MSA Population All modes Drive alone mode
name size Metro CBD Sub- Dis- Metro CBD Sub- Dis-
centers persed centers persed
half to 1 millions 22.9 23.8 22.2 22.8 22.3 23.3 21.7 22.2
Richmond 99651225.727.224.8 25.5 25.0 26.8 24.1 24.8
Greenville 96244123.124.020.3 23.2 22.6 23.3 20.5 22.7
Dayton 95055821.623.920.8 21.0 21.3 23.7 20.5 20.7
Fresno 92251622.523.020.0 22.6 21.1 22.3 19.0 21.2
Birmingham 92110628.128.628.7 28.0 27.4 28.0 27.8 27.2
Albany 87558323.527.322.0 23.1 23.3 27.0 22.3 22.9
Tucson 84374623.722.7 23.822.922.5 22.9
Tulsa 80323522.723.522.922.622.223.0 22.522.0
Syracuse 73211721.022.4 20.720.922.3 20.5
Omaha 71699820.522.221.2 19.8 19.9 21.9 20.5 19.1
Albuquerque 71273822.923.0 22.8 22.2 22.9 22.1
Knoxville 68724924.723.819.0 25.0 24.3 24.3 18.9 24.5
El Paso 679622 22.6 25.3 22.1 22.5 21.7 23.4 20.9 21.7
Bakersfield 66164522.821.2 23.321.019.8 21.4
Allentown 63795821.322.3 21.321.322.4 21.3
Harrisburg 62940123.927.1 23.523.526.1 23.2
Scranton 62477620.220.320.7 20.1 20.2 20.7 20.8 20.0
Toledo 61820320.922.421.9 20.7 20.7 22.3 21.3 20.5
Baton Rouge 602894 26.1 26.6 24.6 26.1 25.2 25.8 23.8 25.2
Springfield 60156420.722.5 20.621.022.3 20.9
Youngstown 59474620.420.820.9 20.3 20.2 20.7 20.7 20.1
Sarasota 58995922.423.221.5 22.3 21.8 22.5 21.4 21.8
Little Rock 583845 24.8 27.0 24.2 24.5 24.0 26.2 23.8 23.6
McAllen 56946320.320.1 20.419.719.9 19.7
Stockton 56359823.723.519.9 24.3 22.9 22.3 19.3 23.5
Charleston 54903325.224.9 25.224.424.4 24.5
Wichita 54522020.121.5 19.919.720.9 19.5
Mobile 54025825.126.324.9 25.0 24.1 25.7 23.9 23.9
Columbia 53669125.025.8 24.924.625.6 24.4
Colorado Spring 516929 22.1 22.4 21.6 22.1 21.7 21.7 20.6 21.8
Fort Wayne 502141 21.7 21.9 21.7 21.3 21.7 21.3
1) All commute times are calculated based on the place of work.
2) The table is sorted by the size of metro population.

119
Table 5-2. Commute times by location type of workplace defined by the MD procedure
MSA Population All modes Drive alone mode
name size Metro Core Sub- Dis- Metro Core Sub- Dis-
center centers persed center centers persed
3 millions and plus 29.4 37.0 30.9 27.2 27.8 35.2 29.8 26.3
New York 21199865 34.3 50.5 39.9 28.9 28.5 52.8 34.6 26.9
Los Angeles 16369949 29.0 35.8 30.9 27.1 27.8 34.1 29.6 26.1
Chicago 9157540 31.343.032.5 28.5 28.9 38.3 32.3 27.6
Washington 7608070 32.140.6 33.7 29.3 30.3 37.8 32.6 28.3
San Francisco 7039362 30.4 38.5 32.3 28.0 28.4 35.4 30.8 26.8
Philadelphia 6188463 27.736.5 29.5 26.0 26.1 34.1 28.3 25.3
Boston 5828672 28.338.528.3 26.0 27.1 37.6 28.2 25.8
Detroit 5456428 26.632.028.3 25.5 26.2 31.0 28.3 25.0
Dallas 5221801 28.133.828.6 26.0 27.4 32.8 28.0 25.4
Houston 4669571 29.233.629.9 27.3 28.1 31.7 29.0 26.5
Atlanta 4112198 31.937.335.0 29.9 30.9 35.6 34.0 29.0
Miami 3876380 28.934.329.3 27.3 27.9 33.0 28.7 26.4
Seattle 3554760 27.932.928.7 25.8 26.2 29.9 27.5 24.7
Phoenix 3251876 26.230.025.8 24.9 25.4 29.3 24.9 24.1
1 to 3 millions 24.827.425.324.024.126.5 24.623.4
Minneapolis 2968806 24.427.3 26.3 23.2 23.7 26.2 25.6 22.7
Cleveland 2945831 24.531.224.3 23.4 23.8 29.6 23.7 23.0
San Diego 2813833 25.3 26.1 27.9 24.3 24.5 24.3 27.0 23.6
St. Louis 2626411 26.4 30.5 30.0 24.6 25.5 29.3 28.9 23.9
Denver 2581506 26.130.926.6 25.0 25.2 28.9 26.0 24.3
Tampa 2395997 25.530.225.9 24.2 25.2 29.6 25.5 23.9
Pittsburgh 2358695 25.831.524.4 24.1 24.9 30.4 24.1 23.8
Portland 2265223 24.428.623.7 23.4 23.2 26.6 22.6 22.6
Cincinnati 1979202 24.927.727.4 24.0 24.3 26.8 26.8 23.6
Sacramento 1796857 25.428.1 27.7 24.5 24.8 27.1 27.2 24.0
Kansas City 1776062 24.0 27.7 23.5 23.3 23.4 27.0 22.9 22.8
Milwaukee 1689572 22.524.521.9 22.2 21.9 23.8 21.8 21.7
Orlando 1644561 28.129.130.4 27.4 27.5 28.3 29.8 26.9
Indianapolis 1607486 24.828.4 26.1 23.4 24.4 27.9 25.7 23.0
San Antonio 1592383 24.9 26.9 25.1 24.4 24.0 26.0 24.0 23.7
Norfolk 1569541 24.026.225.1 23.4 23.6 25.8 23.8 23.3
Las Vegas 1563282 24.1 25.3 23.0 22.6 23.8 21.4
Columbus 1540157 25.025.926.4 24.6 24.4 25.4 25.5 24.0
Charlotte 1499293 27.430.624.826.526.629.7 24.325.7
New Orleans 1337726 28.6 31.5 29.2 27.5 27.0 29.7 27.6 26.1
Salt Lake City 1333914 23.0 26.5 19.5 21.9 21.9 24.8 19.0 21.1
Greensboro 1251509 22.922.9 22.9 22.9 22.4 22.7 22.5 22.4
Austin 1249763 26.427.527.5 25.9 25.5 26.7 26.5 24.9
Nashville 1231311 27.629.729.7 26.6 26.9 29.2 29.2 25.9
Providence 1188613 20.922.818.6 20.6 20.8 22.8 19.0 20.5
Raleigh 1187941 26.427.224.2 26.7 25.9 26.6 24.6 26.0
Hartford 1183803 23.727.421.0 22.9 23.5 26.7 21.6 22.8
Buffalo 1170111 21.425.320.4 20.9 21.0 24.3 19.7 20.6
Memphis 1135614 25.826.426.7 25.5 25.0 26.0 25.5 24.6
Jacksonville 1100491 27.029.3 30.0 25.8 26.5 28.6 28.7 25.5
Rochester 1098201 21.622.923.7 21.1 21.4 22.3 23.1 21.0
Grand Rapids 1088514 22.1 23.1 22.9 21.9 21.7 22.8 23.1 21.5
Oklahoma City 1083346 22.6 25.9 24.2 21.9 22.3 25.5 23.6 21.6
Louisville 1025598 24.425.625.7 23.9 23.7 24.9 24.9 23.2

120
Table 5-2. Continued
MSA Population All modes Drive alone mode
name size Metro Core Sub- Dis- Metro Core Sub- Dis-
center centers persed center centers persed
half to 1 millions 22.923.622.422.722.323.2 22.022.2
Richmond 996512 25.726.724.9 25.3 25.0 26.1 24.2 24.7
Greenville 962441 23.123.621.8 23.1 22.6 23.7 21.5 22.5
Dayton 950558 21.623.920.4 20.9 21.3 23.7 20.7 20.6
Fresno 922516 22.522.820.2 22.7 21.1 21.7 18.9 21.3
Birmingham 921106 28.128.528.0 27.9 27.4 27.9 27.2 27.1
Albany 875583 23.526.922.0 22.8 23.3 26.6 22.4 22.6
Tucson 843746 23.723.222.3 24.1 22.9 22.9 21.4 23.2
Tulsa 803235 22.723.223.222.522.222.7 22.821.8
Syracuse 732117 21.022.325.720.220.922.0 24.820.1
Omaha 716998 20.522.120.9 19.8 19.9 21.8 20.2 19.2
Albuquerque 712738 22.923.5 22.6 22.2 22.9 21.9
Knoxville 687249 24.724.424.0 24.8 24.3 24.6 23.6 24.3
El Paso 679622 22.624.0 22.421.722.7 21.6
Bakersfield 661645 22.821.1 23.521.019.7 21.6
Allentown 637958 21.321.723.3 21.2 21.3 21.9 22.7 21.2
Harrisburg 629401 23.927.6 23.423.526.4 23.2
Scranton 624776 20.220.320.5 20.1 20.2 20.6 20.6 20.1
Toledo 618203 20.922.218.4 20.9 20.7 22.1 18.3 20.7
Baton Rouge 602894 26.1 26.8 25.4 26.3 25.2 26.0 24.7 25.3
Springfield 601564 20.722.619.5 20.6 21.0 22.5 22.8 20.8
Youngstown 594746 20.421.421.0 20.2 20.2 21.1 20.9 20.0
Sarasota 589959 22.423.021.7 22.3 21.8 22.3 21.7 21.8
Little Rock 583845 24.8 26.2 22.8 24.3 24.0 25.3 22.3 23.5
McAllen 569463 20.320.4 20.319.719.7 19.7
Stockton 563598 23.723.520.7 24.1 22.9 22.4 20.1 23.2
Charleston 549033 25.224.8 25.224.424.4 24.5
Wichita 545220 20.121.221.5 19.7 19.7 20.7 20.4 19.4
Mobile 540258 25.125.3 25.124.124.4 23.9
Columbia 536691 25.025.1 25.024.624.8 24.5
Colorado Spring 516929 22.1 21.8 24.2 22.1 21.7 21.3 23.1 21.7
Fort Wayne 502141 21.7 22.6 22.4 21.5 21.3 22.4 21.8 21.1
1) All commute times are calculated based on the place of work.
2) The table is sorted by the size of metro population.

121
M etro wide commute time v s. M etro population siz e
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
50.0
55.0
60.0
13.0 13.5 14.0 14.5 15.0 15.5 16.0 16.5 17.0
Ln (pop)
Min u t e
Y = -7.428 + 2.220 X
Metro wide commute time vs. Metro population size
CB D commute time v s. M etro population
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
50.0
55.0
60.0
13.0 13.5 14.0 14.5 15.0 15.5 16.0 16.5 17.0
Ln (pop)
Min u t e
Y =-58.734 + 6.065 X
CBD commute time vs. Metro population size
Y = -7.428 + 2.220 X
Figure 5-1. Mean commute time by workplace location type versus metro population size

122
Commute time outside centers (Dispersed) v s. M etro
population
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0
50.0
55.0
60.0
13.0 13.5 14.0 14.5 15.0 15.5 16.0 16.5 17.0
Ln (pop)
M inute
S ubcenter commute time v s. M etro population
10 .0
15 .0
20 .0
25 .0
30 .0
35 .0
40 .0
45 .0
50 .0
55 .0
60 .0
13.0 13.5 14.0 14.5 15.0 15.5 16.0 16.5 17.0
Ln (pop)
Mi n u t e
Subcenter commute time vs. Metro population size
Dispersed location commute time vs. Metro population size
Y =-18.063 + 2.933 X
Y = -4.613 + 2.002 X
Figure 5-1. Continued
1) All commute times are calculated based on the place of work.
2) All figures present commute times by location type defined by the GWR procedure.

123
5.3. STATISTICAL ANALYSIS: DETERMINANTS OF AVERAGE COMMUTE
TIME

5.3.1. Model Specification
As Sarzynski and his colleagues (2006) correctly suggested, the relationship
between commute time and urban spatial structure is simultaneous and there should be
considerable time lag in the spatial adjustment in response to commute time due to the
durability of built environment. While the endogenous relationship between the two can be
properly dealt with by using two stage least square (2SLS) regression estimation, the time
lag in spatial adjustment can not be addressed in a cross section analysis without spatial
structure variables in prior period.
Thus, I treat two dimensions of urban spatial structure, employment dispersion and
polycentricity, as endogenous in commute time models (equation 6-1). The first stage OLS
estimation for the two endogenous variables includes in right hand side metropolitan age
dummies, industrial structure, and municipality structure in addition to all exogenous
variables in equation 6-1. I will also present the results of OLS estimation and Durbin-Wu-
Hausman test of exogeneity. Summary statistics of included variables are shown in Table 5-
3.

Commute time = f (Dispersion, Polycentricity, P, D, ∆P, PC, TRAN, DEMO, BAY) (6-1)
, where P denotes log population size; D log population density; ∆P log population growth;
PC population centralization; TRAN transportation infrastructure variables including percent
transit commuters and freeway lane miles per 1000 persons; DEMO demographic variables
including median household income and percent multi-worker families; and BAY dummy
indicating the presence of bay.

124
Table 5-3. Descriptive statistics of variables
Variable Mean Std. Min. Max. No. of
Dev. metros
Commute times
Metro mean commute time by all modes 24.8 3.1 20.1 34.3
Metro mean commute time by drive alone mode 24.1 2.6 19.7 30.9
log mean commute time 3.175 0.107 2.982 3.430
CBD (GWR) commute time by drive alone 27.3 5.9 19.8 55.6
log CBD commute time 3.287 0.192 2.986 4.019
Main center (MD) commute time by drive alone 26.7 5.3 19.7 52.8
log main center commute time 3.269 0.176 2.980 3.967

Spatial structure
Dispersed employment share (GWR) 81.6 6.0 62.9 95.6
Subcenters' share of center employment (GWR) 38.1 25.5 0.0 91.0
CBD employment share (GWR) 10.7 4.8 2.8 28.2
Dispersed employment share (MD) 72.8 7.8 53.3 89.1
Subcenters' share of center employment (MD) 32.8 20.0 0.0 80.5
Main center employment share (MD) 17.8 6.8 6.3 46.7
Employment concentration indices factor score 0.000 1.000 -1.938 3.455
Employment centralization indices factor score 0.000 1.000 -2.376 3.130

Control variables
Log population size 14.183 0.858 13.127 16.870
Log population growth 0.142 0.110 -0.022 0.606
Log population density (per acre)
1)
0.463 0.490 -0.366 1.872
Population centralization indices factor score 0.000 1.000 -2.320 3.730
Freeway lane-miles per 1,000 population
2)
0.665 0.255 0.110 1.255
Percentage transit commuters 2.883 3.415 0.247 24.700
Median household income ($10,000) 4.313 0.595 2.486 6.202
Percentage multi-worker families 57.589 4.217 43.449 68.373
Dummy indicating the presence of bay 0.152 0.361 0.000 1.000 12

Instrument variables for urban spatial structure
Metropolitan age dummies
35 years and less 0.291 0.457 0.000 1.000 23
35 to 60 years 0.405 0.494 0.000 1.000 32
60 to 100 years (=reference) 0.266 0.445 0.000 1.000 21
More than 100 years 0.038 0.192 0.000 1.000 3
Percentage core central city population 30.156 19.095 3.913 88.763
Number of cities over pop. 10k per 100k pop. 0.946 0.425 0.288 2.252
Industrial structure factor score
Business services 0.000 1.000 -1.952 2.612
Convention and construction 0.000 1.000 -1.583 5.815
Public sectors 0.000 1.000 -2.642 3.245
Retail and services 0.000 1.000 -2.526 3.737
1) Population density is calculated after excluding extremely low density census tracts under 100
persons/sq miles.
2) Core urbanized area freeway lane-mile data are used because metro level data are not reported.

125
5.3.2. Estimation Results
Table 5-4 presents three sets of estimation results of commute time regression
models, a) base OLS model without spatial structure variable, b) OLS and 2SLS models
using spatial variables defined by the GWR procedure, and c) OLS and 2SLS models using
spatial variables defined by the MD procedure. Durbin-Wu-Hausman test rejects the null
hypothesis of exogenous spatial structure in all specifications. Thus, coefficients estimated
by OLS are potentially biased. The results show fair goodness of fit, explaining 60 to 70%
of the variation in average commute times by drive alone mode in OLS estimations.
Most control variables were significant with expected signs: Average commute time
increases with population size as expected; Population growth in a recent decade lengthened
commute time in the first two sets of estimations; Higher density was associated with shorter
commute time, but was significant in three specifications; Population decentralization had
positive effects on commute time, but was also significant in three specifications; and the
presence of bay as in San Francisco and Seattle increases average commute time by about 5
percent (about 1.2 minute in an average metropolitan area).
Both transportation variables were not significant perhaps because they were not
properly measured. Freeway density was measured for core urbanized areas while all other
variables were for metropolitan areas. Actual transit infrastructure variables should be used
instead of the proxy variable, transit use, which might have also introduced endogeneity
problem. Demographic variables, median household income and percent multi-worker
families, were significant with expected signs, but were significant only in OLS estimations.
The commute time saving effect of spatial restructuring was only partially identified.
Employment dispersion helps reduce metro-wide average commute time either by shortening

126
commute distance or by lowering congestion level. This effect was found significant only in
models using spatial measures defined by the minimum density procedure. The size of the
effect was considerable, -.0128 in the 2SLS estimation. Further dispersion of one percent of
metropolitan employment share lowers commute time by approximately 1.3% point, all else
being equal. An increase of dispersed employment share by a standard deviation (7.8%)
saves about 10% of commute time, 2.4 minutes in average size metropolitan areas.
However, the effect of polycentricity, subcenters’ share of center employment, had
expected negative sign but was not significant at 10 percent level. This result is surprising
given the big commute time differential between CBD and subcenters. I retested this point
in regression models for average commute time of CBD/main center commuters (Table 5-6).
Relative size of CBD/main center was significant in 2SLS estimations in both GWR and MD
results. Average commute time of CBD/main center workers increases with CBD/main
center employment share by non-trivial size, holding constant metropolitan population size.
That is, the bigger the CBD/main center, firms need to draw labors from farther distance and
hence increasing local congestion level.
Thus, there are two paths of potential commute time savings in spatial changes from
monocentric to polycentric structure. Both groups of workers who move workplace to
subcenters and who remain in CBD/main center can potentially benefit from the spatial
restructuring. The reasons why polycentricity is not significant in metro wide commute time
models despite the potentials can also be interpreted in two ways. First, CBD/main center
employment share may be already too small to affect metro wide average commute time.
Second, it is also possible that the spatial restructuring may have system wide effects such as
increased cross commutes, offsetting potential commute time saving.

127
Table 5-4. Mean metro commute time regression results
a) Base OLS b) GWR results c) MD results
OLS GWR 2SLS GWR OLS MD 2SLS MD
Beta t Beta t Beta t Beta t Beta t
Dispersion -0.0013 -0.77 -0.0054 -1.04 -0.0033 -2.44 ** -0.0128 -3.16***
Polycentricity -0.0006 -1.29 0.0002 0.16 -0.0003 -0.74 -0.0026 -1.29
Log population size 0.0890 5.24 *** 0.1010 5.16 *** 0.0784 2.44** 0.0834 4.64 *** 0.0874 2.48**
Log population density -0.0343 -1.43 -0.0336 -1.35 -0.0595 -1.89 * -0.0434 -1.83 * -0.0755 -2.1 **
Log population growth 0.2396 3.28 *** 0.2190 2.91 *** 0.2159 2.21 ** 0.1356 1.61 -0.2140 -1.12
Population centrality -0.0118 -1.44 -0.0170 -1.85 * -0.0162 -1.03 -0.0178 -2.14 ** -0.0347 -2.67 **
% transit commuters -0.0017 -0.47 -0.0032 -0.82 0.0014 0.26 -0.0011 -0.30 -0.0039 -0.51
Freeway 0.0709 1.86 * 0.0543 1.30 0.02670.39 0.0333 0.81 -0.1006 -1.19
Median hh income 0.0496 2.02 ** 0.0439 1.70 * 0.0260 0.73 0.0422 1.75 * 0.0252 0.74
% multi-worker families -0.0067 -2.26 ** -0.0059 -1.90 * -0.0034 -0.75 -0.0055 -1.89 * -0.0015 -0.34
Bay dummy 0.0524 2.34 ** 0.0530 2.37 ** 0.0538 2.13** 0.0454 2.07 ** 0.0253 0.82
Constant 2.0154 8.47 *** 1.9736 7.15 *** 2.5506 5.92*** 2.3501 8.36 *** 3.0691 5.73***
R-squared 0.667 0.676 0.591 0.695 0.424
Adj R-squared 0.624 0.622 0.524 0.644 0.330
DWH χ
2
test (df=2) 6.083 ** 16.292 ***
3) The dependent variable of all models is log mean commute time by drive alone mode. The number of
observations is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
4) Durbin-Wu-Hausman statistic tests the null hypothesis of exogenous spatial variables. Instrument variables
include % core central city population, the number of municipalities, metro age dummies, and industrial
structure variables.

128
Table 5-5. Mean CBD/Main center commute time regression results
GWR results OLS1 OLS2 2SLS
Beta t Beta t Beta t
CBD employment share 0.0033 1.31 0.0108 1.69 *
Log population size 0.1484 7.50 *** 0.1599 7.42 *** 0.1862 6.08 ***
Log population density -0.0343 -1.23 -0.0329 -1.18 -0.0296 -1
Log population growth 0.1096 1.29 0.0725 0.81 -0.0123 -0.11
Population centrality -0.0198 -2.08 ** -0.0262 -2.46 ** -0.0410 -2.55 **
% transit commuters 0.0165 3.94 *** 0.0149 3.41 *** 0.0111 2.01 **
Freeway 0.0787 1.77 * 0.0556 1.17 0.0028 0.04
Median hh income 0.0331 1.15 0.0266 0.92 0.0117 0.35
% multi-worker families -0.0044 -1.27 -0.0033 -0.94 -0.0009 -0.22
Bay dummy 0.0549 2.11 ** 0.0558 2.15 ** 0.0578 2.09 **
Constant 1.1839 4.27 *** 0.9772 3.08 *** 0.5053 1.02
R-squared 0.859 0.863 0.844
Adj R-squared 0.841 0.842 0.822
DWH χ
2
test (df=1) 2.214

MD results OLS1 OLS2 2SLS
Beta t Beta t Beta t
Main center employment share 0.0023 1.46 0.0142 2.34 **
Log population size 0.1451 7.65 *** 0.1498 7.84 *** 0.1740 6.14 ***
Log population density -0.0436 -1.63 -0.0465 -1.74 * -0.0612 -1.66
Log population growth 0.1125 1.38 0.0356 0.37 -0.3624 -1.57
Population centrality -0.0164 -1.80 * -0.0199 -2.13 ** -0.0378 -2.48 **
% transit commuters 0.0138 3.43 *** 0.0123 2.96 *** 0.0042 0.61
Freeway 0.0746 1.75 * 0.0486 1.06 -0.0859 -0.96
Median hh income 0.0227 0.83 0.0208 0.76 0.0106 0.28
% multi-worker families -0.0038 -1.14 -0.0028 -0.84 0.0021 0.41
Bay dummy 0.0571 2.28 ** 0.0556 2.24 ** 0.0479 1.42
Constant 1.2356 4.65 *** 1.1166 4.04 *** 0.5007 1.05
R-squared 0.846 0.850 0.725
Adj R-squared 0.825 0.828 0.685
DWH χ
2
test (df=1) 9.394 ***
1) The dependent variable of GWR result models is log CBD mean commute time by drive alone mode; while
the dependent variable of MD result models is log Main center mean commute time by drive alone mode.
The number of observations is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Durbin-Wu-Hausman statistic tests the null hypothesis of exogenous spatial variables. Instrument variables
include % core central city population, the number of municipalities, metro age dummies, and industrial
structure variables.

129
5.4. DISCUSSION
This chapter presented a study on the commuting impacts of metropolitan level
spatial structure in an attempt to empirically test the prediction from urban economic
theories that spatial evolution from monocentric to polycentric and dispersed structure
reduces average commute time.
Descriptive analysis identified a large potential of commute time saving from the
spatial adjustment. There was a large gap in commute time between CBD workers and those
who work at subcenters and dispersed places particularly in large metropolitan areas. The
edges in commuting economies will be realized to the extent that much of agglomeration
economies become available outside the CBD.
Notwithstanding with the potential, I found only partial commute time saving effects
of spatial adjustment. While employment dispersion effect was found significant in the MD
results, polycentricity effect was not significant. The insignificance of polycentric spatial
dimension in explaining metro wide average commute time variation despite the potentials
can be interpreted in three ways. First, CBD/main center employment share may be already
too small to affect metro wide average commute time. Second, it is also possible that the
spatial restructuring may have system wide effects such as increased cross commutes,
offsetting potential commute time savings. Finally, as discussed in chapter 4, each
metropolitan area may have developed its own development path reflecting unique history,
transportation networks and industrial structure in a way that checks commute time increase
entailed by metropolitan growth.

130
CHAPTER 6.
URBAN SPATIAL STRUCTURE AND METROPOLITAN GROWTH

6.1. INTRODUCTION
Does urban spatial structure influence economic performance or growth in
metropolitan areas? Is one type of spatial form more efficient than another? Or is an
efficient spatial structure contingent on the size and other contexts of each metropolitan
area? To address these unanswered questions, this chapter investigates the impacts of urban
spatial structure on economic growth in a cross section of 79 US metropolitan areas. In
particular, it probes how the links between spatial structure and metropolitan growth depend
on metropolitan size.
Considerable evidence has been collected on the existence and extent of
agglomeration economies (for surveys of the literature, see Moomaw 1983; Gerking 1994).
In general, firms in large cities or regions have higher productivity because agglomeration
economies (see the footnote in chapter 2 for the various types of agglomeration economies)
lower production costs or facilitate competitive innovation. These productivity edges of
larger cities are found to be larger in non-manufacturing sectors (Moomaw 1981). Whereas
manufacturing sectors tend to benefit more from specialization and localization economies
(Henderson 1986; Moomaw 1988; Henderson, Kuncoro, and Turner 1995), urban diversity
and localized competition are found to contribute to more innovation and growth in more

131
knowledge intensive sectors (Glaeser et al. 1992; Harrison, Kelley, and Gant 1996; Feldman
and Audretsch 1999; Combes 2000).
Not only firms but also individuals tend to have more chance to learn and acquire
skills in urban agglomeration that ensure them higher returns (Glaeser 1999). Moreover,
consumers in larger cities enjoy a variety of specialized goods and services, and cultural and
entertainment opportunities (Clark, Kahn, and Ofek 1988). High end restaurants and
Broadway shows in Manhattan and professional sports teams in big cities are good examples.
However, urban growth is not without costs. Firms and households in large cities
suffer from negative externalities such as congestion, pollution, and higher crime rate.
Productivity advantages of larger cities should be exploited at least partially by higher rents
and negative externalities should be compensated by higher rents.
The balance between these agglomeration economies and diseconomies has
provided a rationale for optimal city size theories (Carlino 1987). Optimal city size is
derived as the maximal point of net agglomeration economies that are supposed to be
inverted U-shaped function of city size (Begovic 1991), while earlier studies attempted to
find the minimal point of U-shaped cost curve of urban public services (Clark 1945; Hirsch
1959).
However, the notion of optimal city size was heavily criticized by the fact that there
coexist cities of a wide range of sizes in a national urban hierarchy, through which
specialized goods and services are delivered and innovations and other economic functions
are channeled across cities in the urban system (Richardson 1972). Indeed, relative size
distribution of cities has been remarkably stable over time in most countries (Eaton and
Eckstein 1997; Black and Henderson 1999, 2003; Nitsch 2003), which has been known to

132
obey Zipf’s law or rank size rule (Gabaix 1999). Although economic reasons for the
empirical regularity are still murky, it implies parallel urban growth patterns in relation to
city size rather than convergence to a certain optimal city size (Barnard and Krautmann
1988; Eaton and Eckstein 1997; Glaeser, Scheinkman, and Shleifer 1995)
6
.
Within the hierarchical urban system, each city at different ranks takes on different
economic functions, with varying ‘efficient sizes’ (Richardson 1972; Capello and Camagni
2000). Hence, the prospective growth of a city depends on “its ability to move to ever
higher urban ranks, developing or attracting new and superior functions (Camagni, Diappi,
and Leonardi 1986)”. Empirically, upward mobility in the US urban hierarchy for the last
century was found to be promoted by better geographical amenities – climate and coastal
location – and higher market potential (Black and Henderson 1999, 2003).
The links between urban form and growth should be understood in this context of
urban evolution. To the extent that a city can readjust its spatial form such that it mitigates
negative externalities of city size, it can afford continued growth or higher probability of
upward mobility in the urban hierarchy. If this hypothesis holds, we will observe that
efficient urban spatial structure may vary across different stages of urban development.
After all, this proposition is consistent with urban economic models introduced in chapter 3
in which a city transform from monocentric to polycentric or more dispersed structure.
Prud’homme and Lee (1999), in the same logic, suggested that good city
management – transportation and land use policy – can shift marginal benefits of city size
upward and costs downward, increasing the efficient city size. Cervero (2001) also

6
Wheeler (2003) found an ‘inverted U-shaped’ growth pattern in a cross-section of US
counties, but not in metropolitan level data in the 1980s. I interpret the county level results as the
growth with limited geographical boundary.

133
contended that more compact, centralized and accessible cities have higher productivity and
presented empirical analyses at both inter- and intra metropolitan levels. But, both level
analyses are seriously flawed in supporting his ideas. First, the results with labor
productivity without controlling capital input may be misleading. Second, productivity
comparisons are generally done within an industrial sector, with more disaggregated data
giving better estimates (Rigby and Essletzbichler 2002). Otherwise, industry mix should be
controlled more completely. Finally, given his assumption on the same labor productivity of
a SIC sector within California, his results – higher labor productivity in sub-metro areas with
larger labor market shed and higher accessibility between residence and firms – actually
mean that industrial sectors with higher productivity tends to locate those areas.
These problems are largely due to the limited economic output data at metropolitan
and sub-metropolitan area level. This is one reason why I examine the variation in the
growth rate instead of productivity across metropolitan areas with different spatial forms and
varying sizes. Moreover, a productivity study can give only information about
(dis)economies in production sphere although consumer externalities are also important
factors in urban prospect (Glaeser, Kolko, and Saiz 2001; Gabriel and Rosenthal 2004).
Thus, I believe urban growth is a more complete market test.
Empirical modeling strategy is shortly discussed in the next section, followed by the
presentation of estimation results and discussion.

6.2. MODEL SPECIFICATION
There are two empirical modeling frameworks available in the literature that fit into
the purpose of this chapter, a supply side urban economic growth model by Glaeser and his

134
collegues (Glaeser 2000; Glaeser, Scheinkman, and Shleifer 1995; Glaeser and Shapiro
2003) and population and employment adjustment models (Steinnes and Fisher 1974;
Carlino and Mills 1987; Clark and Murphy 1996; Mulligan, Vias, and Glavac 1999). A key
difference is that the former assumes a spatial equilibrium while population and employment
partially adjust to the equilibrium each period. Whereas both models can serve to identify
the growth effects of spatial structure controlling other determinants, it is easier to interact
spatial structure variables with metropolitan population and employment size in the latter.
Thus, I adopt the empirical framework of Glaeser (2003), in which a certain attribute
of cities contributes to growth in three ways: 1) becoming more important in the production
process; 2) attracting more consumers either by reducing the cost of living or enhancing
amenity level; or 3) increasing the technological growth rate. His empirical analysis found
that higher level of human capital, warmer and drier climate and automobile oriented
transportation system are three key factors explaining the varied growth rates among US
cities in the 1990s.
I add spatial structure variables measured in chapter 2 to his empirical model and
also include interaction terms between the spatial measures and metropolitan population
(employment) size. Both spatial structure and metropolitan size variables are centered by
subtracting mean value for the ease of interpretation. Coefficients of the centered size and
spatial measure variables are the main effects of corresponding variables when the other
variable is at the average level. Thus, β
2
is the spatial structure impacts on urban growth at
the sample average of log population/employment size (1.25 millions and .6 million). But,
the spatial structure effect depends on metropolitan size and the sign of β
3
determines
whether the effect increases or decreases with metropolitan size as in equation 5-2.

135
F N F N X
N
N
Log
t t t
t
t
) log( ) log(
1 3 2 1 1 1
1
− − −
−
+ + + =
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
β β β α
(5-1)
[ ]F N N X
t t t
) log( ) log(
1 3 2 1 1 1 − − −
+ + + = β β β α (5-2)
, where N
t
and N
t-1
denote population (employment) size in 2000 and 1990, respectively; X a
vector of metropolitan attributes listed in Table 6-1 including constant; F a vector of spatial
structure variables.

Two spatial structure variables each indicating urban dispersion and polycentricity
are used in the model estimations. The coefficient of dispersion, measured as the share of
metropolitan employment dispersed outside centers, will show whether clustered or
dispersed urban form is more amenable to urban growth. After controlling the degree of
employment dispersion, it will be tested whether monocentric or polycentric structure is
associated with faster growth. The two spatial indicators are measured as of 2000 due to
data limitation assuming that there’s been little change in urban spatial structure in the ten
year period.
All other explanatory variables are measured as of 1990. These include log
population density, industrial mix, local amenities, and human capital and other
demographic variables (Table 6-1). While density is suggested to have productivity effects
by lowering transportation costs, positive externalities and specialization (Ciccone and Hall
1996), it tradeoffs congestion (Carlino and Chatterjee 2002). Human capital accumulation is
increasingly emphasized in urban economic growth literature (Simon and Nardinelli 2002;
Glaeser and Shapiro 2003; Shapiro 2006).
Main data sources are Population Census; Regional Economic Information System
(REIS 1969-2000) published by the Bureau of Economic Analysis; and 1994 County and
City Data Book. Descriptive statistics are shown in Table 6-2.

136
Table 6-1. Definition of vairables
Variables Descriptions
Dependent variables
Log population growth
Log employment growth

Log(pop2000/pop1990)
Log(emp2000/emp1990)
Metropolitan size and
spatial structure
Log population
Log employment
Log population density

Dispersion
Polycentricity

Log(population 1990)
Log(employment 1990)
Log(population 1990/acre), measured for 95% population area
and excluding extremely low density census tracts under 100
persons per square miles

Percent dispersed location share of total employment
Subcenters’ share of center employment: subcenter’ emp. /
(subcenters’ emp. + CBD/maincenter emp.) * 100
Industrial mix
Percent manufacturing

Percent manufacturing’s share of total earnings
Human capital and
demographic variables
Percent college
Percent poverty
Percent nonwhite
Percent immigrants
Percent pop over 64

Percent of 25+ years persons with bachelor’s degree or higher
Percent persons with income below poverty level
Percent nonwhite population
Percent foreign-born population
Percent population over 64 years
Amenities
Mean Jan. temperature
Annual precipitation
Violent crime rate
Coastal location

January mean of average daily temperature (F°) for 1961-1990
Average annual precipitation for 1961-1990
Violent crimes known to police per 100,000 population
Dummy variable indicating coastal location
Regions
Eight dummies of Census
Division

New England, Middle Atlantic, East North Central, West
North Central, South Atlantic, East South Central, West South
Central, Mountain (Reference= Pacific)
1) All explanatory variables are measured as of 1990 except dispersion and polycentricity.
2) Population data are from Population Census; employment data are from Regional Economic
Information System (REIS 1969-2000) published by the Bureau of Economic Analysis, US
Department of Commerce; all other demographic and amenity variables are from 1994 County
and City Data Book.

137
Table 6-2. Descriptive statistics of variables
No. Mean Std. Min. Max. No.
obs. dev. Metros
Log population growth
79 0.1423 0.1099 -0.0217 0.6061
Log employment growth
79 0.2366 0.1239 -0.0015 0.6157
Population 1990 (thousands) 79 2,015 2,938 384 19,550
Employment 1990 (thousands) 79 1,003 1,465 102 9,407
Population density 1990 (per acre) 79 1.80 1.02 0.69 6.50

GWR results
Dispersion (%, dispersed emp. share) 79 81.59 6.05 62.86 95.59
Polycentricity (%, subcenters / center)
79 38.07 25.49 0.00 91.04

MD results
Dispersion (%, dispersed emp. share) 79 72.78 7.79 53.29 89.06
Polycentricity (%, subcenters / center)
79 32.85 19.98 0.00 80.47

Percent manufacturing' share 79 18.69 7.75 3.29 36.21
Percent nonwhite
79 19.16 9.04 1.65 41.95
Percent immigrants
79 6.21 5.89 1.12 27.17
Percent pop over 64
79 12.05 3.35 7.38 30.37
Percent college
79 21.04 4.41 11.50 31.71
Percent poverty
79 12.55 4.89 7.05 41.88

Mean Jan. temperature 79 36.79 12.27 11.80 67.20
Annual precipitation
79 36.23 14.33 4.13 63.96
Violent crime rate
79 814 338 172 2190
Coastal location
79 0.34 0.48 0 1 27

New England 79 0.05 0.22 0 1 4
Middle Atlantic 79 0.13 0.33 0 1 10
East North Central 79 0.15 0.36 0 1 12
West North Central 79 0.06 0.25 0 1 5
South Atlantic 79 0.19 0.39 0 1 15
East South Central 79 0.08 0.27 0 1 6
West South Central 79 0.14 0.35 0 1 11
Mountain 79 0.09 0.29 0 1 7
Pacific 79 0.11 0.32 0 1 9

138
6.3. ESTIMATION RESULTS
Table 6-3 show OLS regression results without spatial structure variables. All
significant variables had expected signs. Metros with more industrial base in manufacturing
experienced slower growth. Metros with larger nonwhite and aged segments of population
also grew slowly. Lower density metros grew faster. Mean January temperature was highly
significant confirming the economic shift from the frostbelt to sunbelt regions; whereas other
amenities were not significant. It should be noted that the dependent variable is not metro
size but the change in size. The (dis)amenities of these insignificant variables might have
been already reflected in the population/employment distributions in the beginning period to
the extent that the distributions approximate the equilibrium status rather than that these
variables don’t affect firms’ productivity and/or households’ utility level.
Some other variables were significant in either population or employment growth
model, but not in the other. Metros with larger proportion of foreign born population
attracted more people but not jobs. This perhaps presents the role of gateway metros. On
contrary, the coefficient of percent college graduates was significantly positive only in
employment growth models. Metropolitan size was not significant in both set of models,
implying parallel growth patterns. It is interesting to see that regional effects after
controlling all these variables are somewhat different from the regional growth patterns that
we observe. The most consistently fast growing region was Mountain division. Full
specifications of models 3 and 6 explained about 70 and 80 percent of cross-sectional
variation in metropolitan population and employment growth, respectively.
In Tables 6-4 and 6-5, I show a series of regression results that examine the effects
of spatial structure variables generated by the GWR procedure after dropping some

139
insignificant control variables from Table 6-3. Model 6 in each table is the final model
which I refer to in interpretation. Both population and employment models show moderate
increase in explanatory power compared with estimations without spatial structure variables
in Table 6-3. Including spatial structure variables did not change the sign and significance
of most control variables except that log population size became significant with positive
sign and percentage manufacturing’s share turned out insignificant in population models.
Estimations with spatial structure variables generated by the MD procedure (Table 6-6 and
6-7) also present similar results for control variables.
Turning to the key variables of main interest, global coefficients of neither
dispersion (dispersed employment share) nor polycentricity (subcenters’ share of center
employment) was not significant when both spatial dimensions were included (model 6 in
each table). However, the interaction term of employment dispersion was significant with
expected positive sign in the estimations with GWR results although this was not significant
in the estimations with MD results. Interaction term of polycentricity was not significant in
any estimated models.
The findings are significant. First, whether employment location is more clustered
or dispersed matters more with respect to metropolitan growth than whether the clustering
occurs in the CBD or in subcenters. Second, consistently with hypothesis put forward in the
introduction, growth effects of spatial form (dispersion) do depend on metropolitan size. A
metropolitan area with more clustered spatial form grows faster when it is small; whereas
more dispersion leads to higher growth rate as it grows large.
In Table 6-8, to help understand the second finding, I present varying growth effects
of spatial form at different metropolitan sizes using estimated coefficients from GWR results

140
(Table 6-4 and 6-5). As explained above, the coefficient of dispersion variable (-0.0009 in
population model 6 and -0.0005 in employment model 6), though not significant and close to
zero, is its growth impact when log metropolitan size is about sample mean (1.25 million
population and .6 million employment).
When population is about half the average size, the coefficient changes is -0.0069,
meaning that the dispersion of additional one percent of metro employment leads to
approximately 0.7% slower growth rate. 1 standard deviation increase of dispersed
employment share (about 6% point) implies about 4.2% point slower growth. On contrary,
1% point and standard deviation increase of dispersed employment lead to about .5% and
3% higher growth rate, respectively, when population size is about 3.4 million. Given that
average log population growth was about 14% in the 1990s, these growth effects of
employment dispersion are substantial. A similar pattern is also found in employment
growth, but in a little smaller magnitude.
I also calibrated the growth effects of polycentricity at different employment sizes
although none of the coefficients was significant. Overall, the effects were very small
except that monocentric structure (after controlling dispersion) was more amenable to
population growth to a considerable extent in small metropolitan areas with about 1 million
residents or less.

141
Table 6-3. Metropolitan growth models without urban spatial structure variables
Dependent variable: Log population growth 1990-2000 Dependent variable: Log employment growth 1990-2000
Model 1 Model 2 Model 3 (stepwise) Model 4 Model 5 Model 6 (stepwise)
Beta t Beta t Beta t Beta t Beta t Beta t
log pop (emp) 0.0192 1.33 0.0269 1.65 0.0160 1.54 0.0059 0.45 0.0086 0.59
log pop density -0.0504 -1.65 -0.0841 -2.32 ** -0.0485 -1.63 -0.0510 -1.48 -0.0547 -3.06 ***
% manufacturing -0.0028 -1.63 -0.0033 -2.09 ** -0.0046 -2.71 *** -0.0050 -3.28 *** -0.0043 -3.29 ***
% nonwhite -0.0029 -1.70 * -0.0037 -2.78 *** -0.0057 -3.47 *** -0.0063 -4.88 ***
% immigrants 0.0062 1.90 * -0.0005 -0.17
% pop over 64 -0.0095 -2.48 ** -0.0131 -4.19 *** -0.0089 -2.41 ** -0.0070 -2.15 **
% pop college -0.0006 -0.24 0.0054 2.04 ** 0.0059 2.61 **
% pop poverty -0.0034 -1.00 0.0016 0.47 0.0033 1.52
Mean Jan. temerature 0.0052 3.29 *** 0.0058 6.83 *** 0.0065 4.31 *** 0.0046 4.69 ***
annual precipitation 0.0000 0.01 -0.0010 -0.74
violent crime rate -0.0001 -1.39 0.0000 -1.14
coastal location -0.0342 -1.53 -0.0582 -2.98 *** -0.0109 -0.51
New England -0.1305 -2.70 *** -0.0249 -0.40 -0.1463 -3.00 *** -0.0539 -0.90 -0.1677 -4.61 ***
Middle Atlantic -0.1531 -3.87 *** -0.0286 -0.51 -0.1306 -3.27 *** -0.0171 -0.32 -0.1211 -4.64 ***
East North Central -0.0827 -2.04 ** 0.0729 1.37 -0.0110 -0.27 0.1261 2.46 **
West North Central -0.0544 -1.21 0.0785 1.43 -0.0085 -0.19 0.0847 1.60
South Atlantic 0.0057 0.16 0.0732 1.67 0.0527 2.45 ** 0.0366 1.03 0.0844 2.00 *
East South Central -0.0543 -1.19 0.0663 1.16 0.0218 0.47 0.1251 2.26 **
West South Central 0.0146 0.40 0.0545 1.20 0.0706 1.92 * 0.0832 1.89 *
Mountain 0.1463 3.62 *** 0.1944 4.11 *** 0.1294 4.51 *** 0.1973 4.89 *** 0.1721 3.80 *** 0.1279 4.29 ***
constant -0.0229 -0.12 -0.1388 -0.62 -0.0665 -0.44 0.2578 1.60 0.0863 0.44 0.2286 2.31 **
R sq. 0.570 0.724 0.659 0.655 0.800 0.762
Adj. R sq. 0.499 0.629 0.625 0.598 0.731 0.727
3) The number of observations of all models is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
4) Reference region for the census division dummies is Pacific.

142
Table 6-4. Metropolitan population growth models with urban spatial structure variables (GWR results)
model 1 model 2 model 3 model 4 model 5 model 6
Beta t Beta t Beta t Beta t Beta t Beta t
Dispersion -0.0006 -0.36 -0.0014 -0.82 -0.0003 -0.19 -0.0009 -0.53
Polycentric -0.0005 -1.04 -0.0006 -1.27 -0.0007 -1.53 -0.0006 -1.08
Dispersion * log pop. 0.0049 2.75 *** 0.0060 2.14 **
Polycentric * log pop. -0.0008 -1.62 0.0003 0.41
log pop. centered 0.0294 1.78 * 0.0379 2.14 ** 0.0379 2.14 ** 0.0315 2.01 ** 0.0469 2.56 ** 0.0380 2.07 **
log pop. density -0.0926 -2.77 *** -0.0881 -2.64 ** -0.0882 -2.63 ** -0.0977 -3.07 *** -0.0892 -2.71 *** -0.0934 -2.93 ***
% manufacturing -0.0034 -2.15 ** -0.0031 -1.99 * -0.0028 -1.68 * -0.0015 -0.88 -0.0025 -1.56 -0.0006 -0.37
% nonwhite -0.0037 -2.26 ** -0.0037 -2.29 ** -0.0039 -2.38 ** -0.0035 -2.22 ** -0.0035 -2.18 ** -0.0037 -2.36 **
% immigrants 0.0044 1.52 0.0043 1.49 0.0045 1.57 0.0076 2.53 ** 0.0055 1.88 * 0.0078 2.62 **
% pop over 64 -0.0099 -2.48 ** -0.0089 -2.23 ** -0.0092 -2.30 ** -0.0118 -3.08 *** -0.0093 -2.37 ** -0.0112 -2.89 ***
% pop college 0.0001 0.03 0.0001 0.02 -0.0001 -0.05 0.0007 0.27 0.0005 0.21 0.0004 0.17
mean Jan. temerature 0.0050 3.10 *** 0.0047 2.88 *** 0.0045 2.76 *** 0.0058 3.71 *** 0.0051 3.16 *** 0.0053 3.32 ***
annual precipitation -0.0003 -0.24 -0.0005 -0.38 -0.0005 -0.32 0.0002 0.13 -0.0006 -0.43 0.0002 0.12
violent crime rate -0.0001 -1.48 -0.0001 -1.35 0.0000 -1.24 -0.0001 -1.52 0.0000 -1.30 0.0000 -1.27
New England -0.0312 -0.49 -0.0544 -0.83 -0.0547 -0.83 -0.0419 -0.69 -0.0548 -0.84 -0.0703 -1.11
Middle Atlantic -0.0462 -0.82 -0.0653 -1.13 -0.0659 -1.14 -0.0390 -0.73 -0.0607 -1.06 -0.0615 -1.11
East North Central 0.0507 0.98 0.0381 0.73 0.0350 0.66 0.0526 1.08 0.0471 0.91 0.0323 0.64
West North Central 0.0701 1.33 0.0629 1.19 0.0572 1.07 0.0699 1.39 0.0751 1.42 0.0514 0.99
South Atlantic 0.0706 1.57 0.0713 1.60 0.0759 1.69 * 0.0598 1.40 0.0749 1.70 * 0.0615 1.40
East South Central 0.0512 0.92 0.0507 0.92 0.0457 0.82 0.0425 0.80 0.0533 0.98 0.0343 0.64
West South Central 0.0380 0.95 0.0433 1.08 0.0436 1.09 0.0252 0.66 0.0416 1.05 0.0295 0.77
Mountain 0.1873 4.13 *** 0.1772 3.93 *** 0.1809 3.98 *** 0.2081 4.76 *** 0.1828 4.10 *** 0.2024 4.62 ***
constant 0.2397 2.14 **0.2415 2.17 ** 0.2447 2.19 ** 0.1528 1.38 0.2026 1.81 * 0.1552 1.40
R sq. 0.708 0.713 0.716 0.742 0.725 0.752
Adj. R sq. 0.614 0.620 0.618 0.653 0.630 0.655
1) The number of observations of all models is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Log population size and two spatial structure variables (dispersion and polycentricity) that are used in interaction terms are centered by subtracting the
mean value for the purpose of ease in interpretation.

143
Table 6-5. Metropolitan employment growth models with urban spatial structure variables (GWR results)
model 1 model 2 model 3 model 4 model 5 model 6
Beta t Beta t Beta t Beta t Beta t Beta t
Dispersion -0.0009 -0.61 -0.0007 -0.45 -0.0007 -0.47 -0.0005 -0.30
Polycentricity 0.0002 0.51 0.0001 0.31 0.0000 -0.02 0.0001 0.25
Dispersion * log emp. 0.0044 2.69 *** 0.0047 1.96 *
Polycentric * log emp. -0.0007 -1.62 0.0001 0.17
log emp. centered 0.0052 0.36 0.0040 0.27 0.0037 0.24 0.0071 0.53 0.0121 0.77 0.0052 0.33
log pop. density -0.0612 -1.97 * -0.0628 -2.00 ** -0.0626 -1.98 * -0.0678 -2.29 ** -0.0668 -2.15 ** -0.0684 -2.25 **
% manufacturing -0.0051 -3.43 *** -0.0055 -3.65 *** -0.0053 -3.34 *** -0.0035 -2.26 ** -0.0050 -3.33 *** -0.0036 -2.16 **
% nonwhite -0.0056 -3.64 *** -0.0054 -3.57 *** -0.0055 -3.57 *** -0.0053 -3.63 *** -0.0051 -3.37 *** -0.0053 -3.54 ***
% immigrants 0.0003 0.11 0.0001 0.04 0.0003 0.09 0.0036 1.26 0.0015 0.54 0.0035 1.22
% pop over 64 -0.0095 -2.54 ** -0.0095 -2.53 ** -0.0096 -2.54 ** -0.0111 -3.10 *** -0.0097 -2.64 ** -0.0113 -3.05 ***
% pop college 0.0048 1.95 * 0.0049 2.00 * 0.0048 1.95 * 0.0052 2.23 ** 0.0051 2.13 ** 0.0052 2.19 **
mean Jan. temerature 0.0066 4.38 *** 0.0068 4.42 *** 0.0067 4.31 *** 0.0072 4.98 *** 0.0072 4.69 *** 0.0072 4.80 ***
annual precipitation -0.0010 -0.75 -0.0010 -0.77 -0.0010 -0.73 -0.0005 -0.36 -0.0010 -0.77 -0.0004 -0.32
violent crime rate 0.0000 -1.27 0.0000 -1.37 0.0000 -1.29 0.0000 -1.37 0.0000 -1.43 0.0000 -1.36
New England -0.0473 -0.79 -0.0419 -0.68 -0.0419 -0.67 -0.0571 -1.01 -0.0425 -0.70 -0.0546 -0.91
Middle Atlantic -0.0135 -0.26 -0.0092 -0.17 -0.0092 -0.17 -0.0073 -0.15 -0.0063 -0.12 -0.0049 -0.09
East North Central 0.1326 2.76 *** 0.1374 2.79 *** 0.1360 2.74 *** 0.1357 2.96 *** 0.1458 2.98 *** 0.1367 2.84 ***
West North Central 0.0936 1.90 * 0.0996 2.00 * 0.0966 1.91 * 0.0974 2.07 ** 0.1118 2.24 ** 0.0978 1.98 *
South Atlantic 0.0876 2.08 ** 0.0840 2.00 ** 0.0864 2.03 ** 0.0783 1.95 * 0.0855 2.06 ** 0.0767 1.85 *
East South Central 0.1319 2.53 ** 0.1356 2.60 ** 0.1331 2.52 ** 0.1252 2.52 ** 0.1353 2.63 ** 0.1258 2.48 **
West South Central 0.0947 2.53 ** 0.0934 2.48 ** 0.0935 2.47 ** 0.0860 2.41 ** 0.0922 2.48 ** 0.0849 2.32 **
Mountain 0.1856 4.39 *** 0.1854 4.36 *** 0.1872 4.36 *** 0.2104 5.09 *** 0.1933 4.58 *** 0.2115 5.00 ***
constant 0.2316 2.20 **0.2303 2.19 ** 0.2310 2.18 ** 0.1553 1.50 0.1972 1.87 * 0.1552 1.47
R sq. 0.800 0.799 0.800 0.822 0.808 0.822
Adj. R sq. 0.735 0.735 0.731 0.760 0.742 0.752
1) The number of observations of all models is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Log employment size and two spatial structure variables (dispersion and polycentricity) that are used in interaction terms are centered by subtracting the
mean value for the purpose of ease in interpretation.

144
Table 6-6. Metropolitan population growth models with urban spatial structure variables (MD results)
model 1 model 2 model 3 model 4 model 5 model 6
Beta t Beta t Beta t Beta t Beta t Beta t
Dispersion -0.0018 -1.19 -0.0022 -1.38 -0.0015 -0.96 -0.0023 -1.41
Polycentricity -0.0003 -0.68 -0.0005 -0.97 -0.0005 -1.03 -0.0006 -1.13
Dispersion * log pop. 0.0026 1.66 0.0012 0.63
Polycentric * log pop. -0.0009 -1.69 * -0.0008 -1.34
log pop. centered 0.0224 1.29 0.0336 1.99 * 0.0257 1.45 0.0234 1.36 0.0371 2.21 ** 0.0276 1.58
log pop. density -0.0887 -2.67 *** -0.0919 -2.76 *** -0.0880 -2.65 ** -0.0893 -2.73 *** -0.0957 -2.91 *** -0.0913 -2.79 ***
% manufacturing -0.0029 -1.83 * -0.0032 -2.02 ** -0.0024 -1.44 -0.0022 -1.37 -0.0028 -1.75 * -0.0017 -0.98
% nonwhite -0.0036 -2.26 ** -0.0034 -2.07 ** -0.0033 -2.04 ** -0.0040 -2.49 ** -0.0033 -2.04 ** -0.0034 -2.08 **
% immigrants 0.0052 1.76 * 0.0040 1.36 0.0049 1.65 0.0065 2.15 ** 0.0045 1.54 0.0061 2.01 **
% pop over 64 -0.0095 -2.45 ** -0.0092 -2.30 ** -0.0088 -2.22 ** -0.0102 -2.65 ** -0.0095 -2.42 ** -0.0094 -2.40 **
% pop college -0.0004 -0.15 0.0002 0.07 -0.0004 -0.16 -0.0002 -0.09 0.0003 0.13 -0.0003 -0.12
mean Jan. temerature 0.0045 2.76 *** 0.0050 3.14 *** 0.0045 2.75 *** 0.0046 2.85 *** 0.0053 3.33 *** 0.0047 2.90 ***
annual precipitation -0.0003 -0.24 -0.0003 -0.25 -0.0003 -0.20 -0.0001 -0.08 -0.0008 -0.55 -0.0006 -0.40
violent crime rate -0.0001 -1.42 -0.0001 -1.63 -0.0001 -1.61 -0.0001 -1.49 -0.0001 -1.42 -0.0001 -1.40
New England -0.0194 -0.30 -0.0453 -0.69 -0.0334 -0.51 -0.0452 -0.70 -0.0508 -0.79 -0.0463 -0.70
Middle Atlantic -0.0378 -0.67 -0.0590 -1.02 -0.0516 -0.89 -0.0526 -0.94 -0.0584 -1.02 -0.0546 -0.95
East North Central 0.0538 1.05 0.0430 0.82 0.0438 0.84 0.0373 0.73 0.0430 0.83 0.0378 0.73
West North Central 0.0645 1.23 0.0698 1.32 0.0609 1.16 0.0500 0.95 0.0742 1.43 0.0576 1.09
South Atlantic 0.0777 1.73 * 0.0633 1.39 0.0710 1.56 0.0703 1.58 0.0698 1.55 0.0757 1.67
East South Central 0.0506 0.92 0.0461 0.82 0.0402 0.72 0.0391 0.71 0.0536 0.96 0.0424 0.76
West South Central 0.0378 0.95 0.0370 0.93 0.0354 0.89 0.0294 0.75 0.0364 0.93 0.0312 0.79
Mountain 0.1817 4.09 *** 0.1801 3.98 *** 0.1743 3.86 *** 0.1742 3.96 *** 0.1773 3.98 *** 0.1685 3.79 ***
constant 0.2403 2.17 **0.2322 2.07 ** 0.2311 2.08 ** 0.2378 2.18 ** 0.2248 2.04 ** 0.2242 2.05 **
R sq. 0.714 0.710 0.719 0.727 0.723 0.738
Adj. R sq. 0.622 0.616 0.622 0.633 0.628 0.635
1) The number of observations of all models is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Log employment size and two spatial structure variables (dispersion and polycentricity) that are used in interaction terms are centered by subtracting the
mean value for the purpose of ease in interpretation.

145
Table 6-7. Metropolitan employment growth models with urban spatial structure variables (MD results)
model 1 model 2 model 3 model 4 model 5 model 6
Beta t Beta t Beta t Beta t Beta t Beta t
Dispersion 0.0001 0.07 0.0002 0.11 0.0003 0.18 0.0002 0.15
Polycentricity 0.0001 0.17 0.0001 0.19 0.0000 -0.07 0.0001 0.19
Dispersion * log emp. 0.0017 1.20 0.0014 0.84
Polycentric * log emp. -0.0005 -1.05 -0.0003 -0.50
log emp. centered 0.0073 0.48 0.0062 0.43 0.0069 0.43 0.0072 0.47 0.0073 0.50 0.0064 0.41
log pop. density -0.0608 -1.94 * -0.0607 -1.95 * -0.0610 -1.93 * -0.0606 -1.94 * -0.0629 -2.02 ** -0.0620 -1.96 *
% manufacturing -0.0053 -3.44 *** -0.0054 -3.54 *** -0.0054 -3.31 *** -0.0049 -3.08 *** -0.0051 -3.35 *** -0.0050 -2.95 ***
% nonwhite -0.0055 -3.57 *** -0.0055 -3.54 *** -0.0055 -3.51 *** -0.0057 -3.72 *** -0.0055 -3.52 *** -0.0058 -3.58 ***
% immigrants 0.0001 0.03 0.0002 0.07 0.0001 0.05 0.0012 0.40 0.0007 0.25 0.0014 0.46
% pop over 64 -0.0091 -2.47 ** -0.0092 -2.46 ** -0.0092 -2.44 ** -0.0096 -2.60 ** -0.0095 -2.53 ** -0.0099 -2.59 **
% pop college 0.0049 1.97 * 0.0048 1.98 * 0.0049 1.96 * 0.0048 1.97 * 0.0049 2.02 ** 0.0049 1.96 *
mean Jan. temerature 0.0066 4.28 *** 0.0066 4.39 *** 0.0066 4.24 *** 0.0066 4.25 *** 0.0067 4.45 *** 0.0066 4.21 ***
annual precipitation -0.0011 -0.81 -0.0011 -0.82 -0.0011 -0.82 -0.0009 -0.69 -0.0013 -0.98 -0.0011 -0.79
violent crime rate 0.0000 -1.31 0.0000 -1.23 0.0000 -1.22 0.0000 -1.34 0.0000 -1.09 0.0000 -1.12
New England -0.0522 -0.86 -0.0487 -0.79 -0.0496 -0.79 -0.0701 -1.13 -0.0527 -0.86 -0.0636 -0.99
Middle Atlantic -0.0173 -0.33 -0.0143 -0.26 -0.0149 -0.27 -0.0280 -0.52 -0.0136 -0.25 -0.0206 -0.37
East North Central 0.1318 2.73 *** 0.1337 2.71 *** 0.1336 2.69 *** 0.1197 2.43 ** 0.1343 2.73 *** 0.1257 2.47 **
West North Central 0.0959 1.93 * 0.0959 1.94 * 0.0966 1.92 * 0.0859 1.71 * 0.0998 2.02 ** 0.0907 1.76 *
South Atlantic 0.0844 1.98 * 0.0863 2.02 ** 0.0857 1.97 * 0.0806 1.89 * 0.0911 2.12 ** 0.0868 1.96 *
East South Central 0.1346 2.58 ** 0.1361 2.57 ** 0.1365 2.55 ** 0.1271 2.43 ** 0.1417 2.66 ** 0.1349 2.47 **
West South Central 0.0956 2.55 ** 0.0959 2.56 ** 0.0961 2.54 ** 0.0905 2.41 ** 0.0965 2.57 ** 0.0925 2.42 **
Mountain 0.1820 4.33 *** 0.1829 4.31 *** 0.1834 4.26 *** 0.1769 4.20 *** 0.1816 4.28 *** 0.1794 4.15 ***
constant 0.2310 2.19 **0.2325 2.20 ** 0.2329 2.18 ** 0.2358 2.25 ** 0.2281 2.16 ** 0.2358 2.20 **
R sq. 0.799 0.799 0.799 0.803 0.802 0.805
Adj. R sq. 0.734 0.734 0.729 0.736 0.734 0.728
1) The number of observations of all models is 79. * Significant at 10%; ** significant at 5%; *** significant at 1%.
2) Log employment size and two spatial structure variables (dispersion and polycentricity) that are used in interaction terms are centered by subtracting the
mean value for the purpose of ease in interpretation.

146
Table 6-8. Varying growth effects of spatial structure depending on metropolitan size
Population growth

Growth effects of
1% increase
Growth effects of
1 std. dev. increase
log population population size dispersion polycentricity dispersion polycentricity
13.04 460,632 -0.69% -0.09% -4.2% -2.2%
14.04 1,252,129 -0.09% -0.06% -0.6% -1.4%
15.04 3,403,639 0.50% -0.03% 3.0% -0.7%
16.04 9,252,049 1.10% 0.00% 6.6% 0.1%
Employment growth

Growth effects of
1% increase
Growth effects of
1 std. dev. increase
log employment employment size dispersion polycentricity dispersion polycentricity
12.30 220,261 -0.52% 0.00% -3.1% 0.0%
13.30 598,731 -0.05% 0.01% -0.3% 0.3%
14.30 1,627,519 0.42% 0.02% 2.5% 0.6%
15.30 4,424,054 0.89% 0.03% 5.4% 0.8%
1) This table is calibrated based on the coefficients of spatial structure variables generated by the GWR
procedure in Table 6-4 and 5-5.
2) Sample mean and standard deviation of percent dispersed employment share (dispersion) are 81.6% and
6.05%, respectively; and these values for subcenters’ share of center employment (polycentricity) are 38.1%
and 25.5%, respectively.

147
6.4. DISCUSSION
In this chapter, I examined how the links between metropolitan spatial structure and
economic growth depend on metropolitan size. Consistent with the urban evolution
hypothesis discussed in the introduction, growth effects of employment dispersion were
found to be dependent on metropolitan size. A metropolitan area with more clustered spatial
form grows faster, perhaps enjoying agglomeration economies when it is small; whereas
more dispersion leads to higher growth rate as it grows large. Just as a city needs to
successfully take on higher order functions and economic activities to move upward within
the national urban system, it also needs to restructure its spatial form in such a way to
mitigate congestion or other diseconomies of size for continued growth.
Therefore, attempts to find one particular efficient urban form – at least with respect
to growth – may not be promising, just as the efforts to find the optimal city size have not
been fruitful. Efficient spatial structure may depend not only on the city size but also on
other urban attributes such as industrial structure and the shape of transportation networks,
which are products of the historical path of urban development. Insignificant growth effects
of polycentric versus monocentric structure, combining with findings from Chapter 4, imply
that there may exist many plausibly competitive urban forms and different paths of spatial
evolution.

148
CHAPTER 7.
CONCLUSIONS AND DISCUSSION

The objectives of this thesis were to uncover the current stage of spatial evolution, to
investigate the driving forces of spatial changes, and to examine the links between urban
spatial structure and commuting economies and economic growth in the contemporary US
metropolitan areas. Because each analysis chapter was relatively independent and contained
its own conclusions and discussion, this chapter will summarize the findings in earlier
chapters.
In Chapter 2, I successfully defined and estimated sets of spatial structure variables
by identifying employment centers consistently across 79 metropolitan areas with population
over a half million. One of the most important features of the modern metropolis based on
these spatial descriptors was that it is predominantly dispersed. Average dispersed
employment share was 82 percent by the GWR method and 73 percent by the minimum
density method. Majority of the jobs were diffused outside any type of employment centers
in all 79 metropolitan areas without exception. Findings from this Chapter parallel the
results of Gordon and Richardson (1996) and Lang (2003) – spatial evolution “beyond
polycentricity”.
Chapter 3 presented a series of statistical analyses to examine the determinants of
manifold dimensions of urban spatial structure. Whereas it turned out more intricate than it
appears to explain any regularity in urban spatial structure, the results provided valuable

149
findings that generally conform to the predictions from urban economic theories and path
dependence perspectives.
Larger metropolitan areas tend to have smaller CBD employment shares and more
decentralized and polycentric structures. However, population size was not a significant
factor explaining employment dispersion, the other spatial dimension. Congestion level was
a significant contributor to the subcenter formation in terms of the number of subcenters, but
was less significant in employment share models. Industrial composition was also found to
be an important spatial determinant, confirming that different industries are subject to
different agglomeration economies with varying geographical ambits.
The path dependence in urban spatial structure was indirectly identified in two ways.
The most recently developed metropolitan areas have smaller CBD and are more
decentralized than pre-war metros; while metros that reached the half of 2000 population 35
to 60 years ago had more polycentric structure. Second, metros with a strong agglomeration
in the urban core tended to have less number of subcenters and smaller subcenter
employment shares.
Chapter 4 explored spatial changes in six selected metropolitan areas for recent two
decades in order to address the question whether they are increasingly edgy or edgeless.
Findings paralleled the results of Gordon and Richardson (1996). Jobs continued to
decentralize from the metropolitan core to the suburbs in the 1980s and 1990s and jobs
dispersion was a more common phenomenon than subcentering.
Nevertheless, the results showed significant variation in spatial decentralization
trends rather than a uniform linear process from monocentric through polycentric, and to
dispersed structure. New York and Boston with big and long established CBDs were less

150
subject to decentralization; and polycentricity of Los Angeles and San Francisco was
reinforced in the last decade; while jobs dispersion was predominant in Portland and
Philadelphia. They seem to have developed unique patterns of decentralization, in light of
their histories and circumstances to limit the growth of commuting times.
These findings suggest two important theoretical implications. First, the
geographical and historical contexts of an individual metropolis strongly affect the path that
it takes in response to global trends such as ever decreasing transportation costs and IT
development. It seems that there is a “self reinforcing” pattern also in spatial development
as in technology adoption and industrial development. Second, industrial composition and
restructuring is an important part of the path dependent spatial evolution processes.
Chapter 5 presented a study on the commuting impacts of metropolitan level spatial
structure. Descriptive analysis identified a large potential of commute time saving from
spatial restructuring towards more polycentric and dispersed form, particularly in large
metropolitan areas. Notwithstanding with the potential, however, I found only partial
commute time saving effects of the spatial adjustment in regression analyses. While
employment dispersion helped reduce commute time, polycentricity effect was not
significant.
The insignificance of polycentric spatial dimension can be interpreted in two ways.
First, CBD/main center employment share may be already too small to affect metro wide
average commute time. Second, polycentrization may have system wide effects such as
increased cross commuting, offsetting potential commute time savings.
Chapter 6 examined how the links between metropolitan spatial structure and
economic growth depend on the size of the metropolis. Consistent with theories of urban

151
system and evolution, growth effects of employment dispersion were found to be dependent
on metropolitan size. A metropolitan area with more clustered spatial form grows faster
when it is small; whereas more dispersion leads to higher growth rate as it grows large. Just
as a city needs to successfully take on higher order functions and economic activities to
move upward within the national urban system, it also needs to restructure its spatial form in
a way to mitigate congestion or other diseconomies of size for continued growth.
Therefore, attempts to find one particular efficient urban form may not be promising,
just as the efforts to find the optimal city size have not been fruitful. Efficient spatial
structure may depend not only on the city size but also on other urban attributes such as
industrial structure and the shape of transportation networks, which are products of the
historical path of urban development. Insignificant growth effects of polycentric versus
monocentric structure imply that there may exist many plausibly competitive urban forms
and different paths of spatial evolution.

152
REFERENCES

Anas, Alex, R. Arnott, and K. A. Small. 1998. Urban Spatial Structure. Journal of Economic
Literature 36:1426-1464.
Anas, Alex, and Ikki Kim. 1996. General equilibrium models of polycentric urban land use
with endogenous congestion and job agglomeration. Journal of Urban Economics
40:232-256.
Anderson, N. B., and W. T. Bogart. 2001. The structure of sprawl: Identifying and
characterizing employment centers in polycentric metropolitan areas. American
Journal of Economics and Sociology 60 (1):147-169.
Anderson, W. P., P. S. Kanaroglou, and E. J. Miller. 1996. Urban form, energy and the
environment: A review of issues, evidence and policy. Urban Studies 33 (1):7-35.
Arthur, W. B. 1989. Competing Technologies, Increasing Returns, and Lock-in by Historical
Events. Economic Journal 99 (394):116-131.
———. 1990. Positive Feedbacks in the Economy. Scientific American 262 (2):92-&.
———. 1990. Silicon Valley Locational Clusters - When Do Increasing Returns Imply
Monopoly. Mathematical Social Sciences 19 (3):235-251.
Arthur, W. Brian. 1988. Self-reinforcing mechanisms in economics. In The Economy as an
Evolving Complex System, edited by K. J. Arrow and D. Pines. Reading, MA:
Addison-Wesley.
———. 1988. Urban systems and historical path dependence. In Cities and Their Vital
Systems: Infrastructure Past, Present, and Future, edited by J. H. Ausubel and R.
Herman. Washington D.C.: National Academy Press.
———. 1994. Increasing Returns and Path Dependence in the Economy. Ann Arbor, MI:
The University of Michigan Press.
Audirac, Ivonne. 2002. Information technology and urban form. Journal of Planning
Literature 17 (2):212-226.
Badoe, Daniel A., and Eric J. Miller. 2000. Transportaion-land-use interaction: empirical
findings in North America, and their implications for modeling. Transportation
Research D 5:235-263.

153
Barnard, J. R., and A. C. Krautmann. 1988. Population-Growth among United-States
Regions and Metropolitan Areas - a Test for Causality. Journal of Regional Science
28 (1):103-118.
Baum, Christopher, Mark E. Schaffer, and Steven Stillman. 2003. Instrumental variables and
GMM: Estimation and testing. Stata Journal 3 (1):1-31.
Begovic, B. 1991. The Economic-Approach to Optimal City Size. Progress in Planning
36:93-161.
Black, D., and V. Henderson. 1999. Spatial Evolution of Population and Industry in the
United States. American Economic Review 89 (2):321-327.
———. 2003. Urban evolution in the USA. Journal of Economic Geography 3 (4):343-372.
Boarnet, M. G. 1994. The Monocentric Model and Employment Location. Journal of Urban
Economics 36 (1):79-97.
Boarnet, Marlon G., and Randall Crane. 2001. The influence of land use on travel behavior:
specification and estimation strategies. Transportation Research A 35:823-845.
Bogart, W. T., and W. C. Ferry. 1999. Employment centres in greater Cleveland: Evidence
of evolution in a formerly monocentric city. Urban Studies 36:2099-2110.
Brueckner, Jan K. 1987. The structure of urban equilibria: a unified treatment of the Muth-
Mills model. In Handbook of Regional and Urban Economics, edited by E. S. Mills.
Aldershot: Elsevier Science.
Bruegmann, Robert. 2005. Sprawl: A Compact History: The Univ. of Chicago Press.
Cairncross, F. 1997. The death of distance. Cambridge: Harvard Business School Press.
Camagni, R., L. Diappi, and G. Leonardi. 1986. Urban-Growth and Decline in a Hierarchical
System - a Supply-Oriented Dynamic Approach. Regional Science and Urban
Economics 16 (1):145-160.
Capello, Roberta, and Roberto Camagni. 2000. Beyond optimal city size: an evaluation of
alternative urban growth patterns. Urban Studies 37 (9):1479-1496.
Carlino, G. A., and E. S. Mills. 1987. The Determinants of County Growth. Journal of
Regional Science 27 (1):39-54.
Carlino, Gerald A. 1987. Productivity in cities: does city size matter? Business Review:3-12.
Carlino, Gerald, and Satyajit Chatterjee. 2002. Employment deconcentration: a new
perspective on America's postwar urban evolution. Journal of Regional Science 42
(3):455-475.

154
Castells, Manuel. 1989. The Informational City. Cambridge, MA: Blackwell.
Cervero, R. 2001. Efficient Urbanisation: Economic Performance and the Shape of the
Metropolis. Urban Studies 38 (10):1651-1671.
Cervero, Robert, and Roger Gorham. 1995. Commuting in transit versus automobile
neighborhoods. Journal of American Planning Association 61 (2):210-225.
Cervero, Robert, and John Landis. 1992. Suburbanization of Jobs and the Journey to Work:
A Submarket Analysis of Commuting in the San Francisco Bay Area. Journal of
advanced transportation. 26 (3):275-298.
Cervero, Robert, and Kang-Li Wu. 1998. Sub-centring and Commuting: Evidence from the
San Francisco Bay Area, 1980-90. Urban Studies 35 (7):1059-1076.
Champion, A. G. 2001. A changing demographic regime and evolving polycentric urban
regions: consequences for the size, composition and distribution of city populations.
Urban Studies 38 (4):657-677.
Ciccone, A., and R. E. Hall. 1996. Productivity and the density of economic activity. The
American Economic Review 86 (1):54-70.
Clark, Colin. 1945. The economic functions of a city in relation to its size. Econometrica 13
(2):97-113.
Clark, D. E., and C. A. Murphy. 1996. Countywide Employment and Population Growth: an
Analysis of the 1980s. Journal of Regional Science 36 (2):235-256.
Clark, D., J. R. Kahn, and H. Ofek. 1988. City size, quality of life and the urbanization
deflator of GNP, 1910-1984. Southern Economic Journal 54 (3):701-714.
Clark, W. A. V. 2000. In Monocentric to polycentric: New urban forms and old paradigm,
edited by G. Bridge and S. Watson. Oxford, UK: Blackwell.
Clark, William A. V., and Marianne Kuijpers-Linde. 1994. Commuting in Restructuring
Urban Regions. Urban Studies 31 (3):465-483 (19 pages).
Combes, Pierre-Philippe. 2000. Economic structure and local growth: France, 1984-1993.
Journal of Urban Economics 47:329-355.
Craig, Steven G., and Pin T. Ng. 2001. Using quantile smoothing splines to identify
employment subcenters in a multicentric urban area. Journal of Urban Economics
49:100-120.
Crane, Randall. 2000. The influence of urban form on travel: an interpretive review. Journal
of Planning Literature 15 (1):3-23.

155
Crane, Randall, and Daniel Chatman. 2003. Traffic and sprawl: Evidence from U.S.
commuting, 1985 to 1997. Planning and Markets 6 (1):14-22.
Cutsinger, J., G. Galster, H. Wolman, R. Hanson, and D. Towns. 2005. Verifying the multi-
dimensional nature of metropolitan land use: Advancing the understanding and
measurement of sprawl. Journal of Urban Affairs 27 (3):235-259.
David, P. A. 1985. Clio and the Economics of Qwerty. American Economic Review 75
(2):332-337.
Dear, Michael, and Steven Flusty. 1998. Postmodern urbanism. Annals of the Association of
American Geographers 88 (1):50-72.
Dekle, R., and J. Eaton. 1999. Agglomeration and land rents: evidence from the prefectures.
Journal of Urban Economics 46:200-214.
Eaton, J., and Z. Eckstein. 1997. Cities and growth: Theory and evidence from France and
Japan. Regional Science and Urban Economics 27 (4-5):443-474.
Eberts, R. W., and D. P. McMillen. 1999. Agglomeration economies and urban public
infrastructure. In Handbook of Regional and Urban Economics, edited by E. S. Mills
and P. Cheshire. Aldershot: Elsevier Science.
Erickson, Rodney A. 1986. Multinucleation in metropolitan economies. Annals of the
Association of American Geographers 76 (3):331-346.
Ewing, R., and R. Cervero. 2001. Travel and the built environment - A synthesis. In Land
Development and Public Involvement in Transportation.
Ewing, R., R. Pendall, and D. Chen. 2003. Measuring sprawl and its transportation impacts.
Transportation Research Record 1931:175-183.
Feldman, M. P., and D. B. Audretsch. 1999. Innovation in cities: Science-based diversity,
specialisation and localized competition. European Economic Review 43:409-429.
Forstall, Richard L., and Richard P. Greene. 1997. Defining job concentrations: The Los
Angeles case. Urban Geography 18 (8):705-739.
Freestone, Robert, and Peter Murphy. 1998. Metropolitan restructuring and suburban
employment centers: Cross-cultural perspectives on the Australian experience.
Journal of American Planning Association 64 (3):286-297.
Fujii, Tadashi, and Truman Asa Hartshorn. 1995. The changing metropolitan structure of
Atlanta, Georgia: Locations of functions and regional structure in a multinucleated
urban area. Urban Geography 16 (8):680-707.

156
Fujita, M., J. F. Thisse, and Y. Zenou. 1997. On the endogeneous formation of secondary
employment centers in a city. Journal of Urban Economics 41:337-357.
Fujita, Masahisa, and Hideaki Ogawa. 1982. Multiple equilibria and structural transition of
nonmonocentric urban configurations. Regional Science and Urban Economics
12:161-196.
Fulton, William. 1996. Are edge cities losing their edge? Planning 62 (5):4-7.
Gabaix, X. 1999. Zipf's law and the growth of cities. American Economic Review 89
(2):129-132.
Gabriel, S. A., and S. S. Rosenthal. 2004. Quality of the business environment versus quality
of life: Do firms and households like the same cities? Review of Economics and
Statistics 86 (1):438-444.
Galster, George, Royce Hanson, Michael R. Ratcliffe, Harold Wolman, Stephen Coleman,
and Jason Freihage. 2001. Wrestling sprawl to the ground: Defining and measuring
an elusive concept. Housing Policy Debate 12 (4):681-717.
Garreau, Joel. 1991. Edge City: Life on the New Frontier. NY: Doubleday.
Gerking, S. 1994. Measuring productivity growth in U.S. regions: a survey. International
Regional Science Review 16 (1):155-185.
Giuliano, Genevieve, Christian Redfearn, Ajay Agarwal, Chen Li, and Duan Zhuang. 2005.
Not all sprawl: Evolution of employment concentrations in Los Angeles, 1980-2000.
Lusk Center for Real Estate, University of Southern California Working paper 2005-
1002.
Giuliano, Giuliano, and Kenneth A. Small. 1991. Subcenters in the Los Angeles Region.
Regional Science and Urban Economics 21 (2):163-182.
Glaeser, E. L., and J. Gyourko. 2005. Urban decline and durable housing. Journal of
Political Economy 113 (2):345-375.
Glaeser, E. L., and M. E. Kahn. 2003. Sprawl and urban growth. In Handbooks in Urban and
Regional Economics, edited by J. V. Henderson and J. F. Thisse: Oxford University
Press (forthcoming).
Glaeser, E. L., J. A. Scheinkman, and A. Shleifer. 1995. Economic-Growth in a Cross-
Section of Cities. Journal of Monetary Economics 36 (1):117-143.
Glaeser, E. L., and J. M. Shapiro. 2003. Urban growth in the 1990s: Is city living back?
Journal of Regional Science 43 (1):139-165.
Glaeser, Edward L. 1999. Learning in cities. Journal of Urban Economics 46:254-277.

157
———. 2000. The new economics of urban and regional growth. In The Oxford Handbook
of Economic Geography, edited by G. L. Clark, M. Gertler and M. Feldman. Oxford,
UK: Oxford University Press.
Glaeser, Edward L., and Matthew E. Kahn. 2001. Decentralized employment and the
transformation of the American city. NBER Working Paper 8117.
Glaeser, Edward L., Hedi D. Kallal, Jos A. Scheinkman, and Andrei Shleifer. 1992. Growth
in cities. Journal of Political Economy 100:1126-1152.
Glaeser, Edward L., and Janet E. Kohlhase. 2004. Cities, regions and the decline of transport
costs. Papers in Regional Science 83:197-228.
Glaeser, Edward L., Jed Kolko, and Albert Saiz. 2001. Consumer city. Journal of Economic
Geography 1:27-50.
Gordon, P., and H. W. Richardson. 1996. Beyond polycentricity: the dispersed metropolis,
Los Angeles, 1970-1990. Journal of American Planning Association 62 (3).
Gordon, P., H. W. Richardson, and M. J. Jun. 1991. The commuting paradox: evidence from
the top twenty. Journal of American Planning Association 57 (4):416-420.
Gordon, P., H. W. Richardson, and H. L. Wong. 1986. The distribution of population and
employment in a polycentric city: the case of Los Angeles. Environment and
Planning A 18:161-173.
Gordon, Peter, Ajay Kumar, and Harry W. Richardson. 1989. The influence of metropolitan
spatial structure on commuting time. Journal of Urban Economics 26:138-151.
Gordon, Peter, and H. L. Wong. 1985. The cost of urban sprawl: Some new evidence.
Environment and Planning A 17:661-666.
Greene, William H. 2000. Econometric Analysis. 4 ed. Upper Saddle River, NJ: Prentice-
Hall.
Harrison, B., M. R. Kelley, and J. Gant. 1996. Innovative firm behavior and local milieu:
exploring the intersection of agglomeration, firm effects, and technological change.
Economic Geography 72 (3):233-258.
Hartshorn, Truman A., and Peter O. Muller. 1989. Suburban downtowns and the
transformation of metropolitan Atlanta's business landscape. Urban Geography 10
(4):375-395.
Hausman, J. A. 1978. Specification tests in econometrics. Econometrica 46 (6):1251-1271.

158
Heikkila, E., P. Gordon, J. I. Kim, B. Peiser, H. W. Richardson, and D. Dale-Johnson. 1989.
What happened to the CBD-distance gradient?: land values in a polycentric city.
Environment and Planning A 21:221-232.
Helsley, Robert W., and Arthur M. Sullivan. 1991. Urban subcenter formation. Regional
Science and Urban Economics 21:255-275.
Henderson, J. V. 1986. Efficiency of resource usage and city size. Journal of Urban
Economics 19:47-70.
Henderson, V., and A. Mitra. 1996. The new urban landscape: developers and edge cities.
Regional Science and Urban Economics 26:613-643.
Henderson, Vernon, Ari Kuncoro, and Matt Turner. 1995. Industrial development in cities.
Journal of Political Economy 103 (5):1067-1090.
Hirsch, W. Z. 1959. Expenditure Implications of Metropolitan Growth and Consolidation.
Review of Economics and Statistics 41 (3):232-241.
Horton, F. E., and D. R. Reynolds. 1971. Effects of Urban Spatial Structure on Individual
Behavior. Economic Geography 47 (1):36-48.
Krugman, P. 1991. History and Industry Location - the Case of the Manufacturing Belt.
American Economic Review 81 (2):80-83.
Lang, Robert E. 2003. Edgeless Cities: Exploring the Elusive Metropolis. Washington, D.C.:
Brookings Institution Press.
Lang, Robert E., and Jennifer LeFurgy. 2003. Edgeless cities: Examining the noncentered
metropolis. Housing Policy Debate 14 (3):427-460.
Levinson, D. M., and A. Kumar. 1994. The rational locator: why travel times have remained
stable. Journal of American Planning Association 60 (3):319-332.
Malmberg, A. 1996. Industrial geography: agglomeration and local milieu. Progress in
Human Geography 20 (3):392-403.
Malmberg, Anders, and Örjan Sölvell. 1997. Localised innovation processes and the
sustainable competitive advantage of firms: A conceptual model. In Interdependent
and Uneven Development, edited by M. Taylor and S. Conti. Aldershot: Ashgate.
Maskell, P., and A. Malmberg. 1999. Localised learning and industrial competitiveness.
Cambridge Journal of Economics 23:167-185.
Massey, Douglas S., and Nancy Denton. 1988. The dimensions of residential segregation.
Social Forces 67:281-313.

159
McDonald, John F., and Paul J. Prather. 1994. Suburban employment centres: The case of
Chicago. Urban Studies 31 (2):201-218.
McMillen, D. P. 2001. Nonparametric employment subcenter identification. Journal of
Urban Economics 50:448-473.
McMillen, D. P., and J. F. McDonald. 1998. Suburban subcenters and employment density
in metropolitan Chicago. Journal of Urban Economics 43:157-180.
McMillen, Daniel P. 2003. Employment subcenters in Chicago: Past, present, and future.
Economic Perspectives, Federal Reserve Bank of Chicago 2Q:2-14.
———. 2003. Identifying sub-centres using contiguity matrices. Urban Studies 40 (1):57-
69.
McMillen, Daniel P., and Stefani C. Smith. 2003. The number of subcenters in large urban
areas. Journal of Urban Economics 53:321-338.
Mieszkowski, P., and E. S. Mills. 1993. The causes of metropolitan suburbanization. Journal
of Economic Perspectives 7 (3):135-147.
Mills, Edwin S. 1967. An aggregative model of resource allocation in a metropolitan area.
The American Economic Review 57 (2):197-210.
Mills, Edwin S. Hamilton Bruce W. 1994. Urban Economics. New York: Harper Collins.
Moomaw, Ronald L. 1981. Productivity and city size: a critique of the evidence. The
Quarterly Journal of Economics 96 (4):675-688.
———. 1983. Spatial productivity variations in manufacturing: a critical survey of cross-
sectional analyses. International Regional Science Review 8 (1):1-22.
———. 1988. Agglomeration economies: Localization or urbanization? Urban Studies 25
(2):150-161.
Muller, Peter O. 1997. The suburban transformation of the globalizing American city.
Annals of the American Academy of Political and Social Science 551:44-58.
———. 2004. Transportation and urban form: Stages in the spatial evolution of the
American Metropolis. In The Geography of Urban Transportation, edited by S.
Hanson and G. Giuliano. New York: The Guilford Press.
Mulligan, G. F., A. C. Vias, and S. M. Glavac. 1999. Initial Diagnostics of a Regional
Adjustment Model. Environment and Planning a 31 (5):855-876.
Muth, R. F. 1975. Numerical solutions of urban residential land-use models. Journal of
Urban Economics 2:307-332.

160
Nelson, Richard R. , and Sidney G. Winter. 1982. An Evolutionary Theory of Economic
Change. Cambridge: Harvard University Press.
Nitsch, V. 2003. Does history matter for urban primacy? The case of Vienna. Regional
Science and Urban Economics 33 (4):401-418.
Ogawa, H., and M. Fujita. 1989. Nonmonocentric urban configurations in a two-dimensional
space. Environment and Planning A 21:363-374.
Ozawa, Connie P. 2004. The Portland Edge: Challenges and Sucesses in Growing
Communities. Washington, DC: Island Press.
Pfister, N., R. Freestone, and P. Murphy. 2000. Polycentricity or dispersion?: changes in
center employment in metropolitan sydney, 1981 to 1996. Urban Geography 21
(5):428-442.
Pierson, P. 2000. Increasing returns, path dependence, and the study of politics. American
Political Science Review 94 (2):251-267.
Porter, M. E. 2000. Location, competition, and economic development: Local clusters in a
global economy. Research and Practice 14 (1):15-34.
Prud'homme, R., and C. W. Lee. 1999. Size, Sprawl, Speed and the Efficiency of Cities.
Urban Studies 36 (11):1849-1858.
Richardson, H. W. 1972. Optimality in city size, systems of cities and urban policy: a
sceptic's view. Urban Studies 9 (1):29-48.
Richardson, Harry W. 1995. Economies and diseconomies of agglomeration. In Urban
Agglomeration and Economic Growth, edited by H. Giersch. Berlin: Springer.
Rigby, D. L., and R. Essletzbichler. 2002. Agglomeration economies and productivity
differences in US cities. Journal of Economic Geography 2 (4):407-432.
Rosenthal, S. S., and W. C. Strange. 2001. The determinants of agglomeration. Journal of
Urban Economics 50:191-229.
Sarzynski, Andrea, Harold L. Wolman, George Galster, and Royce Hanson. 2006. Testing
the conventional wisdom about land use and traffic congestion: The more we sprawl,
the less we move? Urban Studies 43 (3):601-626.
Sasaki, K., and Se-Il Mun. 1996. A dynamic analysis of multiple-center formation in a city.
Journal of Urban Economics 40:257-278.
Sassen, Saskia. 1991. The Global City: New York, London, and Tokyo. Princeton, NJ:
Princeton University Press.

161
Saxenian, AnnaLee. 1994. Regional Advantage: Culture and Competition in Silicon Valley
and Route 128. Cambridge: Harvard Univ. Press.
Schwanen, Tim. 2003. Spatial Variations in Travel Behavior and Time Use, Utrecht
University, Netherlands.
Scott, A. J. 1988. Metropolis: from the division of labor to urban form. Berkeley: Univ. of
California Press.
———. 1988. New Industrial Spaces. London: Pion.
Scott, Allen J. 1990. The technopoles of Southern California. Environment and Planning A
22:1575-1605.
Shapiro, J. M. 2006. Smart cities: Quality of life, productivity, and the growth effects of
human capital. Review of Economics and Statistics 88 (2):324-335.
Simon, C. J., and C. Nardinelli. 2002. Human Capital and the Rise of American Cities, 1900-
1990. Regional Science and Urban Economics 32 (1):59-96.
Small, K. A., and S. Song. 1994. Population and employment densities: structure and
change. Journal of Urban Economics 36:292-313.
Stanback, T. M. 1991. The New Suburbanization. Boulder, CO: Westview Press.
Steinnes, D. N., and W. D. Fisher. 1974. Econometric Model of Intraurban Location.
Journal of Regional Science 14 (1):65-80.
Storper, Michael, and Anthony J. Venables. 2004. Buzz: face-to-face contact and teh urban
economy. Journal of Economic Geography 4:351-370.
Sullivan, A. M. 1986. A general equilibrium model with agglomerative economies and
decentralized employment. Journal of Urban Economics 20:55-74.
Timothy, Darren, and William C. Wheaton. 2001. Intra-urban wage variation, employment
location, and commuting times. Journal of Urban Economics 50:338-366.
Van der Laan, Lambert. 1998. Changing urban systems: An empirical analysis at two spatial
levels. Regional Studies 32 (3):235-247.
Waddell, P., and V. Shukla. 1993. Employment dynamics, spatial restructuring, and the
business cycle. Geographical Analysis 25 (1):35-52.
Wheaton, W. C. 1974. A comparative static analysis of urban spatial structure. Journal of
Economic Theory 9:223-237.

162
———. 2004. Commuting, congestion, and employment dispersal in cities with mixed land
use. Journal of Urban Economics 55:417-438.
Wheeler, C. H. 2003. Evidence on Agglomeration Economies, Diseconomies, and Growth.
Journal of Applied Econometrics 18 (1):79-104.
White, M. J. 1976. Firm suburbanization and urban subcenters. Journal of Urban Economics
3:353-343.
Wooldridge, Jeffrey M. 2002. Economic Analysis of Cross Section and Panel Data.
Cambridge, MA: The MIT Press.

Abstract (if available)

Abstract There have been "qualitative changes" in metropolitan spatial structure in recentdecades. While these changes have been widely recognized, much less is known about thespecifics -- the forms, causes, and consequences of the spatial changes. In that regard, thisresearch aims to address several questions: What are the prominent features of emergingurban forms? Are cities becoming more edgy or more edgeless? What are primary forcesdriving the spatial changes? What are the consequences of the spatial changes in dailycommuting and urban economic growth?

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Population and employment distribution and urban spatial structure: an empirical analysis of metropolitan Beijing, China in the post-reform era

PDF

Testing the entrepreneurial city hypothesis: a study of the Los Angeles region

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Urban spatial transformation and job accessibility: spatial mismatch hypothesis revisited

PDF

From structure to agency: Essays on the spatial analysis of residential segregation

PDF

Satisfaction with local "public goods" and services: the effects of household income and privatization in southern California

PDF

Essays on economic modeling: spatial-temporal extensions and verification

PDF

Structure, agency, and the Kuznets Curve: observations and implications for sustainability planning in U.S. cities

PDF

The twilight of the local redevelopment era: the past, present, and future of urban revitalization and urban economic development in Nevada and California

PDF

The relationship between women's household responsibilities and commute lengths: a study on women in the US and Great Britain

PDF

Neighborhood greenspace's impact on residential property values: understanding the role of spatial effects

PDF

Organizational structuring for homeland security: a U.S. state-by-state comparative analysis

PDF

Urbanization beyond the metropolis: three papers on urbanization patterns and their planning implications in the contemporary global south

PDF

Talent migration: does urban density matter?

PDF

Three papers on dynamic externalities and regional growth

PDF

Social entrepreneurship and urban tourism as economic development: best practices from Long Beach, California

PDF

Humanitarian aid and the production of spatial knowledge and practices in the U.S.-Mexico borderlands

PDF

Individual differences in trust development: Assessing the effects of trustor attributes on trust building, stability, and dissolution

PDF

The film industry and urban development in metropolitan Los Angeles, 1920-1975

PDF

Gang injunctions and community participation

Asset Metadata

Creator Lee, Bumsoo (author)

Core Title Urban spatial structure, commuting, and growth in U.S. metropolitan areas

School School of Policy, Planning, and Development

Degree Doctor of Philosophy

Degree Program Planning

Publication Date 11/17/2006

Defense Date 07/17/2006

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag commuting,metropolitan growth,OAI-PMH Harvest,urban spatial structure

Language English

Advisor Gordon, Peter (committee chair), Moore, James Elliott, II (committee member), Redfearn, Christian L. (committee member), Richardson, Harry W. (committee member)

Creator Email bumsoole@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m165

Unique identifier UC191858

Identifier etd-Lee-20061117 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-32544 (legacy record id),usctheses-m165 (legacy record id)

Legacy Identifier etd-Lee-20061117.pdf

Dmrecord 32544

Document Type Dissertation

Rights Lee, Bumsoo

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu