University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

POPULATION AND EMPLOYMENT DISTRIBUTION AND URBAN SPATIAL
STRUCTURE: AN EMPIRICAL ANALYSIS OF METROPOLITAN BEIJING, CHINA
IN THE POST-REFORM ERA

by

Tieshan Sun

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)

May 2009

Copyright 2009 Tieshan Sun

ii

TABLE OF CONTENTS

LIST OF TABLES iv

LIST OF FIGURES v

ABSTRACT vii

CHAPTER 1. INTRODUCTION 1
1.1 Research Background 1
1.2 Research Questions 9
1.3 Organization of the Dissertation 12
1.4 Study Area 14
1.4.1 Specifying the Study Area’s Boundary 14
1.4.2 A Portrait of Beijing’s Urban Form 16
1.4.3 Spatial Development and Planning of Beijing 18
1.5 Summary 22

CHAPTER 2. LITERATURE REVIEW 25
2.1 Evolution of Urban Spatial Structure 25
2.1.1 Review of Urban Economic Theories 25
2.1.2 Empirical Regularities 30
2.2 Urban Spatial Structure and Commuting 35
2.3 Spatial Restructuring of Chinese Cities in the Post-Reform Era 42

CHAPTER 3. CHARACTERIZATION AND EVOLUTION OF POPULATION
DISTRIBUTION IN METROPOLITAN BEIJING 52
3.1 Introduction 52
3.2 Methodology 54
3.2.1 Analytical Techniques 54
3.2.2 Description of Data 57
3.3 Characterizing Population Density Patterns with Density Functions 61

iii
3.4 Decentralization and Subcentering of Population in Metropolitan
Beijing: A Nonparametric Analysis 71
3.4.1 Local Regression 71
3.4.2 Characterizing Density Patterns using Local Regression 73
3.4.3 Subcentering of Population and Urban Structure 85
3.5 Further Discussions 93

CHAPTER 4. EMPLOYMENT DISTRIBUTION AND SPATIAL
ORGANIZATION OF ECONOMIC ACTIVITY IN METROPOLITAN
BEIJING 99
4.1 Introduction 99
4.2 Data Issues 100
4.3 Exploring Spatial Distribution and Clustering of Employment 104
4.3.1 Analytical Techniques 104
4.3.2 Results and Discussion 110
4.4 Employment Centers and Urban Structure 117
4.4.1 Identification of Employment Centers 117
4.4.2 Results and Discussion 123
4.5 Summary and Discussion 136

CHAPTER 5. JOBS-HOUSING BALANCE AND URBAN COMMUTING IN
METROPOLITAN BEIJING 141
5.1 Introduction 141
5.2 Data 145
5.3 Evidence of Jobs-Housing Balance 148
5.4 Commuting Patterns and Urban Structure 155
5.5 Jobs-Housing Balance and Commuting Time 164
5.5.1 E/P Ratio and Mean Commuting Time 164
5.5.2 Model Specifications 166
5.5.3 Regression Results 169
5.6 Summary 176

CHAPTER 6. CONCLUSION 179

REFERENCES 186

iv

LIST OF TABLES

Table 3.1 The Summary Statistics for the Subdistricts 58
Table 3.2 The Distribution of Population in the Beijing Metropolitan Area at Different
Spatial Scales and its Changes 60
Table 3.3 The Estimation Results of the Density Functions for Both the Metropolitan
Area and the Urban Area of Beijing 65
Table 3.4 Final List of Population Subcenters and Estimation Results 88
Table 4.1 Summary Information on Establishments and Employment by Economic
Sector 104
Table 4.2 Distributions of Establishments and Employment at Different Spatial
Scales 113
Table 4.3 Final List of Employment Subcenters and Semiparametric Estimation 129
Table 4.4 Summary of Employment Centers 130
Table 4.5 Aggregate Employment by Sector inside and outside Centers 133
Table 4.6 Location Quotients for Employment by Sector in Employment Centers 134
Table 5.1 Summary of Respondents’ Socioeconomic and Commuting Attributes 147
Table 5.2 The Distribution of E/P Ratios at Different Levels 153
Table 5.3 Commuting Distribution Pattern in the Survey Area 156
Table 5.4 Regression Results 171

v

LIST OF FIGURES

Figure 1.1 Research Scheme 10
Figure 1.2 The Topology, Road System and the Administrative Organization
of Beijing 16
Figure 1.3 The Built-up Areas and the Ring Road System of the Central Beijing 18
Figure 1.4 The Spatial Expansion of Beijing 20
Figure 1.5 The Spatial Development Scheme of Beijing 21
Figure 3.1 Spatial Distribution of Population in Metropolitan Beijing 61
Figure 3.2 Fitted Density Functions for the Metropolitan Area of Beijing 64
Figure 3.3 Fitted Density Functions for the Urban Area of Beijing 67
Figure 3.4 Growth Patterns of Population Densities in Metropolitan Beijing 70
Figure 3.5 Loess Fit of the Logarithmic Density Function 74
Figure 3.6 Loess Surfaces of Population Density in Metropolitan Beijing 77
Figure 3.7 Contour Maps of Loess Density Surfaces 78
Figure 3.8 The Decentralization of Population and the Spatial Development Scheme
of the Beijing Metropolitan Area 80
Figure 3.9 Density Profiles along the East-West Axis 83
Figure 3.10 Density Profiles along the Radial Transport Axes 84
Figure 3.11 The Location of Subcenters and the Development Scheme of the Beijing
Metropolitan Area 90
Figure 3.12 Expansion of the Central Agglomeration of Population 90

vi
Figure 4.1 Distributions of Establishments and Employment 103
Figure 4.2 The Standard Deviational Ellipses for Various Economic Sectors 112
Figure 4.3 The D Functions for Various Economic Sectors 115
Figure 4.4 Kernel Employment Density Surface 124
Figure 4.5 Spatial Autocorrelation Types and Moran Significance Map 126
Figure 4.6 Boundaries and Locations of Candidate Employment Centers 127
Figure 4.7 Locations of Population Subcenters and Employment Centers and the
Development Scheme of Beijing 132
Figure 5.1 The Survey Area 146
Figure 5.2 Employment-Population Ratios and Urban Subcenters 152
Figure 5.3 Loess Curves of Urban Densities of People and Jobs 154
Figure 5.4 Spatial Distributions of Commuting Inflows and Outflows 158
Figure 5.5 Contour Maps for the Population and Employment Density Surfaces 159
Figure 5.6 Major External Commuting Flows and Self-Containment of Subdistricts 162
Figure 5.7 Cumulative Percentage of Commuting Flows with Distances between
Subdistricts of Residence and Workplace 164
Figure 5.8 E/P Ratio and Mean Commuting Time 165

vii
ABSTRACT

From a comparative international perspective, this dissertation explores the spatial
distributions of population and employment in the Beijing metropolitan area in the post-
reform era. This study aims to extend the literature on urban spatial structure, with special
reference to the pattern and process of urban decentralization and restructuring from a
developing and transitional economy context, and to offer further understanding of the
spatial organization of contemporary urban areas that departs from the North American or
European experience.
Beijing is a transition city that has experienced dramatic urban growth and spatial
restructuring since the reforms in China, and its experience sheds light on how the urban
spatial structure changes within a hybrid of an evolving market economy with a central
government that retains significant control. This study focuses on the distribution patterns
of population and employment in metropolitan Beijing, and employs more flexible
techniques, such as nonparametric analysis, geostatistical techniques and demonstrates
more nuanced dynamics than those discussed in previous studies.
Our study finds that the spatial pattern of the Beijing metropolitan area is
becoming alike to that of large Western cities in the post-reform era, with the compact
urban form in the pre-reform era replaced by a more dispersed and polycentric spatial
pattern. The overall trend toward the decentralization and polycentrification of both
population and employment is evident in the Beijing metropolitan area since the reforms.

viii
However, compared with the decentralization of large Western metropolitan areas, the
extent of the decentralization of metropolitan Beijing is quite limited. We show that both
people and jobs that moved out of the inner city tend to re-concentrate in the near suburbs
adjacent to the central area instead of dispersing throughout the metropolitan area. The
rapid growth of the near suburbs has expedited the expansion of the central city, with a
larger central agglomeration emerged dominating the whole metropolitan area. In this
broad sense, the spatial pattern of the Beijing metropolitan area is still highly centralized,
and the tendency toward decentralization at the level of the metropolitan area is
questionable. Besides, although both people and jobs have decentralized in the Beijing
metropolitan area, jobs tend to more concentrate in the central city, and employment has
been shown to be less decentralized than population.
Even though the spatial structure of Beijing is largely characterized by
monocentricity, our study does provide the evidence that significant population and
employment subcenters have emerged in the suburbs of Beijing. However, the number
and size of subcenters are small, and the pattern of the subcenters in metropolitan Beijing
is highly related to the planned development scheme of the city, so the polycentricity
emerged in the Beijing metropolitan area is very different by nature from that observed in
Western cities, and has different origins. Although the common features of spatial pattern
and trend broadly analogous to those of Western cities have been observed in post-reform
Beijing, the driving forces and the process involved still need be understood with
reference to the peculiar Chinese context, and the similar factors that caused the

ix
suburbanization in the West have taken their effects on the suburbanization of Beijing in
a totally different context.
An interesting finding of our study is the similar distribution patterns of both
population and employment in the Beijing metropolitan area, with the coincidence of
population and employment subcenters in space. The comparison of the distributions of
people and jobs in the Beijing metropolitan area shows that the suburban areas adjacent
to the central area are the most balanced with people and jobs, corresponding to the
emerged subcenters in the near suburbs. This result might be used to argue that jobs-
housing balance occurs as part of the urban development process, and the
decentralization of both population and employment in the Beijing metropolitan area
from the inner city to the near suburbs has induced jobs-housing balance in the near
suburbs. At last, the relationship between jobs-housing balance and urban commuting is
tested through regression analysis. The results show that balancing people and jobs by
configuring land use patterns seems not quite relevant to shortening commuting durations
in our case, so this may suggest that it is more promising to integrate transportation
planning and land use planning to address the transportation problems in Beijing.

1
CHAPTER 1.
INTRODUCTION

1.1 RESEARCH BACKGROUND
Since the late 1980s, there has been revitalized interest in the analysis of urban
spatial structures in the urban economics field. Anas et al. (1998) attributed this to the
fact that urban growth patterns in developed countries have undergone a “qualitative
change” over the last two decades, characterized by the emergence of increasingly large
and diversified suburban employment subcenters that are in direct competition with the
traditional city center, with the continual decentralization of both population and
employment, which have profoundly changed the spatial structure of contemporary
metropolitan areas and led to a more dispersed and polycentric urban form (Coffey and
Shearmur, 2001). Although the nature, causes and consequences of this spatial change
have still been under debate (Lee, 2007; Shearmur et al., 2007), the polycentric urban
phenomenon has been extensively documented and empirical regularities are evident in
the literature (Anas et al., 1998; McMillen and Smith, 2003; Baumont et al., 2004;
McMillen, 2004). Policy concerns have also arisen regarding the changing urban
structure, given the social, environmental and economic impact involved (Lang and
Lefurgy, 2003).
However, empirical studies conducted so far are mostly based on the urban
experience of advanced Western countries, while few studies have been carried out on the

2
developing urban world, and much less has been known about how cities in developing
countries have changed over the decades, and whether the similar development trend to
that of Western metropolitan evolution is also apparent there. Some of the recent
scholarship has revealed that, responding to similar global forces of economic
restructuring and technology advances, such urban processes as decentralization and
suburbanization have also been observed in the cities of developing countries, and new
urban elements such as suburban shopping malls and new towns have emerged too,
which has been cited as the evidence of the urban convergence hypothesis that claims
cities around the world are becoming alike and converging to a set of socio-spatial
attributes similar to those of Western cities (Cohen, 1996; Dick and Rimmer, 1998; Ma
and Wu, 2005a). However, the convergence thesis has been widely challenged and
criticized. As Ma and Wu (2005a) as well as Freestone and Murphy (1998) argued,
despite a general convergence of suburbanization and metropolitan re-centering trends
across countries, the nature of the urban forms emerging, the underlying driving forces
and the processes involved are culturally and historically specific in different countries
and embedded in local economic and political systems. So, some comparative analysis
looking across the different contexts of developed and developing economies to
understand more thoroughly the changing spatial structure of contemporary metropolitan
areas is still needed and of great interest.
Comparative studies are important and necessary because they shed light on the
spatial characteristics of cities under different regulatory regimes, and help understand

3
how urban forms transform driven by not only the market forces, but also the changing
regulations. For instance, through the analysis of the structure of Russian cities, Bertaud
and Renaud (1997) found a totally different spatial pattern of the socialist city - a
perversely positive density gradient - where urban land was allocated administratively in
the absence of land markets, contrasted with the negative gradient normally seen in the
market city. In a study of the evolution of employment centers in Seoul, Jun and Ha
(2002) also revealed that the number, size, and characteristics of subcenters in the cities
of developing countries may be different from those in Western metropolitan areas, and
possibly play different roles, and the government rather than the market forces plays the
key role in determining the initial location and development pace of urban subcenters.
From a policy perspective, comparative studies may also provide insights on whether the
planning and policy experience of Western urban development is applicable or even
relevant to the cities of the developing world, or the cities of developing countries have
their peculiar trajectory of urban evolution that calls for different planning and policy
approaches. This dissertation is seeking to contribute in this way through a case study of
Beijing, one of the largest Chinese cities with rapid growth and spatial transformation
under the changing regulatory regime in the post-reform era.
Scholars engaged in China’s urban development have devoted great efforts to
understanding the changing spatiality of post-reform Chinese cities in recent years. Their
studies have shown that the spatial pattern of Chinese cities in the post-reform era since
the 1980s has changed dramatically and is greatly different from that of previous eras.

4
Especially in the 1990s, with introducing the land and housing market in urban China, the
development process of Chinese cities has become more driven by market forces, which
facilitated the growth of post-reform urbanism in Chinese cities, characterized by the
rapid urban expansion and decentralization, and increasing socio-spatial stratification, etc.
Some studies, e.g. Zhou and Ma (2000), have revealed that since the late 1980s large
Chinese cities have witnessed the relocation of both urban residents and industries from
the inner city to the suburbs, a trend broadly analogous to that characteristic of the
suburbanization in Western cities since the World War II, although the driving forces and
the process are embedded in a quite different context. Moreover, with the rapid
decentralization and suburban growth, the compact urban form of pre-reform Chinese
cities has also been replaced by a more dispersed and polycentric spatial pattern in
today’s Chinese cities (Wu, 1998a; Wu and Yeh, 1999). However so far, little is known
about how the urban space has been restructured with the process of decentralization in
urban China, e.g. whether the traditional monocentric urban form has been transformed to
a polycentric structure or still remains. And, planning issues in contemporary Chinese
cities have, by and large, still not been addressed. These include the impacts of the
decentralized urban form on urban density patterns, land values, housing development,
transport demands and demands for public goods. They are important issues that motivate
further detailed empirical investigations.
This dissertation follows the literature discussed above and presents an empirical
analysis of spatial trend and pattern of Beijing, one of the largest Chinese cities, in the

5
post-reform era. Our study provides findings on the patterns and processes of urban
decentralization and restructuring from a developing and transitional economy context
through the case of Beijing, and offers further understanding of the spatial organization of
contemporary urban areas based on the evidence that departs from the North American
and European experience.
As the capital city of China, Beijing has long been the focus of research on
China’s urban development. In recent years, a large number of empirical studies covering
a variety of topics have been conducted to understand better the changing urban form of
Beijing. It has been shown that departing from the socialist model of urbanization in the
pre-reform era, Beijing has experienced dramatic urban changes bearing similarities to
both Western cities and cities in other developing countries in the post-reform times (Gu
and Shen, 2003). Several studies have shown that with the urban land reform that
introduced land values and market to the Chinese city, urban pattern with a declining
density gradient of population, land values and property prices similar to that of Western
cities is also evident in contemporary Beijing (Zheng and Kahn, 2008; Han, 2004; Wang
and Zhou, 1999). These studies also revealed that the current spatial pattern of Beijing is
by and large centralized and can be largely explained by the monocentric model, although
minor suburban centers may have emerged. However, due to the lack of adequate data,
the distribution pattern of employment in Beijing has been less studied, and as for the
spatial organization of economic activity in the city, studies mainly focused on the retail
sector. For instance, Wang and Jones (2002) studied the retail structure of Beijing and

6
found that retail facilities are highly concentrated in the urban core and new
developments mainly take peripheral locations outside the third or even fourth ring road
in conjunction with the development of new suburban communities.
The persistent spatial trend revealed in the literature for the development of
Beijing since the 1980s has been the rapid urban expansion and the trend of
suburbanization, studied mainly from the perspectives of population redistribution and
land use changes (Zhou, 1997; Xie et al., 2007). Accordingly, the suburbanization of
Beijing was first driven by the government-led relocation of inner-city residents and
factories in the 1980s, and more driven by market forces recently (Zhou and Ma, 2000;
Feng et al., 2008). Although some scholars, e.g. Feng et al. (2008), have argued that the
new round of suburbanization in Beijing since the 1990s is more market-oriented,
planning and land market institutions still play fundamental roles in the process (Deng
and Huang, 2004). So the decentralization and restructuring of urban space in Beijing are
better understood in terms of the interaction between planning interventions and market
forces, but the underlying dynamics of these interactions are still far from clear.
The socio-spatial transformation of post-reform Chinese cities has also posed
fundamental challenges to the practice of urban planning in today’s Chinese cities (Wei,
2005; Ma, 2004; Yeh and Wu, 1999). Comprehensive planning has been a major tool
used by the municipal government to control and guide urban development in Beijing in
the post-reform era. However, the approach has been shown to be insufficient as an
effective way to guide the development of a city like Beijing that is undergoing rapid

7
urban growth and transformation with great uncertainties, because it is too static and the
plan is usually updated slowly and often lags behind actual development. Besides,
planning practices in Beijing as elsewhere in China have also been criticized for placing
too much emphasis on physical and land use planning. As pointed out by Song et al.
(2006), the deficiency of over-relying on physical planning in the urban planning of
Beijing can be illustrated by the gap between the planned and actual distributions of jobs
and housing and its resulted negative impacts on transportation. Physical planning
approach emphasized jobs-housing balance through configuring the physical form and
land use pattern of the city. However, since the establishment of the land and housing
market, the location choices of urban residents and firms have been more driven by
market forces, resulting in the improved residential mobility and the increasing spatial
separation between places of residence and jobs in the city that made the existing planned
organization no longer function effectively. Although the distributions of people and jobs
in Beijing are more driven by market forces nowadays, planning practices have been less
responsive to these forces, which led to the futility of such planning practices and called
for the new approach that can integrate the market mechanism.
In the latest master plan of Beijing, a polycentric spatial pattern has been planned
as the target to solve various challenges that the city is facing (please refer to section
1.4.3 for more details), such as accommodating rapid urban growth as well as mitigating
the deterioration of traffic conditions and protecting cultural, historical, environmental
and social capital, among other goals. It is not clear that it is compatible to the current

8
spatial structure of the city. Meeting these goals and developing better policies to manage
urban growth effectively in Beijing will require better understand the nature of urban
space under the marketization, which is one major objective of our study.
This dissertation is by and large an empirical and explorative study that provides a
thorough and systematic analysis on the spatial trend and pattern of the Beijing
metropolitan area in the post-reform era from the perspective of population and
employment distributions and their interactions. Our study takes a comparative
perspective that highlights how the Chinese urban transformation differs from the
Western experience, and applies the approaches of urban economics instead of the
framework of political economy analysis often used in the literature of China’s urban
transformation. We aim to fill the gap in the literature in two ways. First, this study
provides evidence on the spatial evolution of contemporary metropolitan areas from a
developing and transitional economy context, complementary to the existing empirical
literature of urban spatial evolution. Second, we define urban spatial structure as the
distributions of both population and employment, in other words, people and
socioeconomic opportunities (jobs) in our study, and study their interactions through
analyzing the commuting pattern, which tend to provide a more integrated and complete
image of the spatial structure of metropolitan Beijing than previous studies in the
literature, with the focus of our analysis on questions that have not been addressed mainly
due to the lack of adequate data. Furthermore, our findings are also useful to help better

9
understand the changing nature of the urban form of metropolitan Beijing under the
marketization and transition, and have explicit planning implications.

1.2 RESEARCH QUESTIONS
Following the approach of urban economic analysis, urban spatial structure can be
defined as the spatial distribution of people and jobs, as well as the transportation system
that connects people to jobs in the metropolitan area (Shen, 2000). Therefore, an
empirical study analyzing urban spatial structure usually starts with identifying the
distributions of urban population and employment. Meanwhile, of course, the
distributions of people and jobs influence each other, as the former are the labor force and
the product markets for the latter and the latter provide economic opportunities for the
former. Theoretically, people and jobs tend to locate near to each other, but in reality
there is spatial separation between place of residence and workplace, and commuting -
the movement between living and work places - bridges the gap between the two
distribution patterns (Sohn, 2002). Accordingly, our study is based on this notion and
expressed as a conceptual scheme in Figure 1.1. It investigates the spatial structure of the
Beijing metropolitan area in the post-reform era through examining the spatial trend and
pattern of population and employment distributions and their interactions by analyzing
jobs-housing balance and urban commuting.

10
Figure 1.1 Research Scheme

Under this framework, a set of specific research questions are addressed as
followed:
1) Characterizing the spatial pattern and evolution of population distribution in
metropolitan Beijing, 1982-2000
How has the population been distributed within the metropolitan area at
different spatial scales? How have those patterns evolved over time? In
particular, to what spatial extent has the decentralization occurred?
Is there any spatial disparity shown in the process of decentralization?
How does this variation relate to the overall development of the city?
How has the urban spatial structure transformed with the population
decentralization? In particular, has the population uniformly dispersed or
clustered to form subcenters while it spread into the suburbs?
Transportation Network
Urban Spatial Structure
(Land Use Patterns)
Population Distribution
(Residential Areas)
Employment Distribution
(Industrial and Business Areas)
Labor Market
Consumer Market
Commuting Commuting

11
2) Characterizing the spatial pattern of employment distribution in metropolitan
Beijing, 2001
What is the overall distribution pattern of employment throughout the
Beijing metropolitan area at different spatial scales? Is it different across
economic sectors? How do the levels of agglomeration of economic
activity in different sectors for multi-spatial scales differ from each other?
Has the employment decentralized as the population in the Beijing
metropolitan area? If so, how has the decentralization of employment
affected the spatial structure of metropolitan Beijing? In specific, have the
employment subcenters emerged, what are the major characteristics of the
subcenters, and what factors account for the emergence of the various
types of employment centers?
3) Jobs-housing balance and urban commuting in metropolitan Beijing
Does the commuting pattern reveal a balanced urban structure of Beijing?
Has urban spatial structure, in terms of jobs-housing balance, an important
effect on commute duration, and what are the main factors that explain
variations in commute time when both socioeconomic characteristics of
commuters and urban structure are considered?

12
1.3 ORGANIZATION OF THE DISSERTATION
The remainder of the dissertation is organized as followed. Chapter 2 provides a
review of the literature along the two lines: theories and empirics on the evolution of
urban spatial structure and its relationship to urban commuting, and the spatial
restructuring of post-reform Chinese cities. This chapter aims to set the background and
benchmark for our study through summarizing the previous research related to ours, and
to highlight the contribution of our study to the literature.
The main research occupies Chapters 3 through 5. Chapter 3 focuses on the
spatial trend and pattern of population distribution in the Beijing metropolitan area from
1982 to 2000. Compared with the previous studies, this chapter extends the study period
to the 1990s, and employs more flexible analytical techniques in addition to the classic
density function approach. The focus of the analysis is on the questions that have been
less addressed in previous studies, such as the spatial extent, spatial variation and
structural transformation of the population decentralization in metropolitan Beijing. This
chapter aims to better understand the changing characteristics of population distribution
within the Beijing metropolitan area, given the context of the rapid urban growth and the
economic and societal transition.
Chapter 4 studies the employment distribution and the spatial organization of
economic activity in the Beijing metropolitan area, based on an original dataset drawn
from the 2001 basic establishment census of Beijing. Two general approaches are applied
and distinguished in this chapter. The first concerns the overall distribution and

13
agglomeration of employment throughout the metropolitan area and the second focuses
on the local structure and subcentering of economic activity. Both approaches are built on
some geostatistical techniques. This chapter aims to build up some stylish facts about the
spatial organization of contemporary Chinese metropolitan areas in terms of employment
distribution through the case study of metropolitan Beijing.
Chapter 5 further investigates the spatial structure of the Beijing metropolitan area
by analyzing detailed individual-level survey data pertaining to the commuting pattern.
We compare the distribution patterns of population and employment throughout the
metropolitan area by analyzing jobs-housing balance, and further examine the spatial
structure within the central city based on both the distributions of population and
employment as well as the commuting pattern. To examine urban structure at different
spatial scales helps better understand the nature of the urban form of metropolitan Beijing.
Finally, we focus on the planning debate over the effectiveness of land use planning to
transportation problems, and examine the main factors that explain variations in commute
times in the central Beijing. This chapter aims to better understand the nature of urban
spatial structure of Beijing and its impacts on urban commuting and to provide
implications for planning practices in Beijing.
At last, Chapter 7 summarizes major findings of our study and discusses the
implications of the results for the planning of the Beijing metropolitan area. Some
additional needs and themes for future research are also discussed.

14
1.4 STUDY AREA
1.4.1 Specifying the Study Area’s Boundary
It is important to understand the administrative organization of Chinese cities
before we specify the boundary of the study area. The city (shi) is an officially designated
urban administrative entity in China, which contains both urban and rural areas (Zhang
and Zhao, 1998). Especially after the reorganization of the Chinese urban administrative
system since the economic reforms, which has been well documented in Ma and Cui
(1987) and Ma (2005), a large rural area has often been included within the
administrative boundary of a city. As Ma and Cui (1987) argued, the so-called “city” is
actually a city region administrated by the central city, and this region is also called the
administrative region (xing zheng qu). This dissertation refers to the municipality
covering the whole administrative region of Beijing as Beijing or “the city”.
With a total area of 16,808 km
2
, Beijing had 13 districts and 5 counties under its
jurisdiction in 2000. These districts and counties are normally divided into three zones:
the inner city, the inner (or near) suburb, and the outer suburb. The inner city is also
called the old city (lao cheng qu) that is the historic urban core. The inner city and the
near suburb together cover the continuously urbanized (built-up) area of the city, and
form the so called central city (zhong xin cheng) according to the latest master plan of
Beijing, which is comparable to the central city of the large metropolitan area in western
countries. The study area of this dissertation is the metropolitan area of Beijing, the
boundary of which is still not clearly defined in the literature. Considering the rapid

15
urbanization of the city during the past decades, we refer to the central city including both
the inner city and the near suburb as the urban area (cheng qu) of Beijing, and define the
metropolitan area to include the urban area and its adjacent outer suburb districts and
county. The urban area of Beijing consists of 8 districts (Dongcheng, Xicheng, Chongwen,
Xuanwu, Chaoyang, Haidian, Shijingshan, Fengtai) with an area of 1,370 km
2
(8.1% of
Beijing’s total area), and is the urbanized core of the city. The metropolitan area contains
5 outer suburb districts (Changping, Shunyi, Tongzhou, Mentougou, Fangshan) and 1
county (Daxing) in addition to the urban area, and covers an area of 9,073 km
2
(54.0% of
Beijing’s total area). These outer suburb districts and county composing the fringe of the
metropolitan area are mostly partial urbanized or rural. (Figure 1.2)

16
Figure 1.2 The Topology, Road System and the Administrative Organization of Beijing

1.4.2 A Portrait of Beijing’s Urban Form
Beijing is located in the north China plain, which opens to the south and east of
the city. Mountains dominate in the north, northwest and west of the city, accounting for
62% of Beijing’s total area (Figure 1.2). The urban area is on flat land and situated in the

17
south-central part of the city, spreading out in bands of concentric ring roads. Tian’anmen,
to the south of the Forbidden City, former residence of the emperors of China, is the
urban center. Unlike most western cities with the CBD as the urban center, the heart of
Beijing is occupied by the old imperial city, which makes the city to develop two separate
central business areas in both the east and the west part of the inner city. The so-called
central business district (the CBD) in the Chaoyang District sits in the east and the
Beijing Financial Street in the Xicheng District sits in the west.
The spatial pattern of Beijing reflects its imperial traditions rooted in serving as
the capital city of China for nearly 800 years (Song et. al., 2006). The older part of the
city enclosed by the 2
nd
ring road, with a horizontally symmetrical layout and a grid street
system, overlays most of the inner city. The newer parts of the city, mainly built up after
the 1950s, have expanded outward in every direction from the historic city core to the
near suburb, resulting in the pancake-style urban form of Beijing (Deng and Huang,
2004).
The road system of Beijing is characterized by the concentric ring roads
connected by the radial highways, all built or expanded after the 1980s. This ring and
radial road system greatly impacts the current urban pattern of Beijing, and is often used
as a kind of reference system to define geographical locations and boundaries. The
continuously built up area of the city is mostly contained within the 5
th
ring road, which
can be roughly regarded as the edge of the central city. The 6
th
ring road was just

18
completed in 2008 and is the furthest from the urban core, designed to connect the major
new towns in the fringe of the metropolitan area. (Figure 1.2 and 1.3)
Figure 1.3 The Built-up Areas and the Ring Road System of the Central Beijing

1.4.3 Spatial Development and Planning of Beijing
As the capital city of China, Beijing experienced rapid economic and population
growth in the past decades. This rapid growth has been greatly changing the nature and
structure of the city. Especially after the 1990s, with the establishment and boom of the

19
land and housing markets, real estate development has become one of the fundamental
forces in shaping the physical pattern of the city. Market forces, together with the
planning efforts of the municipal government, are transforming Beijing from a socialist
capital city into a modernized international metropolis.
The spatial transformation of the city has been reflected in two aspects: the
renewal and reconstruction of the inner city and the rapid expansion and transition of the
urban periphery (Gu and Shen, 2003; Lei, 2007). With the restructuring and
internationalization of the urban economy, Beijing has endeavored to promote its image
as a global city and implemented large renewal projects in the inner city to improve its
physical environment for business, investment and living. Old factories and residential
buildings in the inner city were replaced by large scale commercial complexes, modern
apartment buildings, and the road and highway systems (Huang, 2004), which caused
substantial changes in the land use structure and urban landscape. Meanwhile, large scale
urban development projects have burgeoned at the urban periphery, expediting the sprawl
of the central city along the ring and major radial roads (Figure 1.4). Numerous
residential communities were constructed by developers in the suburbs to meet the
intense demands of housing generated with the increase of urban residents’ income and
the growth of urban population. With the industrial relocation from the inner city into the
suburbs, development zones were set up by governments for the concentrated industrial
development, accelerating the expropriation of the farmland and greenbelt on the urban
fringe and incurring the inefficient patterns of urban expansion (Deng and Huang, 2004).

20
The new development at the periphery has brought new centralities as various functional
centers to the urban structure; however, as the central area expanded rapidly between the
bands formed by the concentric ring roads, it assumed the adjacent centers and became a
larger agglomeration dominating the metropolitan area. So the spatial development of
Beijing is largely characterized by the continuous compactness and the rapid spread of
the central city (Lei, 2007).
Figure 1.4 The Spatial Expansion of Beijing

(Source: The Beijing Municipal Commission of Urban Planning, The Master Plan of
Beijing City, 2004~2020.)

21
Figure 1.5 The Spatial Development Scheme of Beijing

(Source: The Beijing Municipal Commission of Urban Planning, The Master Plan of
Beijing City, 2004~2020.)

As the rapid growth of the city has been far beyond the control of the previous
city plans (Huang, 2004; Song et. al., 2006), the municipal government of Beijing and its
planning commission has revised the master plan of the city, to accommodate to the new
social, economic and political conditions, and to address the development challenges that
the city has been facing. The latest master plan of Beijing for the period 2004-2020

22
provides new guidelines for the future development of the city, and plans the polycentric
structure with two urban axes in the central city, two belts for ecological conservation
and economic development respectively, and multi-centers both in the central city and
throughout the metropolitan area (Figure 1.5).

1.5 SUMMARY
From a comparative international perspective, this dissertation explores the spatial
distributions of population and employment in the Beijing metropolitan area in the post-
reform era. This study aims to extend the literature on urban spatial evolution, with
special reference to the pattern and process of urban decentralization and restructuring
from a developing and transitional economy context, and to offer further understanding
of the spatial organization of contemporary urban areas that departs from the North
American or European experience.
Beijing is a transition city that has experienced dramatic urban growth and spatial
restructuring since the reforms in China, and its experience sheds light on how the urban
spatial structure changes within a hybrid of an evolving market economy with a central
government that retains significant control. With the emerging urban land and housing
market, the development process of post-reform Beijing has become more market-
oriented. Therefore, it is interesting to examine whether the Beijing metropolitan area is
becoming alike to its Western counterparts as argued by the urban convergence
hypothesis, or it has its peculiar trajectory of urban evolution. Some specific questions

23
are raised as well, such as how the urban pattern and development process of Beijing is
different from cities in Western countries; what drives the common features of the urban
form of Beijing analogous to those of Western cities, if any; what features of Chinese
planning or regulatory framework manifest themselves in what we find in Beijing; does
the era of development matter, etc.
To address these questions, this study focuses on the distribution patterns of
population and employment in metropolitan Beijing. Typical empirical tools used in the
literature often pose the restrictive assumptions of the urban structure (e.g.
monocentricity, symmetry), which may lead to biased measured outcomes and mask
significant internal dynamics in the spatial distributions of population and employment.
In this study, we apply more flexible techniques, such as nonparametric analysis,
geostatistical techniques, and demonstrate more nuanced dynamics than those discussed
in previous studies.
Our study finds that the spatial pattern of the Beijing metropolitan area is
becoming alike to those of large Western cities in the post-reform era, with the compact
urban form in the pre-reform era replaced by a more dispersed and polycentric spatial
pattern. The overall trend toward the decentralization and polycentrification of both
population and employment is also evident in the Beijing metropolitan area in the post-
reform era. However, compared with the decentralization of large Western metropolitan
areas, the extent of the decentralization of metropolitan Beijing is quite limited. We show
that both people and jobs that moved out of the inner city tend to re-concentrate in the

24
near suburbs adjacent to the central area instead of dispersing throughout the
metropolitan area. The rapid growth of the near suburbs has expedited the expansion of
the central city, with a larger central agglomeration emerged dominating the whole
metropolitan area. In this broad sense, the spatial pattern of the Beijing metropolitan area
is still highly centralized, and the tendency toward decentralization at the level of the
metropolitan area is questionable.
Even though the spatial structure of Beijing is largely characterized by
monocentricity, our study does provide the evidence that significant population and
employment subcenters do have emerged in the suburbs of Beijing. However, the number
and size of subcenters are small, and the pattern of the subcenters in metropolitan Beijing
is highly adherent to the development scheme of the city, so the polycentricity emerged
in the Beijing metropolitan area is very different by nature from that observed in Western
cities, and has different origins. Although the common features of spatial pattern and
trend broadly analogous to those of Western cities have been observed in post-reform
Beijing, the driving forces and the process involved still need be understood with
reference to the peculiar Chinese context, and the similar factors that caused the
suburbanization in the West have taken their effects on the suburbanization of Beijing in
a totally different context.

25
CHAPTER 2.
LITERATURE REVIEW

2.1 EVOLUTION OF URBAN SPATIAL STRUCTURE
2.1.1 Review of Urban Economic Theories
The evolution of intra-urban spatial structure has been long of interest to urban
economists. Urban economic theories have shed important lights on the nature of
changing urban form. Standard urban economic theories suggest that the fundamental
determinant of the spatial structure of cities is a trade-off between two opposite forces:
the propensity of economic agents to interact (the cost of interaction) and their aversion
to crowding (the cost of congestion), and different combinations of the forces engender
different spatial distributions of agents, such as an even or concentrated pattern, or a
monocentric or polycentric configuration (Papageorgiou and Pines, 1999).
Monocentricity is a spatial character of the “nineteenth century city” (Anas et al.,
1998), consisting of a compact production core surrounded by an apron of residences,
which has been well modeled in the urban literature since Alonso (1964), Mills (1967)
and Muth (1969), and has dominated urban economics for nearly three decades. The
monocentric model provides important insights about urban spatial structure and its
evolution. In this model, production is concentrated at the central business district (CBD),
and rents are high near the center while commuting costs are low. So, locational choice is
solely based on the distance to the employment center. In the equilibrium, households

26
living at the central locations will consume small quantities of housing and spend little on
commuting, while households which commute longer distances will consume more
housing that is cheaper at more distant locations (Mieszkowski and Smith, 1991). Under
certain assumptions, the model implies a declining population density with distance from
the CBD. Based on the model, the empirical studies also find the declining employment
density and the density gradient is larger than that for population but has been falling
faster (Mieszkowski and Mills, 1993), which partly supports the hypothesis of the greater
centrality of employment than that of population.
The comparative statics of the model suggest that population growth, or changes
in real incomes or in commuting costs result in changes in equilibrium urban structure
(Wheaton, 1974). So, the model can well explain the general trend of decentralization of
cities over the last century or more. According to the model, rising incomes and declining
transportation costs cause the density gradient to decline, which accounts for the
decentralized evolution of urban structure. But the simple model predicts an increasing
density gradient with population growth, which suggests larger cities are more centralized
and contradicts empirical evidences. Papageorgiou and Pines (1999) argue the
comparative statics of the basic monocentric model are usually based on constant returns
to scale in production, while, in a case where production exhibits scale economies, they
show the gradient can decrease with increasing population size. Mills and Tan (1980)
also suggest the observed negative correlation between the density gradient and
population size may be due to the polycentric structure of large cities.

27
The monocentric model has been increasingly criticized recently for its
inadequacy to describe the spatial pattern of large modern urban areas, where
decentralization of population and employment has taken a more polycentric form, with
the emergence of suburban subcenters independent or subsidiary to the older CBD (Anas
et al., 1998). Several studies, e.g. Clark and Kuijpers-Linde (1994), Kloosterman and
Musterd (2001), and Champion (2001), have noted that economic restructuring in the
globalization era, the development of new transport and information technologies, and
changing household composition and commuting patterns all contribute to the current
change in urban structure, which has undermined some underlying assumptions of the
monocentric model. Furthermore, the standard static monocentric model assumes the
urban center and its location to be exogenous, which makes it not useful in explaining
how urban agglomerations become monocentric or polycentric, although it can be
extended to a polycentric context to examine the effects of establishing subcenters on
urban spatial structure, as in White (1976).
To investigate how urban spatial structure evolves to a monocentric or polycentric
pattern, we need models that allow for endogenous clustering of economic activity.
Beckmann (1976) is among the first to consider urban structure without a predetermined
center. He derives a bell-shaped population density profile, which peaks at an
endogenous urban center, but some limitations of his model have excluded the
polycentric pattern as a possible outcome. Fujita and Ogawa (1982) provide a more
thorough analysis and advance the understanding of urban monocentricity versus

28
polycentricity. Their model is capable of yielding multi-centric patterns as well as
monocentric and non-centric patterns, and there are multiple equilibriums under the same
parameter values, which suggests the model may not yield unambiguous hypotheses to
relate the subcenter formation process with major parameters, such as commuting rate for
the households or production level and locational potential parameters for the business
firms. But, as summarized by McMillen and Simth (2003), through simulations, the
Fujita-Ogawa model suggests that the equilibrium number of subcenters is likely to
increase with population and the per-unit cost of commuting. This hypothesis has been
empirically supported by McMillen and Smith (2003). The comparative statics of the
Fujita-Ogawa model is quite different from those of the monocentric model. For example,
in their model, the monocentric city can only be sustained under a certain level of
commuting cost, and beyond that level, the city is dispersed and new and smaller centers
emerge, which suggests the effect of commuting cost is no longer monotonic
(Papageorgiou and Pines, 1999). Actually, this catastrophic structural transition of the
urban configuration when the parameters take critical values has been a main finding of
Fujita and Ogawa’s (1982) analysis.
Fujita and Ogawa (1982) view urban configurations as determined by decisions
made by atomistic agents, but some other studies suggest different mechanisms for
subcenter formation. Henderson and Mitra (1996) develop an edge city model to adapt
the Fujita-Ogawa model to include land developers and a history. They argue large land
developers play an important role in the formation of “edge-city” type subcenters, as

29
suggested by Garreau (1991). These developers initiate massive planned private
developments on the scale of medium-size cities, manipulating the decisions of the
atomistic agents, which leads to equilibriums typically not considered in a purely
atomistic world (Henderson and Mitra, 1996). Another mechanism has been suggested by
Fujita et al. (1997), which claims the emergence of a subcenter as the result of the
establishment of a large firm within a small city. They show since the firm is large
relative to the city size, the entrance is likely to affect drastically both the local labor and
land markets, and the total impact ultimately depends on the location choice of the
entrant (Fujita et al., 1997).
In the static model, the structural transition of the urban configuration has been
analyzed through comparative statics, but it is more appropriate to be examined in a
dynamic model. Helsley and Sullivan (1991) develop a model to formalize a polycentric
city as a system of employment centers and subcenters within a growing metropolitan
area. They show in a growing city, subcenters arise from the tradeoff between external
scale economies in production and diseconomies of scale in transportation. And their
model predicts a development pattern of spatial structure with three phases: an initial
phase of exclusive central-city development, a second phase of exclusive subcenter
development, and a final phase of simultaneous development. The model replicates the
stylized fact that subcenters generally form after the dominant city center has been
established (Helsley and Sullivan, 1991). Their model has been further refined by Sasaki
and Mun (1996).

30
A brief review of theoretical literature has shown the recent development of urban
economic theories has moved far beyond monocentric models. Empirical literature also
provides ample evidences for the qualitative change of spatial structure of contemporary
metropolitan areas (Anas et al., 1998). While theoretical models focus on examining the
equilibrium spatial configuration of polycentric cities, empirical studies identify
subcenters and examine impacts they have on land values, population and employment
distributions, and travel patterns (McMillen and Smith, 2003). Some empirical
regularities are evident and summarized as follows.

2.1.2 Empirical Regularities
A sketch of how urban form evolved in modern times has suggested that cities
have kept spreading out and the decentralization of population is a common trend. Some
cross-country studies find the growth and suburbanization of metropolitan areas have
been international trends, which are the most notable in the U.S. (Mieszkowski and Mills,
1993). Mieszkowski and Mills (1993) summarize two theories to explain suburbanization:
the natural evolution theory that emphasizes rising real incomes and changes of intra-
urban transportation over time, and the “flight from blight” theory that stresses fiscal and
social problems of central cities. Glaeser and Kahn (2003) suggest the technological
superiority of the automobile as the root cause of U.S. sprawl, which they believe has two
fundamental effects on population decentralization: to reduce transport costs and to
eliminate the scale economies involved in older transportation technologies.

31
Glaeser and Kahn (2001) claim that the decentralization of the U.S. cities
proceeds in two waves: the decentralization of population followed by the
decentralization of employment. Mieszkowski and Mills (1993) show that 57 percent of
metropolitan residents and 70 percent of metropolitan jobs were located in central cities
in the U.S. in the 1950s, while the percentages were only about 37 and 45 in 1990.
Glaeser and Kahn (2001) also find most metropolitan areas across the U.S. were
remarkably decentralized in 1996, with less than one-quarter of their employment within
three miles of their CBD. They further test and support the hypothesis that jobs have been
following people, and argue spatial patterns of cities nowadays seem to be driven as
much by consumption advantages experienced by workers as by the productivity
advantages of particular locales for firms.
The decentralized spatial structure of metropolitan areas has taken two forms: a
polycentric pattern that emphasizes the concentration of employment and commercial
activities within subcenters, and a dispersed pattern that emphasizes the generalized
dispersion nature of decentralization. The subcentering phenomenon characteristic of
polycentricity has been well documented in the U.S. Garreau (1991) identifies 123
existing and 77 emerging edge cities in the 35 largest U.S. metro areas and provides a
journalistic interpretation of polycentric structure within large cities. Multiple subcenters
have also been identified using various methods in many of the largest U.S. cities, among
which the most widely examined cases are Los Angeles (Giuliano and Small, 1991; Song,
1994; Gordon and Richardson, 1996; McMillen, 2001) and Chicago (McDonald and

32
McMillen, 1990; McDonald and Prather 1994; McMillen and McDonald, 1997;
McMillen, 2001). As there exists no generally accepted method of defining metropolitan
employment subcenters, studies even working on the same area have tended to yield
different results. For example, the number of subcenters in Los Angeles identified by the
studies mentioned above varies from 6 to 32. And the results are also sensitive to
definition and spatial units used in the analysis. Certain employment clusters can be
viewed as several large subcenters or one gigantic mega-center using different criteria.
With defined subcenters, several studies, e.g. Small and Song (1994), and
McDonald and Prather (1994), have found the polycentric model a better fit relative to
the monocentric model, and subcenters help explain density and land-value patterns in
large U.S. metro areas. McMillen (2004) uses McMillen and Smith’s (2003) subcenter
lists for 62 large U.S. metro areas to explain the spatial distribution of employment
density within these metro areas, and he finds that the subcenter distance variable is
statistically significant in all but five cities. Some studies, e.g. Giuliano and Small (1991),
and Anderson and Bogart (2001), also show the employment centers within metropolitan
areas tend to form an interdependent system, with a size distribution and a pattern of
specialization analogous to the system of cities in a larger regional or national economy
(Anas et al., 1998). This suggests decentralization and subcentering of employment
represent a systematic change of metropolitan structure rather than a random sprawling of
firms (Anderson and Bogart, 2001).

33
Polycentric urban forms have also been observed in other countries, such as
Canada (Coffey and Shearmur, 2001), Australia (Freestone and Murphy, 1998), and
Korea (Jun and Ha, 2002), although they are less advanced than in the U.S.. Despite a
general trend of decentralization and subcentering across countries, the forms emerged
and the processes involved may be different (Freestone and Murphy, 1998). Jun and Ha
(2002) note that market forces may be the dominant factors in subcenter formation in
developed countries, but the state plays a more important role in determining the
formation of subcenters in developing countries.
Subcenters emerge for the same reasons that explain the formation of the CBD, to
exploit agglomeration economies through clustering. But, if the benefits from clustering
in subcenters diminish or even subcentering is as costly as locating in the CBD, the
generalized dispersion of firms may be more plausible (Fulton, 1996). Gordon and
Richardson (1996) argue, by virtue of ubiquitous auto access, dispersion mitigates
congestion costs, while the benefits of agglomeration can still be enjoyed from most
dispersed locations. So, the advantages of location in centers are diminishing (Lee, 2007).
Their rationale for “generalized dispersion” has been supported by the fact that more than
80 percent of employment located outside centers in Los Angeles, and the percentage had
increased from 80 to 88 over two decades by 1990 (Gordon and Richardson, 1996). This
dispersion view has also been supported by the notion “edgeless cities” introduced
recently by Lang and Lefurgy (2003).

34
But it is still arguable whether polycentricity or generalized dispersion
characterizes the nature of emerging urban spatial structure. Giuliano and Redfearn (2005)
examine spatial trends of employment concentrations in Los Angeles from 1980 to 2000,
and argue that agglomeration economies at the intra-metropolitan scale continue to be a
significant organizational factor in the space economy. Lee (2007) examines spatial
trends in selected U.S. metropolitan areas and finds that generalized dispersion was more
common than subcentering in the 1980s and 1990s, and three urban patterns (dispersed,
polycentric and monocentric) coexisted. He suggests that decentralization patterns of
employment may be distinct for cities due to their different histories and circumstances.
The traditional CBD turns out to be less important as the spatial pattern of
employment in modern metropolitan areas is becoming dispersed or polycentric. But
decentralization has not eliminated the importance of the CBD. Several studies, e.g.
Small and Song (1994), and Cervero and Wu (1998), have indicated that the CBD is still
the largest and densest urban center, and usually has a larger effect on surrounding
densities and land prices than does any subcenter (Anas et al., 1998). McMillen (2004)
also finds that the traditional CBD distance gradient remains statistically significant in
explaining the spatial distribution of employment density within all 62 metro areas as of
1990. This suggests the monocentric city model may not be obsolete (Papageorgiou and
Pines, 1999). Central city locations still matter because they facilitate innovative
activities based on the exchange of richly layered information demanding a high
frequency of face-to-face contacts (Kloosterman and Musterd, 2001). Glaeser and Kahn

35
(2001) find that the relative centralization of skilled industries appears to be a fairly
robust phenomenon in their study, so they argue the primary force fighting against
decentralization in modern urban areas seems to be the advantage of urban centers in
speeding the flow of ideas. Besides, some very large cities, especially in Europe and Asia,
retain a strong urban center, which provides attractions for prestige retail, entertainment
and culture, due to their historical and cultural accumulation.
In summary, despite the broad consensus in empirical literature on the overall
trend of metropolitan decentralization and subcentering, it is still less known about the
nature and form of this spatial change (Lee, 2007). Furthermore, existing empirical
evidences so far about the emergence of polycentric urban forms have largely been drawn
from North American and European experiences, comparative analyses are still needed
from a developing country context.

2.2 URBAN SPATIAL STRUCTURE AND COMMUTING
A critical aspect of understanding changing spatial structure of contemporary
metropolitan areas centers on its impacts on changes in urban transport generally and
commuting specifically (Clark and Kuijpers-Linde, 1994). How the continuing
decentralization and polycentrification affect travel behavior in large metropolitan areas
has sparked a lively and longstanding policy debate in recent times. However, the
evidences provided by the empirical literature are still controversial and the relationship

36
between urban form and travel patterns, and in specific, the effect of spatial structure on
urban commuting, is still poorly understood.
As the monocentric city becomes inefficient with urban growth for increasing
congestion in the urban center, polycentricity has been regarded as a more efficient
spatial form that helps to reduce commuting and congestion costs and to improve
mobility and accessibility (Clark and Kuijpers-Linde, 1994). The hypothesis is that with
decentralization of both residences and jobs, firms that desire accessibility to the labor
force locate with reference to where potential employees reside, which leads to the jobs-
housing balance that helps maintain constant commuting durations and distances
(Cervero and Wu, 1998), and meanwhile, the emergence of suburban employment centers
also conduces residential site choices with shorter commute times. But empirical studies
on the effects of polycentricity on commuting provide contradictive evidences. Some
studies, e.g. Gordon et al. (1989, 1991), show decentralization is associated with constant
or shorter average commutes through subcentering, while others, e.g. Rosetti and
Eversole (1993), show the opposite results.
Using data from the San Francisco Bay Area, Cervero and Wu (1998) show that
paralleling the region’s subcentering trend has been a substantial increase in average
commute vehicle miles traveled per employee between 1980 and 1990. And using
decomposition analysis, they find that increasing commute distances contribute the most
to rising commute vehicle miles traveled per employee. More recently, with data from the
1998 Netherlands National Travel Survey, Schwanen et al. (2003) indicate that the role of

37
urban form in the explanation of commute times is limited in comparison with micro-
level personal and household characteristics, and car commuting times are shown to be
higher in most polycentric systems. Through analyzing detailed locational data pertaining
to commuting patterns in a medium-sized metropolitan area, the Quebec metropolitan
area, Vandersmissen et al. (2003) also find that once travel mode and key social factors
are controlled for, the shift from a monocentric to a dispersed city form is responsible for
increasing commuting time. These findings all provide the evidences that are contrary to
the argument that suburbanization of jobs maintains stability in commuting duration.
Nevertheless, through a case study of the Randstad and the Southern Californian urban
region, Clark and Kuijpers-Linde (1994) argue that increasing automobile dependence
and an increasingly polycentric urban structure are inevitable, and during the transition to
a polycentric region, there will be increased commuting times and congestion, but over
time and on even greater scale, the development of polycentric urban structure may avoid
severe traffic diseconomies in large metropolitan areas through spatial restructuring.
In this section, we do not aim to provide a comprehensive review of the vast body
of literature exploring the relationship between urban form and commuting. Specifically,
we restrict ourselves to a brief summary of one strand that focuses on the commuting
impacts of balancing jobs and housing, to situate our own study on the Beijing
metropolitan area appropriately. Commuting is still the major source of congestion and
air pollution in large cities nowadays. Journey to work research sheds light on several
important issues in urban planning and policies. Many planners hold the view that land

38
use patterns fundamentally affect commuting behavior. The locations of jobs and housing
reflect development patterns of cities conditioned by planners and policymakers. And, the
jobs-housing balance, a term that is used to describe the relative locations of jobs with
respect to housing in a given area (Giuliano and Small, 1993), has been considered as a
plausible strategy to mitigate congestion and related environmental problems in large
metropolitan areas, through reducing the spatial separation between workplace and
residence, which consequently decreases excess commuting. However, the merits and
implications of the jobs-housing balance have still been widely debated (Horner, 2004).
Cervero (1989) is among the first to correlate long commutes in suburban areas
with a severe jobs-housing imbalance, and to promote the jobs-housing balance strategy
for combating growing traffic congestion and air pollution in American cities. But his
view has been questioned by other researchers. In the explanation of their finding of “the
commuting paradox”, Gordon et al. (1991) suggest that average commute times may be
contained in metropolitan areas by the location adjustments that households and firms
make, without the need for planning interventions, which implies market forces are best
suited for bringing about the balance. Similarly, in a study investigating the relevance of
jobs-housing balance to transportation problems, Giuliano (1991) indicates that jobs-
housing balance occurs as part of the urban development process, and commuting
patterns are not closely related to jobs-housing balance, which depresses the viability of
jobs-housing balance policy. However, as Levine (1998) argues, critics of jobs-housing
balance as transportation policy have identified policies for jobs-housing balance with

39
regulation, while a more consistent policy conclusion would identify these policies with
the relaxation of such regulation on suburban land use. His study argues that the
significance of jobs-housing balance is not in reducing congestion, whereas the goal of
the jobs-housing balance approach to transportation problems is better framed as
supporting a broader range of residential and transportation choices for individuals and
households.
So far empirical studies have produced polarized conclusions regarding the
relationship between jobs-housing balance and urban commuting. Through examining
commuting patterns for the Los Angeles region in 1980, Giuliano and Small (1993) find
jobs-housing balance, whether measured by the ratio of resident workers per job in a
broad subarea or by the required commuting time, has a statistically significant but not
very large influence on actual commuting times, and they conclude that commuting
distance and time are not very sensitive to variations in urban structure, and policies
aimed at changing the jobs-housing balance will have only a minor effect on commuting.
Based on a study that tracks changes in commuting patterns between 1984 and 1990 for
30,000 employees of a major health care provider in Southern California, Wachs et al.
(1993) also suggest that choices of residential location for these employees are influenced
by many factors in addition to the home-work separation, and they find little evidence to
support the argument that the jobs-housing imbalance increases commuting distance or
time. In follow-up work, Cervero (1996) examines changes in jobs-housing imbalance in
the San Francisco Bay Area during the 1980s, and he finds little association between

40
jobs-housing balance and self-containment. And in another study that examines the
association between how self-contained new towns are and how their residents and
workers commute, drawing on experiences in the U.S. and the U.K. (Cervero, 1995), he
also finds that jobs-housing balance and self-containment matter little in shaping
commuting choices of new town residents and workers. In both studies, he suggests
policies that eliminate barriers to residential mobility and housing production more
strongly influence commuting than initiatives aimed at jobs-housing balance and self-
sufficiency, which is quite similar to Levine’s argument (1998). More recently, several
studies all find moderate support for a relationship between jobs-housing balance and
commuting. Peng’s study (1997) finds a non-linear relationship between the jobs-housing
ratio and vehicle miles travelled and trip length in the Portland metropolitan area, and he
indicates only when the jobs-housing ratio is less than 1.2 or larger than 2.8, commuting
patterns in terms of vehicle miles travelled vary noticeably as the jobs-housing ratio
changes. Similarly, Levinson (1998) finds that jobs-rich residential areas and housing-
rich workplaces correspond to shorter commutes in Washington, DC. And, Sultana’s
(2002) analysis of the Atlanta region highlights that the imbalance between the locations
of jobs and housing is still the most important determinant for longer commuting, and she
advocates higher quality housing growth close to the jobs-rich communities.
Diverse views and conclusions show that the relationship between jobs-housing
balance and commuting patterns is far from clear. Shen (2000) suggests that the
controversy over jobs-housing balance is in part rooted deeply in philosophical grounds,

41
and partly is also attributable to shortcomings in measurement and analysis. He argues
that socioeconomic characteristics of commuters also have significant impacts on
commute duration, while previous studies often excluded them in analysis, which tends to
produce misleading results. Another body of literature on urban commuting explains the
commuting pattern of workers by their characteristics. One rich body of such research
revolves around the spatial mismatch hypothesis, which argues that central-city blacks
have less access to jobs and consequently commute more due to the persistent
employment decentralization and residential segregation (Kain, 1968; Wang, 2001).
Some other studies point to the shorter length of commutes for female commuters (White,
1986; Turner and Niemeier, 1997). And, there are other studies that explain commuting
by worker’s characteristics beyond race and gender. For instance, Punpuing (1993)
examines the relationship between demographic, socioeconomic and social environment
factors and commuting patterns in Bangkok, Thailand, and he finds that age and home-
ownership status are related to commuting time, and commuting distance relates to
occupation and home-ownership status. Wang’s study (2001) also demonstrates the
promise of explaining commuting times by worker’s characteristics such as race, gender,
household types, home-ownership status, educational attainment and wage rate. However,
these studies usually do not include urban structure as explanatory variables in analysis,
and Wang (2001) argues that a model including both kinds of independent variables
requires different interpretations. As correctly pointed out by Shen (2000), commuting
patterns have both social and spatial dimensions. In a study that examines commute times

42
in the Boston metropolitan area in 1990, he finds strong evidence that urban spatial
structure, which is determined jointly by transportation and land use, has a statistically
significant and important effect on commuting when characteristics of workers are
controlled for. By taking both social and spatial dimensions of commuting into account,
he concludes that mobility enhancement and land use planning are worth exploring as
possible approaches to transportation problems, and it is necessary to integrate physical
planning and social service planning to address transportation problems.

2.3 SPATIAL RESTRUCTURING OF CHINESE CITIES IN THE POST-
REFORM ERA
Spatial restructuring of Chinese cities in the post-reform era has been of great
interest to many urban scholars. However, as Ma and Wu (2005a) argued, relative to the
cities of the advanced capitalist countries, much less has been known about how the
internal socio-spatial patterns in the former socialist cities have changed during the post-
socialist era. Scholars engaged in China’s urban development have attempted to
conceptualize and theorize the spatial restructuring and transformation of Chinese cities
since the economic reforms, and such efforts have become several collections of studies
that provided insightful understanding of the changing spatiality and the underlying
processes and mechanisms (e.g. a volume edited by Logan, 2002; two theme issues of
Environment and Planning A edited by Lin and Wei, 2002 and Wei and Lin, 2002; a
special issue of Progress in Planning edited by Lin, 2004a; and a theme issue of Urban

43
Geography edited by Wu, 2005). One recent collection, a volume edited by Ma and Wu
(2005b), brought together the latest scholarship and provided the most comprehensive
treatment on the topic.
As correctly pointed out by Ma and Wu (2005a), China’s post-reform urban
transformations are a consequence of the interplay among the diverse forces emanating
from the global, national and local scales that converge in the city. They argued that the
global changes in the mode of regulation and the regime of accumulation that affected the
production of space, urban consumption, and the circulation of capital, people and
technology in Chinese cities have had great impacts on China’s urban transformations.
Moreover, China’s post-reform urban restructuring also involves several major domestic
institutional shifts, such as the shift to an economy of market, the shift to fiscal
decentralization and greater local economic autonomy, and the shift to paid land use
rights and commodified housing production, etc. The consequence of such changes is a
new urban spatiality for China that is visibly different from that of the pre-reform era.
Some other scholars also provided the similar views. For instance, Lin (2002) argued that
China’s urban development has been the direct outcome of national political strategizing,
state articulation and reconfiguration, and shifts in global capital accumulation, so the
dynamics of the post-reform urban restructuring in China can not be understood solely
based on the growth deterministic interpretation of urban change, while should be
examined within the broader context of globalization and national development strategies
of China.

44
Through a review of the literature on China’s urban transformation in the second
half of the 20
th
century, Ma (2002) emphasized the urgent need for the context-based
country-specific theorization of urban change in post-reform China. Such efforts to
conceptualize and theorize China’s urban restructuring from the perspectives of political
economy have recently been made by urban geographers, e.g. Wu (1997). Wu (1997)
developed a framework for understanding urban restructuring in post-reform China, and
applied several ideas of political economy analysis, such as capital switching, the
structure of building provision, rent gap and property right. He presented a descriptive
model that revealed how the built environment transformed with the changing logic of
the production of the built environment in Chinese cities since the reforms, namely the
political economy of decentralization, reorganizing the production of the built
environment through adopting the new ways of urban development, and the penetration
of the global capital. Whereas, Ma (2002) argued that the applications of the existing
Western political economy approaches are not enough for theorizing China’s urban
restructuring, and fresh perspectives based on China’s own experiences are needed,
which he believed should center on the strong party-state of China at the central and local
levels and the increasingly close but complex reciprocal economic relations between
businesses and government units. Similarly, in a recent study, Lin (2007a) also examined
the changing spatiality of Chinese urbanism with special reference to changes in state-
society relations. He suggested that the reformation of state-society relations in the post-
reform era has facilitated the growth of modern urbanism in China, which is

45
characterized by the dramatic urban expansion, high inner-city density and growing urban
diversity, heterogeneity and inequality.
Besides the theoretical reasoning, empirical studies have also been conducted
widely covering diverse subject-matters to establish the reality of China’s urban
restructuring. The consequences of the restructuring of post-reform Chinese cities are
manifested in the emergence of the new urban spaces, such as new business districts,
gentrified gated communities, dilapidated migrant enclaves, large peripheral residential
areas, development zones and high-tech industrial parks, etc. Several scholars have
underscored that the dramatic changes in China’s contemporary urban landscape are
mainly attributed to the changes in urban land use resulted from the establishment of the
land and housing market. Wu (2001) has argued that globalization and marketization in
the arena of urban land and housing are the fundamental factors determining the post-
reform urban structure in China, which have led to the spatial patterns of post-reform
Chinese cities greatly analogous to those of Western cities. Through a case study of the
land use changes in Guangzhou, Wu and Yeh (1999) uncovered the transformation of
urban spatial structure in Guangzhou since the reforms, characterized by the rapid
decentralization through leap-frog developments in peripheral areas and the re-emergence
of business and service areas in the city center, and they explained the changing spatial
pattern as a result of the reforms that introduced land values and markets to the urban
areas, which fundamentally changed the urban development process and organization in
China. In a more comprehensive study that analyzed changes in China’s non-agricultural

46
land in relation to the growth and structural changes of Chinese cities, Lin (2007b) found
that rapid urban sprawl of large cities driven by the expansion of ring-roads and the
construction of development zones has contributed to the conversion of farmland to urban
areas, and meanwhile rural industrialization and a housing boom have led to a dispersed
pattern of urban land development all over the country, so he argued that since the mid-
1990s, China’s urban spaces have been reproduced through a land-centered development
process.
Some scholars have also emphasized the relationship between globalization,
economic restructuring and spatial transformation of Chinese cities. This strand of
literature mainly focuses on the de-industrialization and tertiarization of China’s urban
economy, and the rapid growth of the tertiary sector in urban China. Lin (2004b) argued
that economic tertiarization has been one of the key forces driving the dramatic
expansion and transformation of large Chinese cities in the post-reform times. Through a
case study of Guangzhou, Lin (2004b), as well as Yang (2004), found that the tertiary
sector in urban China has grown substantially and become a major source of employment
and a powerful engine for reorganizing urban land use, further transforming the
contemporary urban landscape of China. This relationship between the growth of services
and urban development has also been evident and documented in the studies on other
large Chinese cities, such as Beijing (Wang and Jones, 2002), Shanghai (Wang and
Zhang, 2005), and Xi’an (Yin et. al., 2005).

47
Another strand of the literature on China’s urban transformation has focused on
the impacts of the rural-urban migration on the urban restructuring. The large scale rural-
urban migration and the emergence of migrant enclaves as a result of massive influx of
rural migrants to large and medium-sized Chinese cities since the reforms have been
widely documented and studied in the literature (Liu and Liang, 1997; Ma and Xiang,
1998; Zhang, 2001; Fan and Taubmann, 2002). These studies have revealed that migrant
enclaves in Chinese cities are mostly self-organized and self-managed and created
through kinship and native place ties as well as through carefully cultivated personal
connections with local authorities, and therefore they can be considered as non-state and
quasi-privatized urban spaces produced in the interaction between migrants and
government, which are culturally, economically, and politically different from the larger
urban space in which they are embedded (Ma, 2004; Zhang, 2005). They also emphasized
that such settlements are not like the ethnic ghettos in American cities, and they are not
spaces of urban poor, although the enclaves are generally crowded and disordered. Ma
and Xiang (1998) showed that employment and income patterns of migrant groups
differed markedly, and they suggested that they should not be seen as a single urban
lower social class. In a recent study, Zhang (2005) examined the role of migrant enclaves
in China’s urbanization more closely, and argued that migrant enclaves play a positive
role to promote urbanization in today’s China by housing massive rural migrants and
assimilating them into cities without using government resources. Through a detailed
study of the settlement pattern of migrant households in Shanghai, Wu (2005, 2008)

48
found that migrants have asserted their influence on urban spatial structure, and the
migrant distribution in Shanghai displayed a strong centralized tendency until the late
1990s when the inner suburbs became the main locations where new migrants
concentrated, and he concluded that the migrant residential redistribution coincided with
the overall trend of decentralization of population and industries in Shanghai, which
implies that migrant distributions contributed to the suburbanization.
Aside from migrant enclaves, post-reform Chinese cities have also witnessed the
emergence of other new residential spaces, such as illegal housing constructed in the old
neighborhoods, gentrified gated communities built for the rich and large scale peripheral
residential communities with varying housing quality, which has led to increasing socio-
spatial differentiation and residential inequality in contemporary Chinese cities. In a case
study of Beijing, Huang (2005) found significant housing inequality across education and
occupation and unprecedented residential segregation at the neighborhood level in
Beijing, which she believed can be attributed to persistent socialist institutions such as the
household registration system and work-units, newly introduced market forces, and their
interactions. Similarly, through examining the demographic composition of selected
subdistricts in Shanghai, Wu and Li (2005) found that social spaces have been
differentiated in Shanghai by education, occupation and household registration status.
Based on a case study of three neighborhoods in Shanghai, Li and Wu (2006) argued that
residential differentiation of Shanghai is constituted by sorting stratified residents
towards differentiated neighborhoods, and they found that the neighborhoods in the

49
central areas are becoming gentrified, with the remainder turning into deteriorating
workers’ villages, and the suburbs are becoming increasingly heterogeneous. Ma (2004)
emphasized that unlike in Western cities where housing of different quality tends to be
located in different spatial sectors, housing areas of different quality in Chinese cities
tend to be more mixed in spatial distribution, so it is usual to see the co-existence of high-
class communities and dilapidated low-end housing in the same general area in Chinese
cities, which is a unique spatial feature of post-reform urban China.
One major strand of the literature on the spatial restructuring of post-reform
Chinese cities that is the most relevant to our study focuses on the suburbanization and
polycentrification of large Chinese cities. Zhou and Ma (2000) first provided the most
thorough study on suburbanization of urban China, and they demonstrated that population
has relocated from the urban core to the suburbs in several large cities of China since the
1980s. Wang and Zhou (1999) modeled population densities in Beijing from 1982 to
1990, and they found the density gradient became flatter and the city-center density
decreased over time, which signaled the initiation of suburbanization in Beijing in the
1980s. The driving forces for the decentralization of large Chinese cities since the
reforms, as discussed by Zhou and his colleagues, include transportation improvement,
continual urban growth, the marketization of urban land and housing, and the renovation
of the central city (Zhou, 1997; Wang and Zhou, 1999; Zhou and Ma, 2000). Generally,
the passive side of suburbanization has been emphasized in these studies, and
suburbanization in Chinese cities is considered to be more led by government than a

50
spontaneous process driven by individual preferences like in Western cities. In a more
recent up-to-date research on suburbanization in China, Feng et. al. (2008) suggested that
the driving forces of suburbanization in large Chinese cities have changed since the
1990s and the new round of suburbanization has been more driven by market forces, and
they argued that the differences in the operation of the processes underlying
suburbanization in China and the West have begun to fade off, and the existing
differences are more related to the different stages of suburbanization than being caused
by the dichotomy of market and planned economies. However, the impacts of
government and institutions on China’s suburbanization have still been emphasized in
other studies. Through analyzing the pattern of residential moves in Guangzhou, Li and
Siu (2001) found that work units and the municipal housing bureau are the primary
driving forces behind suburbanization in China today rather than the market per se. Deng
and Huang (2004) argued that urban sprawl in large Chinese cities, characterized by the
inefficient urban expansion and suburban growth, is mainly attributed to China’s uneven
land reform between the city and the countryside, and the rapid suburban growth in an
inefficient way can be viewed as the consequence of political manipulation of land
development on the urban fringe.
As noted by Ma (2004), empirical studies on China’s suburbanization so far
mainly focused on the process and the underlying driving forces of suburbanization,
while relatively little has been known about how the internal spatial structure of Chinese
cities has been restructured through the process. A few studies do have shown the

51
changing urban form with the suburbanization. For instance, Feng and Zhou (2005) found
that suburbanization of population and industries gradually reconfigured the spatial form
of the Hangzhou urban area from a hand-like to a fan-like shape. And, Wu (1998a) also
revealed that with the suburbanization the pre-reform compact urban form of Chinese
cities has been replaced by a more dispersed polycentric spatial morphology. However,
systematic studies on the changing nature of urban spatial structure with the
decentralization of population and industries in large Chinese cities are still quite needed.

52
CHAPTER 3.
CHARACTERIZATION AND EVOLUTION OF POPULATION DISTRIBUTION
IN METROPOLITAN BEIJING

3.1 INTRODUCTION
The pattern of population distribution is a crucial economic and social feature of
an urban area (McDonald, 1989), and a basic concern of studies on urban spatial structure.
It is important for planners and policy-makers to understand urban population
distributions because of their explicit implications for labor supply, housing needs and
requirements of publicly provided facilities (Champion, 1976). The empirical analysis on
the evolving distribution of urban population is also the starting point for investigations
into the changing nature and structure of cities, and the forces that influence the spatial
organization of human activities in urban areas (Champion, 1976).
Beijing experienced rapid population growth in recent decades. During the period
from 1982 to 2000, the population of the Beijing metropolitan area grew by 50.2% from
7.99 million to 12.01 million, with an average annual growth rate of 1.87%. The
explosive growth of population has changed the spatial form of the city. The
decentralization of population from the inner city into the suburbs since the 1980s has
been identified in several previous studies (Wang and Zhou, 1999; Zhou and Ma, 2000).
Wang and Zhou (1999) modeled the population density in Beijing for the first time, and
applied population density functions to study the population distributions in the Beijing

53
urban area for 1982 and 1990. Their results found the similar patterns and development
trends of population distribution in Beijing to what has occurred in most Western cities,
and indicated the beginning of suburbanization in Beijing in the 1980s. However, more
detailed investigations into the spatial dynamics of population distribution in the Beijing
metropolitan area are still needed. Although the fact that the population has decentralized
with urban growth since the 1980s in Beijing has been well established, the form and
extent of this decentralization are not well-understood. The broad trend to
decentralization may mask significant rearrangement of the spatial distribution of
population. Local, asymmetric, trends would be highly relevant as the city has become
significantly more differentiated and segregated than before over the post-reform period,
which is most clearly reflected in residential space (Ma and Wu, 2005a).
This chapter presents an empirical analysis on the patterns of population
distribution in the Beijing metropolitan area and their evolution from 1982 to 2000.
Compared with the previous studies, this study extends the study period to the 1990s, and
employs more flexible analytical techniques in addition to the classic density gradient
function approach. It mainly addresses three questions: (1) How has the population been
distributed within the metropolitan area at different spatial scales? How have those
patterns evolved over time? In particular, to what extent has the decentralization occurred?
(2) Is there any spatial disparity shown in the process of decentralization? How does this
variation relate to the overall spatial development of the city? (3) How has the urban
spatial structure transformed with the population decentralization? In particular, has the

54
population uniformly dispersed or clustered to form subcenters while it spread into the
suburbs? The objective of this study is to better understand the changing characteristics
of population distribution within the Beijing metropolitan area, given the context of the
rapid urban growth and the economic and societal transition.

3.2 METHODOLOGY
3.2.1 Analytical Techniques
The study of urban population distribution has a long history in urban economics.
Starting with the pioneering work of Clark (1951), urban population densities have been
studied intensively by urban researchers for a wide range of urban areas in different
countries and at different time periods. Population density gradient functions have been
employed as the main instruments for describing the spatial patterns of urban population
densities owning to their simplicity and comparability among urban areas and time points
(Mills and Tan, 1980). Despite the extensive uses of urban population density gradient
functions, there is little agreement on its specific functional form (Kau and Lee, 1976).
By far the most popular functional form used for depicting urban population density
gradient patterns remains the negative exponential, not only because it performs very
well in many studies to serve as a simple but useful summary measure of population
density patterns (McDonald, 1989), but because it has its roots in the theoretical
framework by Muth (1969) and Mills (1972) and can be directly derived from the classic
monocentric urban model. But, through the years, scholars are not convinced that the

55
negative exponential is the best functional form, and several studies, e.g. Kau and Lee
(1976) and Lahiri and Numrich (1983), have shown that the negative exponential
function may not be the appropriate specification in all cases, especially for describing
the more complex population density patterns of the contemporary, relatively dispersed
urban area (McDonald, 1989). Some researchers keep experimenting with more flexible
functional forms, such as spline functions (Anderson, 1982, 1985), and looking for
empirical generalization through more powerful techniques capable of representing the
empirical patterns in population density more accurately, while still no single functional
form can be claimed to be generally superior to the negative exponential (Alperovich,
1995).
The use of the negative exponential function or its relatives assumes a
monocentric and symmetrical urban structure. Such functions do not account for the
differences of population densities in different directions or for the existence of possible
subcenters. There is strong evidence emerging that the decentralization of the large
contemporary metropolitan area has taken a more polycentric form, and subcenters have
become an increasingly important element in determining urban structure (Anas et al.,
1998; Jun and Ha, 2002). The major step in advancing the modeling of population
densities for the contemporary urban area has been the reconceptualization of the density
function from the monocentric to the polycentric form (Griffith, 1981a, 1981b). With
defined subcenters, several studies, e.g. Gordon et al. (1986) and Small and Song (1994),
have found the polycentric density function a better fit relative to the monocentric one for

56
decentralized urban areas. But, as Griffith and Wong (2007) correctly pointed out,
although the more sophisticated conceptualization has advanced the modeling of urban
population densities with more complex model specifications, the approach is still
suffered from reducing the three-dimensional density surface into a two-dimensional
function without considering local spatial structure.
The analysis of the spatial patterns of urban population requires estimating the
density surface at several points in time, in order to capture both systematic changes of
global trends and local variation, such as subcenters. The pervasiveness of polycentricity
has motivated more flexible techniques applied to modeling urban population densities.
To address the rigidity and drawbacks of the parametric approach, McMillen and
McDonald (1997) first employed a nonparametric estimation procedure, locally weighted
regression, in modeling polycentric cities, and they emphasized the flexibility of
nonparametric estimation has distinctive advantages for modeling the polycentric
structure of the contemporary decentralized urban area. Afterwards, this methodology has
been further refined by McMillen (2001) and Redfearn (2007). Nonparametric
procedures do not require the specification of a global function to fit a model to the data,
while instead they infer the regression surface from the data, smoothing the data while
maintaining as much complexity as necessary to produce unbiased estimation (McMillen
and McDonald, 1997). This flexibility makes them ideal for modeling complex processes,
such as the evolution of the spatial patterns of large urban areas. This study employs
nonparametric analysis in addition to the density function approach to characterize the

57
spatial patterns of population densities in the Beijing metropolitan area and their
evolution during the post-reform period. For the case of Beijing, the results indicate that
spatial distributions of population - while broadly monocentric - are sufficiently irregular
that simple parametric forms miss important dynamics locally.

3.2.2 Description of Data
This study uses the population data drawn from the 1982, 1990 and 2000
population censuses of Beijing. The data are at the subdistrict (jie dao), town (zhen) and
township (xiang) level. The subdistrict, town and township are the smallest
administrative divisions of China. In general, urban areas are divided into subdistricts,
while rural areas are divided into towns and townships. The subdistrict, town and
township (hereafter referred to as the subdistrict) are the finest geographical units where
the census data are available. For each subdistrict, we observe the tract’s centroid and
land area. The distance is measured as a straight line distance between the centroids of
tracts. The spatial boundaries of subdistricts are aggregated in some cases to
accommodate to the changing administrative boundaries of subdistricts over time. Table
3.1 provides the summary statistics for our data.

58
Table 3.1 The Summary Statistics for the Subdistricts
Count
1982 1990 2000 Subdistrict
Tracts
168 204 232
Mean Min. Max. Std. Deviation Total
1982 47.57 5.78 457.10 43.90 7,992.27
1990 46.28 6.03 292.39 35.85 9,440.59
Population
(in thousands)
2000 51.75 4.22 214.20 36.92 12,006.09
1982 54.26 0.98 383.50 61.66 9,116.30
1990 44.69 0.98 383.50 57.57 9,116.30
Area
(in km
2
)
2000 39.29 0.94 383.50 54.75 9,116.30
1982 7.71 0.04 60.40 13.27 -
1990 7.83 0.04 52.71 11.55 -
Density
(thousand per
km
2
) 2000 8.70 0.03 40.47 10.28 -

The Beijing metropolitan area covers more than nine thousand square kilometers
with nearly eight million people in 1982 and over twelve million in 2000. There are 168
tracts in 1982 and 232 in 2000. The increase of the number of tracts is mainly due to the
division of large subdistricts into smaller ones for the administrative purposes. The
subdistrict tracts vary in land area from less than one square kilometer to hundreds of
square kilometers. There is similarly great variation in the population of tracts, with the
mean at about 50 thousand. The average population density of tracts increases over time,
from 7.71 thousand per square kilometer in 1982 to 8.70 in 2000. While the lowest
density remains generally the same through the years, marked shifts in the location of
population can be observed at the other end of the distribution - the highest density drops
greatly by 33% from 60.40 thousand per square kilometer in 1982 to 40.47 in 2000.

59
The spatial distributions of population in the Beijing metropolitan area in 1982,
1990 and 2000 are depicted in Figure 3.1. A relatively uneven pattern is present with less
population distributed in the western and northwestern mountain areas, where the
population densities are generally under 200 per square kilometer. The population density
patterns have changed steadily in Beijing during the two decades (Table 3.2 and Figure
3.1). About one-third of population lived in the inner city of Beijing in 1982, while this
percentage declined to 17.31% in 2000. The inner city has lost a significant amount of
local residents during the two decades with the negative population growth rate of -
13.41%. The most drastic decrease in population in the inner city mainly occurred in the
1990s. The suburbs, particularly the near suburbs, have experienced rapid population
growth. The population in the near suburbs more than doubled in the two decades. In
2000, over half of population in the metropolitan area lived in the near suburbs, whereas
only less than one-third in the outer suburbs. Even though the population in the outer
suburbs has increased continuously during the decades, the proportion of population
living there continued to decline, albeit slowly.

60
Table 3.2 The Distribution of Population in the Beijing Metropolitan Area at Different
Spatial Scales and its Changes
Population (in thousands) Population Growth Rates
Periods
Spatial Scales
1982 1990 2000
1982-
1990
1990-
2000
1982-
2000
Metro Area 7,992 9,441 12,006 18.12% 27.18% 50.22%
Urban Area 5,226 6,295 8,365 20.47% 32.87% 60.07%
Inner City 2,400 2,317 2,078 -3.43% -10.34% -13.41%
Near Suburb 2,826 3,978 6,287 40.76% 58.05% 122.46%
Outer Suburb 2,766 3,145 3,641 13.69% 15.77% 31.62%
Population Proportion Changes of Proportion
Metro Area
Urban Area 65.39% 66.68% 69.67% 1.30% 2.99% 4.29%
Inner City 30.02% 24.55% 17.31% -5.48% -7.24% -12.72%
Near Suburb 35.36% 42.14% 52.37% 6.78% 10.23% 17.01%
Outer Suburb 34.61% 33.32% 30.33% -1.30% -2.99% -4.29%

Figure 3.1 illustrates the evolution of population distributions in the Beijing
metropolitan area, and shows the spatial spread of population from the inner city to the
suburbs. While the inner city remains at fairly high densities, the density of the urban
core has fallen since 1982. The near suburbs immediately adjacent to the inner city,
particularly the eastern, northeastern and northwestern parts, have accommodated more
population and become the densest areas, with densities generally over 20 thousand per
square kilometer. The observation from the density maps shows people spread out
significantly, especially in the 1990s, along the major radial roads in almost every
direction, particularly to the northwest, east, south and southwest. In 2000, most areas
within the 5
th
ring road (the continuously built-up urban areas) had densities generally
over 2,000 per square kilometer.

61
Figure 3.1 Spatial Distribution of Population in Metropolitan Beijing

3.3 CHARACTERIZING POPULATION DENSITY PATTERNS WITH
DENSITY FUNCTIONS
This section applies the density function to characterize the spatial pattern of
population densities in the Beijing metropolitan area. Three functions are used: (1) the
negative exponential (Clark, 1951)
0
() ( 0)
x
Dx De
β
β = <

62
(2) the square root negative exponential (Ajo, 1965)
0.5
0
() ( 0)
x
Dx De
γ
γ = <
and (3) the quadratic exponential (Newling, 1969)
2
0
() ( 0; 0)
ax bx
Dx De a b
+
= ><
where D(x) denotes gross population density at distance x from the city center, and D
0
is
density extrapolated to distance zero. The negative exponential is the most widely used
functional form in the literature, and is usually regarded as a reasonable first
approximation of density patterns within the metropolitan area (Parr, 1985a). For a larger
region that contains both metropolitan and surrounding non-metropolitan areas, Parr
(1985b) indicated the overall density profile may not be consistent with the negative
exponential function, and he suggested the square root negative exponential as a
simplified approximation. Considering the Beijing metropolitan area specified in this
study contains large rural areas, we would try the square root negative exponential as a
reasonable substitute of the normally-used negative exponential function. The quadratic
exponential function is proposed by Newling (1969) to capture the density crater
surrounding the city center. The existence of a crater of population density at the city
center has been observed for many large Western cities, Washington, D.C. or Paris, for
example. This situation also applies to Beijing (Figure 3.1), because the old imperial city
occupies a large amount of land at the city center, which now serves as museums, public
parks and central government headquarters that for the most part are nonresidential.

63
The density functions are estimated for both the metropolitan area and the urban
area of Beijing. The conventional method of estimation is to convert the density functions
to logarithmic forms and then use ordinary least squares (OLS). One problem with this
method of estimation concerns the sampling bias related to heteroskedasticity (Frankena,
1978). An analysis of the residuals from the OLS regression using the White test (1980)
indicates significant heteroskedasticity does exist in our data for most of the years. To fix
the problem, we employ the weighted least squares (WLS) estimation method as
suggested by Frankena (1978). Compared with the WLS estimation, the OLS procedure
overestimates the coefficients in general. Both the OLS and WLS estimation results are
reported in Table 3.3 and Figure 3.2 and 3.3.
Among the three functions, the square root negative exponential fits best for the
density patterns within the metropolitan area of Beijing. Although the quadratic
exponential fits better than the negative exponential, the sign of the coefficients in the
quadratic exponential model is not as expected, which suggests the density patterns may
not be properly represented by the quadratic exponential function. The absolute value of
the coefficients β and γ, known as the density gradient, reflects the decline in density with
increasing distance from the city center. In both the OLS and WLS regressions, the
gradient has become steeper and the value of the constant (LnD
0
) has increased over time,
indicating a trend toward the centralization of population in the metropolitan area. Figure
3.2 shows there was not much change in density patterns between 1982 and 1990, while a
notable change occurred from 1990 to 2000. The WLS estimation results of the negative

64
exponential function show the densities at the places within 40 kilometer from the city
center have risen during the 1990s, indicating the growth of population in the urban area
of Beijing during the period. (Figure 3.2)
Figure 3.2 Fitted Density Functions for the Metropolitan Area of Beijing

65
Table 3.3 The Estimation Results of the Density Functions
for Both the Metropolitan Area and the Urban Area of Beijing
Ordinary Least Squares (OLS): DepVar=Ln(Den)
Metropolitan Area
Year 1982 1990 2000
Constant 9.52311.33310.3929.56111.347 10.3889.93511.71910.516
Dis -0.088 -0.182-0.087 -0.178-0.093 -0.157
Dis^2 0.002 0.002 0.001
Dis^1/2 -0.874 -0.860 -0.887
Adj. R
2
0.7240.8210.8180.6890.775 0.7720.7230.7610.761
9.820 0.180 15.590 0.520 2.540 28.790 15.230 6.940 32.110
White Test
0.007 0.913 0.008 0.769 0.282 <0.0001 0.001 0.031 <0.0001
Urban Area
Constant 10.58412.14811.28010.60512.148 11.15710.64011.86210.669
Dis -0.180 -0.344-0.178 -0.304-0.153 -0.159
Dis^2 0.006 0.005 0.000
Dis^1/2 -1.140 -1.116 -0.916
Adj. R
2
0.7280.7500.7840.6730.676 0.7030.6460.5950.643
10.810 2.660 19.670 23.870 5.450 26.950 20.070 11.280 30.870
White Test
0.005 0.265 0.001 <0.0001 0.066 <0.0001 <0.0001 0.004 <0.0001

66
Table 3.3, Continued
Weighted Least Squares (WLS): DepVar=Ln(Den) Weight=Area^1/2
Metropolitan Area
Year 1982 1990 2000
Constant 8.55710.4769.4978.73910.672 9.6499.28911.46310.189
Dis -0.066 -0.134-0.069 -0.136-0.081 -0.150
Dis^2 0.001 0.001 0.001
Dis^1/2 -0.746 -0.765 -0.886
Adj. R
2
0.6560.7370.7160.6470.721 0.7030.7070.7580.752
1.710 7.380 14.250 3.530 9.420 18.320 9.540 15.320 35.930
White Test
0.635 0.061 0.014 0.317 0.024 0.006 0.023 0.002 <0.0001
Urban Area
Constant 10.04111.78411.12010.20611.984 11.19210.50912.07910.871
Dis -0.154 -0.348-0.161 -0.336-0.154 -0.218
Dis^2 0.006 0.006 0.002
Dis^1/2 -1.092 -1.120 -1.033
Adj. R
2
0.6810.7350.7680.6470.680 0.7100.6740.6530.681
9.880 3.380 12.290 28.150 7.270 28.180 22.670 8.690 27.590
White Test
0.020 0.337 0.056 <0.0001 0.064 <0.0001 <0.0001 0.034 0.000

67
Figure 3.3 Fitted Density Functions for the Urban Area of Beijing

For the urban area of Beijing, the quadratic exponential function provides the best
fit, but the sign of the coefficients is still not as expected. The square root negative
exponential fits better than the negative exponential for the early two years, while the
negative exponential fits better for the year 2000. The quadratic exponential model
captures the decreasing density levels at the city center, while the negative exponential
and the square root negative exponential model mostly predict an increasing value of the

68
constant. The estimation results show the gradient has flattened through the years,
indicating the spread of population from the inner city to the near suburbs, except the
WLS estimation of the negative exponential function, the results of which show a parallel
growth of population in the urban area. (Figure 3.3)
The present analysis using the parametric density functions provides little
consistent evidence for the decentralization of population in the Beijing metropolitan area.
Rather, the analysis reveals an increasing trend of concentration of population into the
urban area in metropolitan Beijing, and the spread of people into the near suburbs in the
urban area. The findings depend on the spatial scale specified in the analysis.
Furthermore, the parametric density functions applied in this section do not fit the density
profile and its change for the central part of the city very well. Because the density crater
is present, the linear density functions overestimate the density levels at the center, and
the quadratic function also fails to capture the cresting of densities around the center.
Furthermore, in most cases, the density functions are incapable of reflecting the fact of
the continuous decline in density at the central area.
Figure 3.4 depicts the growth pattern of population densities in the Beijing
metropolitan area. In both decades, the central area within a radius of 5 kilometer from
the city center (mostly contained within the 2
nd
ring road) lost population resulting in a
decrease in density. From 1982 to 1990, there was a general trend of growth in densities
throughout the whole metropolitan area, except several subdistricts at the western and
southeastern periphery. The most rapid growth of densities mainly occurred at the places

69
with distance 5~20 kilometer from the center. The outer areas experienced a relatively
slow growth. During the 1990s, the growth pattern has changed distinctly. The most
evident change is that the densities at the periphery of the metropolitan area (with
distance generally over 40 kilometer from the center) dropped greatly, compared with the
growth in the 1980s. On the other hand, the areas within the distance range 5~20
kilometer from the center have experienced more rapid growth in densities during the
decade. They are mostly the suburbs immediately adjacent to the inner city, especially in
the northeastern and southern parts. Obviously, the growth pattern of population densities
in the Beijing metropolitan area is more complicated than what the parametric density
functions can capture. It turns out the density functions not to be very useful in our study,
mainly due to the rigidity of the functional forms, which motivates the application of
more flexible analytical techniques in the following section.

70
Figure 3.4 Growth Patterns of Population Densities in Metropolitan Beijing
(a) Spatial Patterns of Growth

(b) Plot of Growth Rates against Distance

71
3.4 DECENTRALIZATION AND SUBCENTERING OF POPULATION IN
METROPOLITAN BEIJING: A NONPARAMETRIC ANALYSIS
3.4.1 Local Regression
This section applies a nonparametric estimation procedure, local regression
(Loess) or locally weighted regression (Lowess), to investigate further the pattern of
population distribution and its evolution in the Beijing metropolitan area. The problem
with the parametric analysis using the density function mainly arises from the rigidity of
the preassumed functional form. The advantage of nonparametric procedures lies in that
it is not required to specify a global functional form to fit the data, this can reduce
misspecification bias to a large extent and allow greater flexibility than traditional
modeling methods.
Local regression is originally proposed by Cleveland (1979) and further
developed by Cleveland et al. (1988), and Cleveland and Grosse (1991). The procedure
approximates a complex regression surface with a series of local approximations
(McMillen and McDonald, 1997). Such a local approximation is obtained by fitting a
low-degree polynomial to a subset of the data within a chosen neighborhood around the
point whose response is being estimated, using weighted least squares, with more weight
given to nearby observations. The size of the neighborhood determines the fraction of the
data included in the local fitting, which is also called the smoothing parameter, and
controls the smoothness of the estimated surface. The smoothing parameter can be
selected using a variety of methods, most of which choose the parameter value to

72
minimize a specific criterion. In this study, we use the bias-corrected AIC criterion (AIC
C
)
proposed by Hurvich et al. (1998), formulated as
2
2( ( ) 1)
ˆ log( ) 1
() 2
C
Trace L
AIC
nTraceL
σ
+
=++
− −

where
2
ˆ σ is the error mean square, n is the number of observations, and L is the
smoothing matrix that satisfies ˆ y Ly = , where y is the vector of observed values and ˆ y is
the corresponding vector of predicted values of the dependent variable. Within the
specified neighborhood, observations are given weights that decline with distance. The
weight function used in this study is a tricube function given by
3
3
max
32
1( )
5
i
i
d
w
d
⎛⎞
=−
⎜⎟
⎝⎠

where w
i
denotes the weight given to observation i, d
i
represents the distance between
observation i and the point of the local fitting, and d
max
is the largest distance from the
point to any observation within the local neighborhood. A maintained assumption is that
the regression surface at any point can be approximated by a simple linear function. And
the flexibility is introduced by estimating a weighted linear regression point by point.
Local regression has its particular advantages in modeling urban spatial structure,
as it is very flexible and capable of depicting the local relationship between the response
and the predictor variable. Nonparametric analysis can help develop a better and more
accurate description of urban density surfaces, while a disadvantage is that it does not
produce a function that is easily represented by a mathematical formula. Therefore, the

73
analysis is not based on specific parameters, like the density gradient in a density
function, but requires visualizing the regression surface by drawing the smoothed curve
or surface on a scatter diagram. Because of this other metrics will be needed beyond
estimated gradient values in order to characterize changes in the spatial distribution of
population. These are discussed below.

3.4.2 Characterizing Density Patterns Using Local Regression
The starting point is still a logarithmic density function
0
() LnD x LnD x β = +
which serves as a base specification for the local regression. Compared with linear or
nonlinear least squares regression, local regression adapts locally to curvature in the
regression surface that is not accounted for adequately by the base equation, and
estimates the regression surface more accurately (McMillen and McDonald, 1997).
Figure 3.5 shows the loess fit. The smoothing parameter is selected to minimize the bias-
corrected AIC criterion. The parameter value is 0.23 for 1982 and 1990, and 0.39 for
2000, which means there are 38, 47 and 91 points included in the local neighborhood
respectively for the three years.

74
Figure 3.5 Loess Fit of the Logarithmic Density Function

The loess fit represents the growth pattern of population densities in the Beijing
metropolitan area much better than the OLS estimates. The smooth curves capture the
density crater at the city center, and indicate the decline in density over time in the central
area near the center. During the 1980s, all densities at the places beyond 5 kilometer from
the center have risen slightly, indicating the spread of population out of the inner city,
while during the 1990s, substantial growth of population has only occurred in the areas
beyond 5 kilometer but within 40 kilometer from the center, and densities at the places
beyond 40 kilometer from the center have decreased in general, which indicates the
concentration of population into the near suburbs from both the inner city and the outer
suburbs. The growth pattern that the loess fit has revealed is exactly the same as that

75
depicted in Figure 3.4, which suggests local regression does provide a more accurate
description of density patterns in the metropolitan area.
The loess fit indicates that the population has decentralized during the decades, as
there is clear evidence that people have moved away from the inner city, but the spatial
extent of decentralization is limited in general. Instead of spreading throughout the whole
metropolitan area, the population was decentralized mostly within the urban area, from
the inner city to the near suburbs. Meanwhile, the rapid growth of population in the near
suburbs has significantly expanded the spatial extent of the central city, making a larger
agglomeration of population within the urban area. From this point of view, the tendency
toward decentralization at the level of the metropolitan area is questionable, as people
have become more concentrated within the urban area, instead of further dispersing to the
outer suburbs.
Although local regression improves the explanatory power of the base density
function, the approach is still constrained by the framework of reducing the density
surface to a two-dimensional function. The analysis of spatial patterns requires three-
dimensional surface modeling. Local regression is also applicable to surface fitting. A
population density surface can be expressed as
(, ) Dfxy =
where x is latitude, y is longitude, D is the density at (x, y), and f is a regression function.
The density surface is formed by fitted values at each point on a grid overlaying the
sample area. To reduce the computational complexity, we get the surface using

76
interpolated fitting, which means local regression is only performed at a representative
sample of points in the predictor space and the regression surface is obtained by blending
these local polynomials. The smoothing parameter is selected using the same procedure
as above, the value of which is around 0.10 for all the three years. Figure 3.6 plots the
estimated density surfaces, which clearly show the spread of population over the urban
area. To facilitate our analysis, we also create a series of contour lines that join points of
equal density to depict increasing or decreasing trends of density over space (Figure 3.7),
as it is usually difficult to characterize spatial patterns by directly observing the three-
dimensional plots.

77
Figure 3.6 Loess Surfaces of Population Density in Metropolitan Beijing

78
Figure 3.7 Contour Maps of Loess Density Surfaces

The arrangement of the contour lines on the map and its changes over time give a
direct indication of the evolution of density patterns in the Beijing metropolitan area. The
density peak remains in the inner city, but has flattened over time. In 1982, two density
peaks appear in both the eastern and southern parts of the inner city, while the peak in the
east has dropped and is no longer visible in 1990 and 2000, with the only peak present in

79
the south. As Wang and Zhou (1999) indicated, the generally high density in the south of
the inner city has its historical root. The 30,000 isoline covering the central area near the
city center shrinks greatly from 1982 to 1990, and further disappears in 2000, with the
density level of the central area declining from 30,000 to 24,000 per square kilometer.
Another notable characteristic that the contour maps show is the continuous expansion of
the areas with moderate densities (2,000~20,000 per square kilometer). The 20,000
isoline, roughly overlapping the boundary of the inner city in 1982, has extended
outwards through the years, especially towards the north, and covers most of the areas
within the 3
rd
ring road in 2000. The 10,000 isoline has expanded more extensively in
bands of concentric ring roads, extending in almost every direction and beyond the 4
th

ring road in 2000. The expansion of the 2,000 isoline shows a more interesting picture of
the spatial dispersion of population. The 2,000 isoline covers most of the urban area in
1982 and 1990, and there is not much change in its spatial extent between the two years.
The notable change during the decade is the extension of the isoline along the east-west
direction. However, the 2,000 isoline has expanded significantly during the 1990s, and
extends far beyond the boundary of the urban area in 2000, particularly toward the
northern, northeastern, eastern, and southwestern outer suburbs. This change is in
keeping with the growth pattern shown in Figure 3.4.

80
Figure 3.8 The Decentralization of Population and the Spatial Development Scheme of
the Beijing Metropolitan Area

The spread of the 2,000 isoline, especially from 1990 to 2000, shows a non-
concentric pattern, indicating population growth disparities in different directions. The
isoline expands more extensively to the northeast and southwest and moderately to the
north and east between 1990 and 2000, while the extension to the west and southeast is
quite limited. The dispersion of population towards the west and northwest is mainly
constrained by the topology, as hills and mountains dominate in the western and
northwestern metropolitan area. The significant spread of population in the 1990s, shown
by the expansion of the 2,000 isoline, mainly benefited from the improved road network
linking the central city to the outside suburbs. The construction of the ring and radial road

81
system of Beijing began in the 1980s. Although many road construction projects started
before 1990, they were all completed in the 1990s. By the year 2000, the road system of
the Beijing metropolitan area has been formed in a layout of combining three (the 2
nd
, 3
rd
,
and 4
th
) ring roads with several radial highways and expressways linked to the gates into
and out of the central city. Figure 3.8 shows the road system, with four major radial axes
along the northwestern, northeastern, southeastern and southwestern directions, and one
major through axis along the east-west direction. The radial axes are composed of the
national highways and the paralleled expressways, with most of the expressways built-up
after 1990. The expansion of the ring roads, the buildup of the expressways, and the
increasing mobility of urban population in the 1990s have speeded up the decentralization
process. The expansion pattern of the isolines shown in Figure 3.7 and 3.8 is clearly
associated with the development of the road network. The construction of the successive
ring roads has accelerated the concentric spread of population into the near suburbs, and
the buildup of the expressways has contributed to the dispersion of population to the
outer suburbs. From Figure 3.8, we can clearly find significant extensions of the 2,000
isoline along the radial axes evident in almost every direction except the southeast.
It is interesting to examine why people did not spread so much toward the
southeast. To investigate this, we refer to the development scheme of the city. Although
the improved transport has a great impact on the dispersion of population, the
development of the city, that is the growth pattern of population, is overall controlled by
the city plan. Under the system of public land ownership in China, urban land is owned

82
by the state and allocated to different uses according to the plan by the planning authority
and municipal government. Market forces had not much influence in shaping the urban
pattern before the establishment of the land and housing market in the 1980s. Planning
always plays a key role in the development of the city, even in more recent years with
increasing impacts of market forces on urban development.
In the first city plan drafted in 1957, the Preliminary Master Plan for Construction
of Beijing, the scattering layout principle was proposed, and it has been emphasized and
maintained as a guideline in the later plans since then, through to the latest one.
According to this principle, several satellite towns have been planned to be developed in
the outer suburbs since the 1980s, aimed at attracting population and industries, to avoid
the undesirable sprawl of the central city. Figure 3.8 shows these satellite towns
designated in both the 1982 and the 1990 master plan. Obviously, the expansion of the
2,000 isoline is closely associated with this development scheme. The less expansion to
the southeast is because no satellite town that is far enough from the central urban area
was planed in this direction.
It is clear that the growth of population in the Beijing metropolitan area may have
different patterns in different directions. Therefore, it is helpful to see the changes of
density profiles along major transport axes in different directions over time. Figure 3.9
plots the density profiles along the east-west axis for the three years. It shows the decline
of the density peak and the spread of population along the axis, indicating the trend of
decentralization of population. Figure 3.10 plots the density profiles along the radial axes.

83
Obviously, the density profiles have flattened over time in all four directions, which
suggests people generally dispersed along transport axes. The density profiles move
outward significantly from 1990 to 2000, especially those along the northwestern and
southwestern axes, indicating significant decentralization of population along the axes
during the 1990s. Meanwhile, with the flattening of the density profiles, two humps show
up in the density profiles along the northeastern and southwestern axes respectively at
about 35 kilometer from the city center in 2000, which implies subcenters may emerge in
the outer areas.
Figure 3.9 Density Profiles along the East-West Axis
year 1982 1990 2000
Pred. Density (per square kilometer)
0
10000
20000
30000
40000
Distance (meter)
-30000 -20000 -10000 0 10000 20000 30000 40000

84
Figure 3.10 Density Profiles along the Radial Transport Axes

85
3.4.3 Subcentering of Population and Urban Structure
Empirical studies on the evolution of metropolitan areas in western countries,
especially in the United States, have indicated the evolving of large decentralized
metropolitan areas towards polycentric urban forms with subcenters emerging in the
suburbs and making their marks on urban spatial structure (Anas et al., 1998). But less
evidence has been presented for metropolitan areas in developing countries. In this
section, we investigate the structural change of the Beijing metropolitan area from 1982
to 2000 with the decentralization of population.
The subcentering of population in the Beijing metropolitan area has not been
examined in depth in previous studies. Although Wang and Zhou (1999) acknowledged
the emergence of subcenters in the suburbs of Beijing in the 1980s, they failed to find
significant subcenters using a polycentric density function. The failure may largely be
due to identification method they applied. The methodology for identifying the
subcenters of polycentric metropolitan areas has been advanced during the past decade,
which has become more flexible to adequately characterize the complex structure of
urban densities. The active set of methods used in the literature generally fall into three
broad categories: clustering methods, parametric methods, and nonparametric methods
(Redfearn, 2007). Clustering methods require ample priori knowledge of the study area
and depend on the subjective determinations. Parametric methods are generally not
flexible enough, with restrictive assumptions of urban structure. Therefore,
nonparametric methods have been applied more widely in recent studies, because of their

86
flexibility, objectivity and easy replication for a variety of cities. In this study, we apply
the nonparametric procedure proposed by McMillen (2001) to identify population
subcenters in the Beijing metropolitan area.
McMillen’s (2001) procedure involves two steps. The first step is to identify
candidate subcenters as significant positive residuals in a smoothed loess density surface.
The loess surface serves as a benchmark, and subcenters have densities that greatly
exceed the loess smooth. So candidate subcenters consist of tracts with residuals that are
significantly greater than 0 at the 5% significance level. To avoid taking nearby tracts as
different potential subcenters, we only choose those whose predicted densities are highest
in a cluster of nearby tracts with significant residuals as candidate subcenters, where
“nearby” is defined as within a radius of 4 kilometer. The second step is to apply a
semiparametric procedure to assess the significance of the candidate subcenters. Only
those that have significant impacts on the overall distribution of population can be
identified as effective subcenters. The semiparametric regression is given by
1
12
1
() ( )
n
ii jij jij
j
Dgx x x δδ
−
=
=+ +
∑

where D
i
denotes the density at tract i, x
i
denotes the distance from tract i to the city
center, x
ij
denotes the distance between tract i and candidate subcenter j, and n is the
number of candidate subcenters. x
i
enters the equation nonparametrically, and McMillen
(2001) suggested various alternatives can be used to estimate g(x
i
). We apply cubic
splines to approximate g(x
i
) as used by Anderson (1982, 1985), written as

87
1
23 3
00 1 0 1
1
() ()() () ( )( )
n
ii i i kkikk
k
gx a bxx cx x d xx d d xx Y
−
+
=
≈+ − + − + − + − −
∑

where
1
( , 1, 2, ..., 1)
kk k
xx x k n
+
<= − are the knots dividing the distance interval from
the city center to the metropolitan area boundary into n segments, and Y
k
is a dummy
variable such that
1 if
0 if
kik
kik
Yxx
Yxx
=≥ ⎧
⎨
= <
⎩

The number of knots is not supposed to be large. Though increasing the number of knots
gives the spline more freedom to bend, it also increases the number of parameters to be
estimated. Therefore, we choose three knots at distance 20, 40 and 60 kilometer from the
city center respectively. To assess the significance of the candidate subcenters, the
hypothesis tests on the coefficients of δ
1j
and δ
2j
are conducted. One problem with this
approach is regarding the severe multicollinearity produced by multiple distance
variables entered into the regression. McMillen (2001) suggested a reverse stepwise
regression procedure to choose the number of subcenter distance variables. The
procedure starts with the equation with all distance variables entered. At each iteration
step, the subcenter distance variable with the lowest t value is eliminated until all
subcenter variables in the regression are significant at the 5% significance level. The
intercept and cubic splines are forced to remain at each stage. The final list of subcenters
includes those whose coefficients have expected signs on either
ij
x or
1
ij
x
−
(or both) at
the end of the stepwise regression.

88
Table 3.4 Final List of Population Subcenters and Estimation Results
1982 1990 2000 Center
ID
Distance
Variable Estimate P-value Estimate P-value Estimate P-value
x
ij
-0.030 0.002
1
x
-1
ij
3.94E-080.009 3.26E-08 0.045
x
ij

2
x
-1
ij
6.60E-08 0.002 1.03E-07 <.0001 6.93E-08 0.000
x
ij
-0.031 0.003
3
x
-1
ij

x
ij
-0.064 0.001
4
x
-1
ij
1.02E-070.005 1.52E-07 <.0001
x
ij
-0.247 <.0001
5
x
-1
ij

x
ij
-0.617 <.0001
6
x
-1
ij
3.69E-08 0.031
x
ij
-0.029 0.000
7
x
-1
ij
8.41E-080.005 1.09E-07 0.001
x
ij

8
x
-1
ij
1.13E-070.032

x
ij
-0.025 0.015
9
x
-1
ij
1.30E-07 <.0001
x
ij
-0.060 0.010
10
x
-1
ij

x
ij
-0.168 <.0001
11
x
-1
ij
3.26E-08 0.003
x
ij
-0.280 <.0001
12
x
-1
ij

x
ij

13
x
-1
ij
1.14E-07 <.0001
Adjusted R
2
0.856 0.889 0.884
Note: Subcenter tract name - 1. Dong gao di (Feng tai) 2. Jin ding (Shi jing shan) 3.
Zhong guan cun (Hai dian) 4. Guan zhuang (Chao yang) 5. Zi zhu yuan (Hai dian) 6. Ba
jiao (Shi jing shan) 7. Ying feng (Fang shan) 8. Yong ding lu (Hai dian) 9. Sheng li (Shun
yi) 10. Dong sheng (Hai dian) 11. Xing cheng (Fang shan) 12. Bei yuan (Tong zhou) 13.
Cheng bei (Chang ping).

89
Table 3.4 reports the final list of subcenters identified using the above procedure
in the Beijing metropolitan area in 1982, 1990 and 2000. Our findings support the
statement of Wang and Zhou (1999) that subcenters have started to emerge in the suburbs
since the 1980s. We find an increasing number of subcenters emerging through the years,
with 3 subcenters in 1982, 7 subcenters in 1990 and 10 subcenters in 2000. The adjusted
R
2
of the semiparametric regression with multiple centers is generally higher than that of
the simple monocentric regression, especially in the later years, which suggests a better
fit of the polycentric model for the metropolitan area in more recent times.
The coefficients of the subcenter variables are highly significant. The distance
variables enter the regression in both level and inverse forms. Levels are preferable when
subcenters have effects that are spread over large areas, while the inverse form is suitable
to modeling subcenters that have more local effects. Figure 3.11 depicts the location of
the subcenters. It is interesting to see the subcenters that are relatively far from the central
agglomeration of population generally have effects spread over large areas with
significant level distance variables. The newly emerged subcenters in each year are
located relatively far from the central agglomeration, so they usually have effects spread
over large areas, while the existed subcenters normally have more local effects in the
later years with only significant inverse distance variables, as they have become more
adjacent to the central agglomeration with its continuous expansion (Figure 3.12), and
their effects on population distribution have been shadowed by the central agglomeration.

90
Figure 3.11 The Location of Subcenters and the Development Scheme of the Beijing
Metropolitan Area

Figure 3.12 Expansion of the Central Agglomeration of Population

91
Figure 3.11 and 3.12 show the evolvement of the subcenters through the years. In
1982 and 1990, most of the subcenters were located in the near suburbs within the urban
area. Though subcenters started to emerge in the outer suburbs in 1990, most of the outer
suburban centers formed in 2000. In 1982, three subcenters formed in the western,
northwestern and southern inner suburbs, with the central agglomeration mainly
contained within the 2
nd
ring road. From 1982 to 1990, the central agglomeration
expanded not so much, only slightly to the northeast. The newly emerged subcenters in
1990 were mostly in the western, northwestern, and eastern inner suburbs and along the
east-west axis. From 1990 to 2000, the central agglomeration expanded more
significantly, especially to the north, experiencing the process of conurbation, and the
subcenters in the west and northwest near the central area have been combined within the
central agglomeration and their effects on population distribution were not significant any
more in 2000. The newly emerged subcenters in 2000 were all located in the outer
suburbs, with subcenters distributed throughout the metropolitan area. Generally
speaking, the existence of the subcenters is persistent over time because of the historical
path-dependence of the city’s development. However, considering the continuous growth
of the central agglomeration, nearby subcenters may form first and then be incorporated
into it later. From this point of view, the central agglomeration may not be monocentric
by nature, and it may also have a polycentric structure. Meanwhile, the influence of the
road system is clear, with all the subcenters organized around the ring roads or along the
transport axes. Several subcenters in the west along the east-west axis are close together

92
and can be considered as a group that forms a corridor extending from the central area all
the way to the western inner suburbs.
From Figure 3.11, we can clearly see the development of the subcenters is highly
associated with the development scheme of the city. As discussed before, planning is a
major tool used by the municipal government to control and guide the urban development
in China. To strictly control the scale of the central city, the planning authority has made
the plan to maintain a scattering layout of Beijing back in the 1950s. In the 1982 master
plan, following this scattering layout principle, in addition to the satellite towns planed in
the outer suburbs, ten scattered residential groups were planned as inner suburban
development areas at the edge of the urban area, to avoid the over concentration of
population in the central city. These groups are mostly residential and have been
developed through large investment in housing during the 1980s and the 1990s.
Obviously, the subcenters within the urban area are mostly located in or around these
edge development areas, while the outer suburban centers are all located in the satellite
towns.
The subcenters have obvious impacts on the patterns of population distribution in
the Beijing metropolitan area, as shown in Figure 3.11. The changes of the 2,000 isoline
are clearly related with the emergence of the subcenters. From 1982 to 1990, the isoline
extends mostly in the east-west direction, which is influenced by the emergence of the
subcenter 4, 6 and 8 in the western and eastern inner suburbs along the east-west axis.
From 1990 to 2000, the isoline expands significantly toward the northeast, east and

93
southwest, which can be explained by the newly emerged subcenters in the outer suburbs.
This suggests the importance of the subcenters in the evolution of population density
patterns in the Beijing metropolitan area.

3.5 FURTHER DISCUSSIONS
The objective of this chapter is to understand better the population distribution
and its evolution in the Beijing metropolitan area during the post-reform era. Our findings
suggest similar trends and patterns for the Beijing metropolitan area to those observed in
large Western cities. The population has spread with rapid urban growth, and the compact
urban form has been replaced by a more dispersed polycentric spatial distribution.
However, compared with the decentralization of Western cities, the spatial extent of the
decentralization of population in the Beijing metropolitan area is quite limited. We find
people have moved out of the inner city, but concentrated in the near suburbs, instead of
dispersing throughout the metropolitan area. The rapid growth of population in the near
suburbs has expedited the sprawl of the central city, with a larger central agglomeration
of population dominating the metropolitan area. In this sense, the spatial pattern of the
Beijing metropolitan area is still characterized by the continuous compactness. This is
also endorsed by the fairly good fit of the monocentric density function applied to
modeling the density patterns in the metropolitan area.
Although most scholars still regard Beijing as a monocentric city, our findings
provide the evidence that the city has been turning to a polycentric structure. We find

94
significant population subcenters have emerged in the suburbs of Beijing since the 1980s.
However, the polycentricity emerged in the Beijing metropolitan area is very different by
nature from that observed in Western cities. Clark and Kuijpers-Linde (1994)
summarized two different models of polycentricity when studying two prototype regions
of polycentric structures – the Randstad and Southern California: the market-driven
polycentricity portrayed as one of emerging urban centers with shifts in the hierarchy of
centers, and the history-based polycentricity portrayed as one of a collection of separated
urban centers in which locality and history play an important role. Clearly, the
polycentricity emerged in Beijing is different from the both models and has different
origins. The structure was initially driven by the planning efforts to promote dispersed
and scattering metropolitan development and the designation of numerous edge
development areas and satellite towns. The subcenters emerged are adherent to the
development scheme planned for the city, so it can be referred to as the so called
“planned polycentricity”.
Suburbanization in the Beijing metropolitan area has been caused by the same
factors as the reasons of suburbanization in the West, such as the improved transport
system, growing affluence, and rapid urban growth (Zhou and Ma, 2000). But these
factors have taken their effects in a totally different context. The process of
suburbanization and urban spatial restructuring that happened in the Beijing metropolitan
area during the post-reform era needs to be understood with reference to the peculiar
Chinese context: transition from a planned to a market economy and integration into the

95
global economy, affected by the economic reforms and open-door policies since 1978.
The process has been influenced by a series of reforms, including the decentralization of
decision-making and fiscal powers from the central to local governments, marketization
of urban land-use rights and housing, deindustrialization and internationalization of the
urban economy, and diversification of investment for urban development (Ma, 2004).
The most important change in China’s political economy ever since the reforms
has been the decentralization of decision-making and fiscal powers, which gave more
autonomy to local governments in revenue mobilization, investment, and urban
management (Wu, 1998b). Wu and Yeh (1999) argued the decentralization policies that
started in the mid-1980s have fundamentally changed the organization of urban
development in China. Before the reforms, urban development in China followed the
project development scheme of the centrally-planned economy, with state enterprises
rather than local municipalities as the main actors in the organization of urban
development. In such a system, state enterprises provided services, facilities and housing
to their employees, and therefore they directly organized the development of housing,
facilities and even infrastructure. The predominant role of state enterprises in urban
development has also been evident in the development of Beijing in the early years of the
reforms. For example, the subcenter 2 and 7 in the western and southwestern suburbs of
Beijing emerged in 1982 and 1990 respectively are just attributed to two of the largest
state enterprises in Beijing: the Capital Iron and Steel and the SINOPEC Beijing Yanshan
Petrochemical. After the reforms, the decentralization has localized the urban

96
development process and strengthened the status of local municipalities in urban
development and management. The role of state enterprises has been weakened. The
municipality began to play a more active role in urban development through city-wide
comprehensive development and manage urban development using urban planning.
Moreover, the decentralization of fiscal powers has enabled local governments to raise
revenues through local investments, which stimulated localities to actively promote
investments in urban development. Meanwhile, the influx of foreign investments since
the 1990s has been another driving force in urban development. All these explain the
pronounced improvement and rapid construction of urban infrastructure in Beijing after
the 1980s that contribute to the decentralization of population.
Another important change that fundamentally affected urban development in
China is the establishment of the land and housing market in the 1980s. Urban land is
owned by the state in China. Before the reforms, actual land use rights were held by
various state enterprises. The reform introduced the land leasing system, which separates
the land use right from the ownership of land and allows for the paid transfer of land use
rights. Under the new system, the municipal land administration bureau has taken charge
of land leasing and allocated land to different uses. This has reinforced the influence of
municipalities on the urban development pattern through comprehensive planning. Most
importantly, the land leasing system has created the differential rent, which enabled the
local government to finance urban infrastructure and redevelopment through relocation.
With the deindustrialization and internationalization of the urban economy, numerous

97
redevelopment projects have been implemented in the inner city of Beijing since the
1980s, with old houses and factories replaced by modern commercial complexes and
apartment buildings. The first wave of suburbanization in Beijing in the 1980s was
mainly driven by the relocation of residents displaced by the redevelopment projects and
by the relocation of industries, from the inner city to the suburbs. Meanwhile, the land
reform has essentially made the municipality play the role of land developers. Land
leasing has become a major way to raise local revenues and to attract investors to
stimulate local economies. To acquire more benefits, the local government expedited the
land transaction, especially the conversion of rural land at the periphery of the urban area
to urban uses, as redevelopment of existing urban land was more expensive and lowered
the potential profits of the development. In Beijing, the cost of land for urban
development was 120 times higher in the inner city than on the urban fringes (Lin,
2007b). This price difference induced rapid growth of the near suburbs of Beijing in the
1990s, with large scale development projects, like industrial zones, science parks and
residential communities. The concentration of population in the near suburbs in the 1990s
can be attributed to this process.
The establishment of the housing market improved the residential mobility in the
city, as people could purchase housing on the market and decide where they lived. Since
land was more readily available and much cheaper in suburban locations, large scale
residential communities were built there, especially in the near suburbs of Beijing, where
urban infrastructure were generally better developed. With increasing purchasing power

98
and rising standard of living, people have moved to the near suburbs, leading to the rapid
suburban growth and the emergence of subcenters. Another factor that contributed to the
suburban growth is the massive influx of rural migrants in the 1990s, with the gradual
relaxing of the migration control by the government. The migrants were mostly
concentrated in the periphery of the urban area, where they were easily accessible to jobs
and the housing rent was generally low. As a result, concentrated areas of rural migrants
have made a specific form of suburban settlements, migrant enclaves (Gu and Shen, 2003;
Deng and Huang, 2004).
In conclusion, the decentralization of population in the Beijing metropolitan area
is broadly analogous to that observed in western large metropolitan areas. However, the
process and driving forces are specific and need be understood with reference to the
peculiar Chinese context. With the transition from a centrally-planned to a market-based
economy, the process of suburbanization in Beijing has become more driven by market
forces (Feng et al., 2008). However, planning interventions will still play an important
role, and the decentralization process will further impact the urban landscape greatly in
the future.

99
CHAPTER 4.
EMPLOYMENT DISTRIBUTION AND SPATIAL ORGANIZATION OF
ECONOMIC ACTIVITY IN METROPOLITAN BEIJING

4.1 INTRODUCTION
Since the late 1980s, an extensive literature has documented the “qualitative
change” in the spatial structure of contemporary metropolitan areas (Anas et al., 1998).
Although there has been less agreement on the nature of emerging spatial structure in
terms of employment distribution in the metropolitan area of the future (Lee, 2007), the
broad consensus is the overall trend of decentralization of employment from the
traditional central business district (CBD) to the suburbs and the emergence of
employment centers outside the CBD. Glaeser and Kahn (2001) acknowledged most
American cities have been decentralized by the end of the 1990s, and the decentralization
of employment has fundamentally changed the spatial organization of contemporary
metropolitan areas. However, almost all of the empirical studies so far conducted have
been based on the North American or European experience, with only few exceptions,
such as Jun and Ha (2002). Responding to similar global forces of economic restructuring
and technical advances, spatial transformation broadly analogous to what has happened in
American metropolitan areas may have become a norm across countries (Freestone and
Murphy, 1998). But empirical evidence from outside North America and Europe,
especially from developing countries, is still rare. Freestone and Murphy (1998)

100
emphasized that despite the general convergence of metropolitan decentralization and
subcentering trends across countries, the spatial form emerging and the process involved
may be culturally, historically and locally specific. Therefore, it is interesting to conduct
comparative analyses that look beyond the North American and European contexts.
This chapter aims to establish some stylized facts about the spatial organization of
contemporary Chinese metropolitan areas in terms of employment distribution through a
case study of the Beijing metropolitan area. To our knowledge, this kind of study has
never been conducted before mainly due to a lack of adequate data. Our analysis is based
on an original dataset drawn from the 2001 basic establishment census of Beijing. The
establishment census is workplace based and collects information of major economic
activity in the city. Two general approaches are applied and distinguished in our study.
The first concerns the overall distribution and agglomeration of employment throughout
the metropolitan area (Section 4.3) and the second focuses on the local structure and
subcentering of economic activity (Section 4.4). Both approaches are built on some geo-
statistical techniques. A description of our data is first provided in the following section.

4.2 DATA ISSUES
The analysis in this chapter relies on establishment-level data drawn from the
2001 basic establishment census of Beijing. The 2001 census is the second national
census of basic units and establishments in China following the first census in 1996. It
surveys all corporate units and establishments except household and self-employed

101
business in both urban and rural areas. The census offers background information, main
attributes and business data of establishments, such as the title, location, industry
category, the amount of employees, operating status, operating income, and so on.
Therefore it is an ideal data source for our study.
We analyze the spatial distribution of establishments and employment using point
data geocoded to street addresses. Most previous empirical studies used aggregate areal
data in their analysis. However, the analysis using aggregated census data is usually
sensitive to the boundary delineation of areal units and the scale of data aggregation.
Besides, aggregate data also mask the spatial heterogeneity within a unit of analysis, and
therefore, to some extent, misrepresent the spatial distribution of economic activity (Luo
and Wei, 2006). Point data have particular advantages in depicting the “real-world”
employment distribution, as they represent the true locations of establishments. However,
the geocoding of the data is generally difficult. Although the census database contains
establishments’ physical addresses, some of this information is simply incomplete or
inappropriate. Moreover, with the rapid development of the city through the years, some
street addresses contained in the database are not valid any more due to the great changes
of the road network of Beijing, and they can not be matched with the latest streets file. In
the end, exact latitude/longitude coordinates can be identified for about 90% of
establishments in the census database. Though the failure to geocode all the
establishments raises the possibility of sample selection bias, we assume this would not

102
greatly affect our analysis, considering the large share actually sampled and the lack of
apparent systematic omissions in the data.
Figure 4.1 maps the distributions of establishments and employment in the
Beijing metropolitan area. Both spatial patterns appear highly concentrated around the
central city within the 5
th
ring road. The concentration of establishments and employment
outside the central agglomeration is also evident mainly along the radial arterials. Table
4.1 reports summary information on the establishments and employment by economic
sector. The total number of establishments in the Beijing metropolitan area is over two-
hundred thousand. The total employment is about nine million and the mean
establishment size is 42.1 employees. Services, retail trade, and manufacturing sectors
constitute the majority of employment and firms in the city, accounting for 70.88% and
56.19% of the total establishments and employment in the metropolitan area respectively.
The mean sizes of establishments in the retail trade, wholesale trade and services sectors
are all below average, and the mean size of establishments in the manufacturing sector is
about 56% larger than average. The variations of size distributions of establishments in
the retail trade, wholesale trade and producer services sectors are generally small. And,
the transportation and warehousing sector has the largest mean value and variation in the
size of establishments among all the sectors.

103
Figure 4.1 Distributions of Establishments and Employment

104
Table 4.1 Summary Information on Establishments and Employment by Economic Sector
Sector Establishments Employment
Code Total PercentTotal PercentMean S.D.
M 2611112.21 171982119.11 65.87 463.46
T&W 2234 1.04 304188 3.38 136.16 2253.83
WT 17885 8.36 417704 4.64 23.35 56.11
RT 47246 22.10 986788 10.96 20.89 83.97
FIRE 7725 3.61 422179 4.69 54.65 562.70
S 7818936.57 235118826.12 30.07 288.86
CS 16595 7.76 573846 6.38 34.58 103.88
PS 46028 21.53 1017673 11.31 22.11 81.05
H&S 14110.66 1517371.69 107.54 326.76
E&R 5505 2.57 480244 5.34 87.24 252.92
A,S&B 3353 1.57 147885 1.64 44.11 154.00
G 1963 0.92 152872 1.70 77.88 242.08
MSA Total 213810 9000852 42.10 464.71
Note: Sector Codes - M: Manufacturing; T&W: Transportation and Warehousing; WT:
Wholesale Trade; RT: Retail Trade; FIRE: Finance, Insurance and Real Estate; S:
Services (including public services, consumer services, producer services, and other
social services); CS: Consumer Services (including personal services, hotels and
restaurants, tourism, entertainment and recreation); PS: Producer Services (including
information and consulting services, computer services, professional and technical
services); H&S: Health Care and Social Welfare; E&R: Education and Research; A,S&B:
Arts, Sports and Broadcasting, TV and Films; G: Government

4.3 EXPLORING SPATIAL DISTRIBUTION AND CLUSTERING OF
EMPLOYMENT
4.3.1 Analytical Techniques
The establishment-level point data makes the analytical methods based on areal
aggregations normally used in the urban literature not suitable for our analysis. Our study
applies an alternative methodology, called “point pattern analysis”, which aims to
identify patterns in spatial point data. Point pattern analysis techniques first began with

105
the work of plant ecologists and botanists in the 1930s (Boots and Getis, 1988), and
thereafter have been widely used in many different fields, such as epidemiology,
criminology, and archeology, etc. Studies that applied the point pattern approach to
examine the structure of urban form are still quite few. The first application is by Getis
(1983) to explore population distribution in the Chicago region.
Spatial point patterns are based on the coordinates of events, such as the locations
of establishments in our case. Points can also have attribute information, known as
marked points. In our study, marks are the number of employees of each establishment.
The objective of the analysis is to describe and characterize distribution patterns of point
events, and to test and detect significant spatial clustering of points in a particular area,
which usually involves the visualization, exploratory analysis and modeling of spatial
point data.
Geographic description of point data has usually involved three aspects: central
tendency, dispersion and orientation in space (Yuill, 1971). One tool that can capture all
these properties of distribution of point sets is the standard deviational ellipse, which was
first devised by Lefever (1926). The standard deviational ellipse is a centrographic
measure designed to summarize the distribution (both the dispersion and orientation) of a
set of points around the mean center. The average variation in the distance of points from
the mean center can be viewed as a circle or set of circles at a set of standard distances,
like standard deviations in univariate statistics (De Smith et al., 2008). If the set of points
exhibits a directional bias, separate variations along axes that are orthogonal to each other

106
can be derived to define a standard distance ellipse, with the major and minor axis
indicating the direction of maximum and minimum spread respectively, and the angle of
rotation that corresponds to the geographic orientation of the point distribution (Wong
and Lee, 2005).
The procedure to compute the standard deviational ellipse is to first determine the
mean center of the point set and define it as the origin of a set of axes for the point
distribution. The mean center is defined by the means of x- and y-coordinates, which is
also the geometric center of the point set. In our study, x- and y-coordinates are both
weighted by the employment to calculate the weighted mean center, the location of which
is thus influenced by the number of employment as well as the location of each
establishment. The equation has the form:
11
11
,
nn
ii ii
ii
nn
ii
ii
xwyw
xy
ww
==
==
==
∑∑
∑ ∑

where x and y are the coordinates of the weighted mean center, x
i
and y
i
are the
coordinates of establishment i, w
i
is the number of employees of establishment i, and n is
the number of establishments. Then the coordinates of the point set are transformed:
,
ii i i
x xx y y y ′ ′ =−=−
since the weighted mean center becomes the new origin. The two axes are also rotated to
minimize the squared deviations from the point set to the rotated axes. The angel of
rotation ( θ) is calculated as:

107
22
22 2 2
11 11 1
1
1
4
tan
2
nn n n n
ii ii ii ii i i i
ii ii i
n
ii i
i
xw y w xw y w xyw
xyw
θ
== == =
−
=
⎛⎞
⎛⎞⎛⎞⎛ ⎞
⎜⎟
′′ ′′ ′′ −+ − +
⎜⎟⎜⎟⎜ ⎟
⎜⎟
⎝⎠⎝⎠⎝ ⎠
=
⎜⎟
⎜⎟ ′′
⎜⎟
⎝⎠
∑∑ ∑∑ ∑
∑

With the value of θ, the two standard deviations along the rotated x- and y-axes are then
calculated as:
() ()
22
11
11
cos sin cos sin
2 , 2
nn
ii i i i i
ii
xy nn
ii
ii
x yw y x w
ww
θθ θ θ
σσ
==
′′
==
′′ ′ ′ −−
==
∑∑
∑∑
ii

Notes that weights are also included in determining the orientation and axes of the ellipse.
The standard deviational ellipse is a useful graphic representation of point data.
Its application is relatively simple and clear. Lefever (1926) suggested that the major axis
of the ellipse indicates the spatial orientation of the point pattern, and the area of the
ellipse indicates the concentration or scatteration of the point set. A relatively small
ellipse indicates that the point set is more concentrated, and vice versa. Therefore,
differences among various point distributions can be identified.
Second-order properties of point patterns can also be useful in understanding the
organization of the data. These properties, also referred to as spatial dependency among
points, are usually investigated based on the distances between pairs of points, so called
“nearest neighbour analysis”. Here, we apply Ripley’s K function to test for spatial
dependence or clustering among establishments (Ripley, 1976). Geographic

108
concentration of industries has long been of interest to urban planners and economists.
Agglomeration economies created when firms cluster together in space stimulate the
emergence and growth of urban centers, and are the fundamental forces in determining
the intra-urban spatial structure. The K function is an effective tool to measure the levels
of agglomeration across industries for multi-spatial scales, and provides insights to the
spatial organization of industries within the metropolitan area.
The K function is defined as:
2
11
() ( )
nn
hij j
ij
ji
R
Kh I d e
E
==
≠
=
∑∑
i
where K(h) denotes the K function evaluated at distance h, I
h
(d
ij
) is an indicator function
that is 1 if point j is located within the circle around point i defined by h, e
j
is the number
of employees of establishment j, E represents the total employment of all establishments,
and R denotes the total area of the study region. To compute the K function, a circle of
radius h is placed around each establishment i, and the number of employees of all other
establishments j within the circle is counted and summed over all establishments within
the study region. Then the radius of the circle is increased incrementally, and the process
is repeated. By the end, K values can be graphed against distances h to reveal whether
there is any clustering at certain scales. In this sense, the K function can be regarded as a

109
super-order nearest neighbour statistic. In practice, the K function is usually transformed
to the L function to make the plot more linear, and the transformation has the form:
()
()
Kh
Lh h
π
= −
It should be noted that the K function above is generally biased due to edge
effects. As points outside the boundary of the study region are excluded, the number
enumerated by any circle for points located near the boundary is generally less than for
points in the center. So the K function need to be adjusted to correct this bias. One way is
to place a weight to let points near the boundary of the study region receive a greater
weight. Therefore the bias-corrected K function becomes:
2
11
() ( )
nn
ij h ij j
ij
ji
R
Kh w I d e
E
==
≠
=
∑∑
ii
To understand whether the observed pattern of establishment locations is different
from complete spatial randomness (CSR), the observed K function is normally compared
to simulated K functions that are constructed under the CSR hypothesis, whereby the
points constitute a partial realization of a homogeneous Poisson process. However, Feser
and Sweeney (2000) noted that it is not appropriate to use the CSR as the baseline in
analyzing the spatial concentration of industries, since the overall distribution of
establishments is inherently heterogeneous. Moreover, the distribution of establishments
within the metropolitan area usually has a strong central tendency, that is, a high degree
of concentration around the city center. Therefore, localized second-order clustering can

110
not be disentangled from the first-order effect that is so dominant through the K function,
especially for large distance ranges. One way to avoid these problems is to employ a
case-control framework (Feser and Sweeney, 2000; Diggle and Chetwynd, 1991). Here,
we set the distribution of K for the employment of all industries within the metropolitan
area as the baseline, against which spatial clustering of various industries is then
compared. As all industries are subject to similar first-order properties, the primary first-
order effect is then canceled out with the differences in the K distribution indicating
whether a certain industry has a greater tendency to cluster than the general pattern of all
industries. The D function is designed to measure these differences, and has the form:

() () ()
industry i total
D h Lh Lh = −
Positive values of D imply spatial clustering, which mean that the average number of
employees found within a circle of a particular radius over all establishments for a certain
industry is greater than that found for the overall distribution of all industries.

4.3.2 Results and Discussion
The estimation of both the standard deviational ellipse and the K function is
carried out using the CrimeStat software package (Levine, 2007), and the visualization of
various results is accomplished using ArcGIS. Figure 4.2 shows the standard deviational
ellipse fitted for each economic sector. The ellipses are overlaid together in space, with
the mean centers for different sectors not far away from each other and all located near
the city center. The general pattern of ellipses suggests a distinct directional bias for most

111
of the sectors. The geographic orientation of employment distribution, represented by the
angle of rotation of the major axis of the ellipse, is mainly along the northeast-southwest
direction for almost all the sectors. The producer services and arts, sports and
broadcasting sectors have the most circular employment distribution, which means the
employment of the two sectors has distributed more equally in all directions around the
mean center. The ellipse for the transportation and warehousing sector is shown as the
most narrow and elongated one, which suggests a polarized employment distribution of
the sector that is largely due to the concentration of the transportation and warehousing
establishments at several spots around the airport, railway hubs, ports mainly located in
the northeast and southwest suburbs.
A relatively large area of the ellipse for the manufacturing sector indicates a much
scattered and dispersed distribution of manufacturing employment. Table 4.2 summarizes
the distribution of both establishments and employment by economic sector at different
spatial scales within the Beijing metropolitan area. It shows that only 8.89% of
employment in the manufacturing sector is located in the inner city, and over 90% of
manufacturing jobs are spread throughout the suburbs, with over 50% in the near suburbs
and about 40% in the outer suburbs. This suggests that manufacturing activity is very
widely distributed in the metropolitan area, and the sector is highly decentralized.
Previous studies of the suburbanization of Beijing have pointed out that manufacturing
industries have suburbanized in Beijing, together with the population, since the 1980s,
first stimulated by the redevelopment of the inner city and further accelerated by the

112
establishment of the land and housing market as well as the large scale development of
suburban industrial parks in the 1990s (Zhou and Ma, 2000; Feng et al., 2008).
Figure 4.2 The Standard Deviational Ellipses for Various Economic Sectors

113
Table 4.2 Distributions of Establishments and Employment at Different Spatial Scales
Establishments
Inner City Within 5
th
Ring Road Near Suburbs Outer Suburbs Sector
Code (%) (%) (%) (%)
Total 54327 25.41 162367 75.94 120143 56.19 39340 18.40
M 2117 8.11 11589 44.38 11700 44.81 12294 47.08
T&W 305 13.65 1577 70.59 1446 64.73 483 21.62
WT 5770 32.26 15051 84.15 9921 55.47 2194 12.27
RT 12773 27.04 37840 80.09 27702 58.63 6771 14.33
FIRE 2339 30.28 6085 78.77 4037 52.26 1349 17.46
S 22568 28.86 68228 87.26 49570 63.40 6051 7.74
CS 6033 36.35 13939 84.00 8856 53.37 1706 10.28
PS 12570 27.31 41478 90.11 31184 67.75 2274 4.94
H&S 324 22.96 803 56.91 655 46.42 432 30.62
E&R 1351 24.54 3650 66.30 2781 50.52 1373 24.94
A,S&B 1379 41.13 2985 89.02 174051.89 234 6.98
G 582 29.65 1079 54.97 606 30.87 775 39.48
Employment
Inner City Within 5
th
Ring Road Near Suburbs Outer Suburbs Sector
Code (%) (%) (%) (%)
Total 2374114 26.38 6584321 73.15 4866114 54.06 1760624 19.56
M 152912 8.89 779694 45.34 881780 51.27 685129 39.84
T&W 126963 41.74 211255 69.45 98594 32.41 78631 25.85
WT 147851 35.40 357602 85.61 222217 53.20 47636 11.40
RT 300350 30.44 796739 80.74 547282 55.46 139156 14.10
FIRE 185084 43.84 355443 84.19 188304 44.60 48791 11.56
S 671897 28.58 2065753 87.86 1507452 64.11 171839 7.31
CS 214635 37.40 486861 84.84 301707 52.58 57504 10.02
PS 258948 25.45 912785 89.69 715081 70.27 43644 4.29
H&S 57851 38.13 111455 73.45 66101 43.56 27785 18.31
E&R 93188 19.40 333191 69.38 293315 61.08 93741 19.52
A,S&B 67524 45.66 136417 92.25 72318 48.90 8043 5.44
G 64476 42.18 111684 73.06 51724 33.83 36672 23.99

114
Figure 4.2 shows that manufacturing is the most dispersed sector, and the
distribution of employment of production-oriented sectors, such as manufacturing and
transportation and warehousing, is generally more dispersed than that of service and
trade-oriented sectors. This is supported by the fact shown in Table 4.2 that more than
80% of employment in the services and trade sectors is concentrated within the 5
th
ring
road that is approximately regarded as the boundary of the central city (or the
continuously built-up urban area) of Beijing. Among all the sectors, economic activity in
the producer services and arts, sports and broadcasting sectors is the most compactly
distributed. About 90% of employment in the two sectors is contained within the 5
th
ring
road, and near to half of employment in the arts, sports and broadcasting sector is located
in the inner city. Whereas, producer services activity is more concentrated in the near
suburbs, with only 25.45% of employment inside the inner city and 64.24% of
employment outside the inner city but within the 5
th
ring road.
To explore further the spatial clustering of employment, we estimate the K
function for various sectors. The K function is designed to measure second-order
clustering, and therefore it is usually analyzed for only a short distance. Previous studies
suggested that the estimate of the K function at larger distances than one-third the linear
extent of the study region is usually inefficient (Cuthbert and Anderson, 2002). In this
study, the maximum distance of analysis is set at 3 R , where R is the area of the study
region, and the number of distance intervals is set to 100. We apply the circular

115
correction procedure in the CrimeStat software package to help reduce the edge-effects
bias in computing the K function (Levine, 2007).
Figure 4.3 The D Functions for Various Economic Sectors
0
2000
4000
6000
8000
10000
12000
0 5000 10000 15000 20000 25000 30000 35000
Distance (Meters)
D Value
A,S&B FIRE CS PS S RT WT

-14000
-12000
-10000
-8000
-6000
-4000
-2000
0
2000
0 5000 10000 15000 20000 25000 30000 35000
Distance (Meters)
D Value
E&R H&S G M T&W

116
Figure 4.3 plots the D value calculated based on the estimate of the K function for
each economic sector against the distance. The results suggest a similar pattern of spatial
clustering of industries to what has been revealed by the standard deviational ellipse.
Manufacturing is the most dispersed sector with large negative D values over all
distances, while the distribution of employment in the arts, sports and broadcasting and
producer services sectors is observed to be the most clustered. More specifically, the
results for different sectors can be roughly categorized to three types. Spatial clustering is
observed for the services, trade, FIRE and arts, sports and broadcasting sectors. In
general, service-oriented sectors are observed to be more concentrated than trade-oriented
sectors indicated by the larger positive D values. The D values of all these sectors peak at
a distance range of 10~15 kilometer. On the other hand, sectors demonstrating spatial
dispersion are manufacturing, transportation and warehousing, education and research,
and health care and social welfare. The dispersion of employment in social service-
oriented sectors, such as health care and education, is simply due to the dispersed nature
of the distribution of health care and educational facilities. At last, different patterns at
different spatial scales have been revealed for the government sector. We find positive
clustering of employment in the government sector at a tight distance range of less than 8
kilometer, and at distances beyond that range, the sector becomes dispersed.
The analysis using global spatial statistics reveals that although manufacturing has
been largely decentralized in the Beijing metropolitan area, service and trade-oriented
activity has not followed the trend of suburbanization. Employment in the services and

117
trade sectors exhibits the strongest agglomeration tendencies between 10 and 15
kilometer, a range comparable to the scale of the central city, which indicates a strong
central agglomeration of services and trade activity. The global statistic is generally not
useful in examining local variations in the distribution of employment, which is the key
to capture the structure of urban form. In the following section, we apply local statistics
to identify employment centers and further explore the intrametropolitan spatial structure
of Beijing.

4.4 EMPLOYMENT CENTERS AND URBAN STRUCTURE
4.4.1 Identification of Employment Centers
Some decentralization of employment has clearly occurred in Beijing, although
mainly in the manufacturing sector, and the phenomenon appears to be much less
advanced than in Western countries. The next question would be how this
decentralization of employment affected urban spatial structure of Beijing. Standard
urban economic theory has suggested that with the growth of a metropolitan area,
decentralized employment may be dispersed throughout the metropolitan area or
concentrated in employment centers outside the CBD, called “subcenters”, depending on
the relative strength of agglomeration economies and congestion costs. Similar to a CBD,
employment subcenters offer agglomeration economies to firms, while potentially
reducing congestion costs of a monocentric city (McMillen, 2004). Therefore, some
scholars argued that it is employment, not population, that is the key to understanding the

118
formation of urban centers and the nature of the spatial structure of a polycentric city
(McDonald, 1987; Giuliano and Small, 1991).
An employment center is normally defined as a large concentration of economic
activity that has significantly larger employment density than surrounding areas and a
significant impact on the overall distribution of urban population, employment and land
prices (McMillen, 2001, 2004). Previous studies have used a variety of techniques to
identify employment centers, however no generally accepted and systematic
methodology exists (see Coffey and Shearmur, 2001, for a review). In this section, we
propose an alternative approach for identifying employment centers to methods based on
aggregate areal data, which employs detailed disaggregate spatial point data.
The development of the methodology starts with a transformation of geo-
referenced points to a continuous employment density surface. The generation of the
surface is accomplished using the kernel density interpolation. Kernel estimation is a
most commonly used interpolation technique that is particularly compatible with spatial
point data. It requires a point pattern as input, and the output is a continuous surface that
provides density estimates for all locations of the entire study region (Maoh and
Kanaroglou, 2007). The estimated kernel surface allows visualizing and further exploring
local variation of employment densities in space.
In statistics, kernel density estimation is a nonparametric way of estimating the
probability density function from observed data. In point pattern analysis, kernel methods
are used to create surface representations of point events in a similar way. The process

119
involves placing a symmetrical surface (the kernel function) over each point, and the
underlying density is estimated by summing the kernels across points for all reference
locations, to produce a smooth cumulative density surface. In practice, various kernel
functions have been employed, while the normal distribution is the most often used,
which has the functional form:
2
2 2
1
( ) exp( )
2 2
n
ij
i
hj i
i
d
E
xW
h h
λ
π
=
=× × −
∑

where d
ij
is the distance between point i and reference location j in the region, h is the
standard deviation of the normal distribution, also known as the bandwidth, W
i
is a
weight at the location of point i, and E
i
denotes the amount of employees of
establishment i. Although alternative kernels can be used, we apply the normal
distribution function in our study. Silverman (1986) has noted that it does not make much
difference in the shape of the interpolated surface as long as the kernel is symmetrical.
And, almost all authors agreed that the choice of the kernel function is less critical than
that of the bandwidth over which the kernel is applied (Thurstain-Goodwin and Unwin,
2000). The bandwidth controls the smoothness of the interpolated surface. The greater the
bandwidth, the smoother the resulting surface. However, in practice, there is no
consensus on how to select a particular bandwidth. Here we adopt an adaptive bandwidth,
which adjusts the bandwidth interval so that a minimum number of points are found. The
advantage of an adaptive bandwidth is that it provides constant precision of the estimate
over the entire region. The degree of precision depends on the sample size of the

120
bandwidth interval, which is set as a minimum of 100 points within the bandwidth radius
in our study.
The estimated kernel surface is then used for identifying employment centers
through examining the local spatial autocorrelation of employment densities. To do this,
we apply local indicators of spatial association (LISAs) (Anselin, 1995) on the
interpolated employment density. LISAs compare each unit’s employment density to that
of its neighbors, and identify significant positive or negative local spatial autocorrelation.
A positive autocorrelation indicates spatial clustering of similar (high or low) values of
employment density between neighboring spatial units. In this sense, LISAs are
particularly useful for identifying employment centers characterized by significantly
higher employment density than nearby locations, as a cluster of units with relatively
high employment density will show a significant positive autocorrelation (Riguelle et al.,
2007). However, we still need discriminate between clusters of high and low employment
density among clusters of positive autocorrelation. In that purpose, Moran scatter-plot
(Anselin, 1996) is used to identify four types of local spatial autocorrelation between a
unit and its neighbors: the H-H (L-L) type refers to a unit with high (low) employment
density surrounded by neighbors with high (low) employment density, and the H-L (L-H)
type refers to a unit with high (low) employment density surrounded by neighbors with
low (high) employment density. Therefore, an employment center is then defined as a
cluster of continuous units with significant local spatial autocorrelation of H-H (or H-L)

121
type. The application of LISAs and Moran scatter-plot together enables both the location
and the boundary of the employment center to be identified.
In this study, we use the local Moran’s I statistic to test for significant local spatial
autocorrelation, which is written as:
2
()
()/
i
iijj
j
i
i
xx
I wx x
xx n
−
= −
−
∑
∑

where x
i
is the employment density of unit i, x is the mean density across spatial units, n
denotes the total number of units, and w
ij
represents elements of a spatial weight matrix,
which indicates how unit i is spatially associated with its neighboring units j, so the
summation over j is such that only neighboring values of i are included. The spatial
weight matrix defines a spatial pattern exogenously, which connects each spatial unit to a
set of neighbors. In practice, various spatial weight matrices can be used, and the choice
of some specific matrices depends on the geographic configuration of the spatial units
(Baumont et al., 2004). Given the grid pattern of the interpolated density surface and the
equal size of the spatial units in our sample, there is not much difference between simple
contiguity, distance-based, and nearest-neighbors matrices. So we employ distance-based
spatial matrices, the general form of which is defined as followed:
*
*
*
() 0 if
() 1 if
() 0 if
ij
ij ij
ij ij
wd i j
wd d d
wd d d
⎧ ==
⎪
⎪
= ≤
⎨
⎪
= >
⎪
⎩

and

122
**
() ()/ ( )
ij ij ij
j
w d wd wd =
∑

where w
ij
(d) is an element of the standardized weight matrix with a critical cut-off
distance d, and d
ij
denotes the distance between unit i and j. We check the results using
different cut-off distances. The greater the cut-off distance, the more likely the value for
the neighborhood will be smoothed out, leading to a more general pattern with less local
variation. Due to the presence of global spatial autocorrelation, statistical inference must
be based on the conditional permutation approach. 999 permutations are used here to
compute the empirical distribution function, and the p-values obtained for the local
Moran’s I are then pseudo-significance levels (Anselin, 1995).
As Baumont et al. (2004) argued, the spatial autocorrelation analysis has its
particular advantage as an employment center identification method. First, it does not
require ample priori knowledge of the study region. Second, it allows formalizing the
notion of neighbors using different spatial weight matrices. Third, and most importantly,
it provides statistical tests that assess the significance of the results, indicating whether
the local clustering of units with relatively high employment density is significant
statistically. Therefore, employment centers identified with the spatial autocorrelation
method can be regarded as candidate centers with higher employment density than
neighboring areas. However, we still need to consider and examine the impacts of these
centers on the overall distribution of urban population and employment. So the
semiparametric procedure proposed by McMillen (2001) applied in Chapter 3 is used

123
here to assess the significance of the impacts of the centers on the overall urban spatial
structure.

4.4.2 Results and Discussion
To identify employment centers, we start with computing the kernel estimates
using the CrimeStat software package (Levine, 2007). Kernel estimation is conducted on
a reference surface with grid cells of 0.5 0.5 × kilometer. The interpolation process then
transforms discontinuous points into a continuous density surface. Figure 4.4 depicts the
resulting surface map, which suggests a multinucleate urban form of Beijing. The map
indicates that the largest concentration of employment is clearly located at the city core,
and its spatial extent is quite large, which implies the distribution of employment in the
Beijing metropolitan area is by and large highly centralized and concentrated. However,
the map also provides the evidence of clustering of employment in the suburbs, which
suggests possible suburban subcenters.

124
Figure 4.4 Kernel Employment Density Surface

Kernel estimates, which depict the expected number of employment per grid cell,
are then used to explore the spatial structure of employment distribution via the local
spatial autocorrelation analysis, which is implemented with the GeoDa software package
(Anselin, 2003). We experiment with different cut-off distances in the spatial weight
matrix. The cut-off distance needs to be larger than the dimension of the grid cell, while
larger distances will decrease local variability, increase aggregation bias, and make it
difficult to capture the local spatial effect that arises from the concentration of
employment. So by the end, we choose a cut-off distance of 1 kilometer as a compromise.

125
Figure 4.5 shows the type of local spatial autocorrelation to which each grid cell belongs
indicated by different colors, and the Moran significance map that portrays the
significance levels of LISAs with the shading colors.
It appears that most of the cells are characterized by positive spatial
autocorrelation. The cells with spatial autocorrelation of H-H type are mostly centrally
located, while some of them also cluster together in the suburban areas. The periphery of
the metropolitan area is clearly characterized by the concentration of low employment
density, indicated by the cells with spatial autocorrelation of L-L type. Although
significant autocorrelation of H-L type may indicate isolated employment centers,
negative autocorrelations only appear occasionally on the fringes of H-H or L-L clusters,
which turn out to be a marginal phenomenon. Therefore, an employment center is defined
as a cluster of continuous cells with significant positive autocorrelation of H-H type.
Figure 4.6 shows the significant positive autocorrelation of high employment density at
the significance levels of 1% and 5%. The boundary of the employment centers is finally
delineated based on the result at the 5% significance level, but we ignore four relatively
small-scale clusters in the southern and northeastern suburbs, which all consist of less
than 10 cells.

126
Figure 4.5 Spatial Autocorrelation Types and Moran Significance Map

127
Figure 4.6 Boundaries and Locations of Candidate Employment Centers

128
Figure 4.6 depicts the location and boundary of ten candidate employment centers,
including both the CBD and the suburban subcenters. The concentration of employment
at the CBD is the largest and almost circular with a radius of over 10 kilometers.
Nevertheless, the dominance and the relatively large spatial scale of the central
agglomeration of employment do not point to a perfect monocentric urban structure.
There are some other clusters of high employment density in the outer suburbs, which are
not included in the CBD and can be considered as subcenters. These subcenters are
scattered in almost every direction in the suburbs, with some close to the CBD and others
far away from it. The peripheral subcenters, such as Center 2, 5, 6 and 9, are all located in
the satellite towns, with one exception that Center 3 is located at the Capital International
Airport. And three subcenters on the fringe of the CBD, Center 4, 7 and 8, are all based
on some large industrial development areas. For instance, Center 8 is located at one of the
largest development zones and the only State-level development zone in Beijing, the
Beijing Economic and Technological Development Area (BDA), and Center 4 is one of
the important bases of high-tech industries (Shang Di) in Beijing. Overall speaking, our
findings suggest that the Beijing metropolitan area appears to have a polycentric urban
pattern.
To further examine the impacts of the candidate subcenters on the overall
distribution of urban population and employment, we apply the semiparametric procedure
as used in Chapter 3. Establishment-level employment data are first aggregated into
subdistricts, matched with the 2000 population census data. And the dependent variable

129
in the semiparametric regression is then the population and employment density of each
subdistrict. The distance variable of subcenters is measured as the straight-line distance
between the centroid of each subdistrict and that of each subcenter. The reverse stepwise
regression procedure is also used to solve the multicollinearity problem. The final list of
subcenters that have significant impacts on the intrametropolitan distribution of
population and employment includes those whose coefficients are significant at the 5%
significance level and have expected signs on the level or/and the inverse form of
distance variables.
Table 4.3 Final List of Employment Subcenters and Semiparametric Estimation
DepVar
ln(Employment Density)
DepVar
ln(Population Density)
Subcenter
ID
Distance
Variable
Estimate P-Value Estimate P-Value
x
ij

2
x
-1
ij
1.641 <.0001 1.462 <.0001
x
ij

3
x
-1
ij
2.405 <.0001 1.000 0.042
x
ij
-0.205 <.0001 -0.165 <.0001
4
x
-1
ij

x
ij

5
x
-1
ij
12.873 <.0001 12.229 <.0001
x
ij

6
x
-1
ij
6.175 <.0001 5.062 <.0001
x
ij
-0.309 <.0001 -0.368 <.0001
7
x
-1
ij

x
ij

10
x
-1
ij
5.173 <.0001 2.730 0.002
Adjusted R
2
0.879 0.864

130
Table 4.3 reports the semiparametric regression results. Two subcenters in the
southern suburbs, Center 8 and 9, are found to be insignificant at the 5% significance
level. As a result, we find seven significant subcenters, and their impacts on the overall
distributions of both population and employment are quite similar. It is shown that
subcenters 4 and 7 that are near the central agglomeration have effects spread over
relatively large areas with significant level distance variables, while the peripheral
subcenters have more local effects with significant inverse distance variables. This
demonstrates the dominance of the central agglomeration and its nearby locations, as well
as the limited impacts of the peripheral subcenters on the overall urban spatial structure.
Although the relatively large value of R
2
indicates a good fit of the polycentric model, the
dominance of the CBD and nearby subcenters highlights the centralized nature of the
spatial structure of Beijing.
Table 4.4 Summary of Employment Centers
Establishments Employment Center
ID
(%) (%)
Area
(km
2
)
Employment
Density
(1000/km
2
)
Distance to
City Center
(km)
CBD
1 157071 73.46 6522939 72.47 472.40 13.81 2.44
Subcenter
2 1096 0.51 45644 0.51 7.46 6.12 37.66
3 205 0.10 70785 0.79 7.71 9.18 24.48
4 3070 1.44 140648 1.56 24.63 5.71 16.63
5 2250 1.05 97722 1.09 13.93 7.01 25.86
6 2188 1.02 96730 1.07 15.42 6.27 21.68
7 766 0.36 53126 0.59 8.21 6.47 19.09
10 425 0.20 61323 0.68 7.46 8.22 43.36

131
Table 4.4 provides summary information of the eight employment centers, and
their locations and boundaries are shown in Figure 4.7. We call the central agglomeration
of employment as the CBD, but it is much larger in extent than the traditionally perceived
central business district. It is shown that the spatial extent of the central agglomeration
almost approximates the scale of the central city, when we compare its boundary with the
5
th
ring road (Figure 4.7). The total area of the central agglomeration is as large as 472.4
square kilometers, and is far larger than that of any subcenter. Considering the large
spatial scale of the CBD, there is no reason to believe it is monocentric by nature.
Actually, a subcenter in the west is shown to be separated from it, when a lower cut-off
significance level of 1% is used (Figure 4.6). It is not surprising to see the dominance of
the CBD from Table 4.4. It contains over 70% of establishments and jobs in the
metropolitan area, and its average density, 13.81 thousand employees per square
kilometer, is also much higher than those of subcenters. Compared with the CBD, the
subcenters are generally quite small in extent. And the area of the largest one is just about
25 square kilometers, and it only contains 1.56% of the region’s employment. The
densest subcenter is Center 3, located at the Capital International Airport in the
northeastern suburb.

132
Figure 4.7 Locations of Population Subcenters and Employment Centers and the
Development Scheme of Beijing

133
Table 4.5 Aggregate Employment by Sector inside and outside Centers
Inside Centers Outside Centers
CBD Subcenters
(%) (%) (%) (%)
Total 7088917 78.76 6522939 72.47 565978 6.29 1911935 21.24
A,S&B 139093 94.05 13706092.68 2033 1.37 8792 5.95
CS 509074 88.71 490133 85.41 18941 3.30 64772 11.29
E&R 337603 70.30 320364 66.71 17239 3.59 142641 29.70
FIRE 376697 89.23 360280 85.34 16417 3.89 45482 10.77
G 125706 82.23 113548 74.28 12158 7.95 27166 17.77
H&S 124211 81.86 113552 74.83 10659 7.02 27526 18.14
M 962961 55.99 771817 44.88 191144 11.11 756860 44.01
PS 956121 93.95 908925 89.31 47196 4.64 61552 6.05
RT 833518 84.47 784111 79.46 49407 5.01 153270 15.53
T&W 263533 86.63 203428 66.88 60105 19.76 40655 13.37
WT 362126 86.69 346756 83.01 15370 3.68 55578 13.31

Table 4.5 shows the aggregate employment by economic sector inside and outside
centers. It is evident that the subcentering of employment in the Beijing metropolitan area
is still quite weak, with only 6.29% of jobs concentrated within the subcenters, and over
20% of jobs just generally dispersed outside the centers. The majority of employment is
still in the CBD, which indicates a centralized pattern of employment location. The
services and trade-oriented sectors are shown to be more centrally located in the CBD,
and a relatively high percent of jobs in the manufacturing and transportation and
warehousing sectors are clustered into subcenters. Especially for the transportation and
warehousing sector, about 20% of jobs are located in the subcenters, which is attributed
to the concentration of employment in the airport subcenter. Manufacturing that is the
most decentralized sector in the Beijing metropolitan area has the most dispersed pattern

134
of employment distribution, with about 44% of jobs located outside the centers. The arts,
sports and broadcasting and producer services sectors are the most centralized, with over
90% of employment in the centers, mostly in the CBD.
Table 4.6 Location Quotients for Employment by Sector in Employment Centers
Location Quotients Center
ID A,S&B CS E&R FIRE G H&S M PS RT T&W WT
1 1.28 1.18 0.921.181.021.030.621.23 1.10 0.921.15
2 0.20 0.65 1.000.96 2.85 2.17 1.72 1.01 0.88 0.27 0.56
3 0.23 0.46 0.090.090.100.000.570.03 0.21 22.40 0.27
4 0.18 0.54 0.580.700.170.67 2.21 2.20 0.72 0.26 0.48
5 0.23 0.58 0.670.56 2.26 2.68 0.45 0.11 1.35 0.60 0.42
6 0.24 0.74 0.961.07 3.13 1.47 1.12 0.32 1.28 0.31 1.28
7 0.41 0.24 0.320.360.060.60 3.02 0.17 0.38 0.44 0.41
8 0.06 0.37 0.260.970.210.10 2.67 0.37 0.49 0.34 0.32
9 0.50 0.74 2.50 0.94 4.19 2.01 1.09 0.27 1.49 0.05 0.61
10 0.07 0.28 0.24 0.400.470.04 4.21 0.13 0.30 0.51 0.51

To characterize the employment composition of each center, we use the location
quotient calculated as the sector percentage of employment in each center divided by the
same sector percentage for the entire metropolitan area, which reflects the specialization
of the employment center as compared to the whole urban economy. Table 4.6 lists the
results of location quotients, with those greater than 2 highlighted. We include all 9
subcenters rather than only those significant in the analysis, as the two subcenters that are
shown to be insignificant can still be regarded as the potential employment centers in the
region. The CBD (Center 1) is shown to have a broad mix of economic activity mainly
specialized in services and trade. We apply a simple cluster analysis based on the Ward
algorithm to categorize the types of subcenters according to their specializations. The

135
result suggests three clusters best classify the subcenters. The first cluster consists of
subcenters 4, 7, 8 and 10, which are highly specialized in manufacturing. These centers,
with the exception of Center 10, are all large suburban industrial development areas.
Center 10 is dominated by one single large firm: the SINOPEC Beijing Yanshan
Petrochemical, which is the base of petrochemical industry of Beijing. The second cluster
contains centers highly specialized in government and health care and social welfare,
including subcenters 2, 5, 6 and 9. These centers are all located at the seats of
government for the districts, and are all planned satellite towns. So they act as regional
administrative centers that normally contain a broad mix of industries and attract other
major functions, such as manufacturing and retail. The third cluster comprises only one
subcenter, Center 3, which is located at the airport and highly specialized in
transportation and warehousing.
Figure 4.7 depicts the locations of both the population subcenters and the
employment centers that are identified in the Beijing metropolitan area. The most striking
observation is the coincidence of population and employment centers. It seems that these
centers have a high concentration of both population and employment, which suggests a
jobs-housing balanced pattern of the urban structure, although the broad mix of jobs and
workers within the center may not be compatible. The figure also shows that the
identified spatial structure is generally associated with the planned development scheme
of the city. Therefore, one could argue that the employment subcenters may not form as a

136
result of pure economic forces in the Beijing metropolitan area, but may be more likely to
be a result of the planning intervention.

4.5 SUMMARY AND DISCUSSION
Through examining the distribution pattern of employment in the Beijing
metropolitan area, our analysis provides the evidence of a polycentric spatial structure of
Beijing. Despite the emergence of suburban employment centers outside the CBD, the
dominance of the central agglomeration of employment at the city core suggests the
urban structure of Beijing is still highly centralized by nature. The central agglomeration
contains the majority (over 70%) of employment in the Beijing metropolitan area.
Compared with the generally dispersed pattern in most American metropolitan areas with
the majority of the region’s jobs (over 70% in general) outside the employment centers
(Giuliano and Small, 1991; Gordon and Richardson, 1996), the CBD in the Beijing
metropolitan area has a much higher concentration of employment and still plays a key
role in describing the location of the region’s jobs. And its spatial extent is quite large.
For example, it is four times as large as the core of the Los Angeles region in 1980 as
defined by Giuliano and Small (1991). Moreover, the employment subcentering
phenomenon in the Beijing metropolitan area is also not so phenomenal, with only about
6% of jobs clustering into subcenters. And the number of subcenters identified in the
Beijing metropolitan area is also far less than that in some American metropolitan areas
of equivalent extent. For example, Lee (2007) identified over 30 and 40 subcenters in

137
New York and Los Angeles respectively in both 1990 and 2000, while we only find 7
significant subcenters in Beijing.
Although the extent of employment decentralization is limited in Beijing, we do
find similar tendencies of decentralization and subcentering that are broadly analogous to
what has occurred in most Western metropolitan areas. North American and European
experiences indicate that contemporary metropolitan areas are mostly characterized by
decentralized patterns. And suburbanization with urban growth has been one major
feature of recent urban development around the world. In the U.S., suburbanization began
with the moving out of people from the central city towards the suburbs, further followed
by several waves of decentralization of economic activity, first involving consumer
services along with manufacturing functions, and then “back-office” activities, and more
recently high-order services, which has resulted in the emergence of subcenters that are
increasingly large and diversified with more comprehensive and higher-order economic
functions (Coffey and Shearmur, 2001). By contrast, the suburbanization of Beijing is
still in its initial stage. It is shown that a certain degree of decentralization of population
and manufacturing activity has happened in the Beijing metropolitan area since the 1980s,
however, high-order functions such as producer services and arts, sports and broadcasting
activities remain highly concentrated in the CBD to take advantage of the agglomeration
economies inherent to this location.
It is different from suburbanization in the U.S. that the process of decentralization
in the Beijing metropolitan area started with the dispersion of both population and jobs

138
simultaneously, initially stimulated by the redevelopment of the inner city and the
relocation of industries resulted from the land and housing reforms and the restructuring
of the urban economy. Therefore, not like in the U.S. where suburbanization was first
driven by the residential relocation, suburbanization in Beijing has witnessed the
redistribution of both population and firms ever since the beginning. But, the
decentralization of jobs is not the same in extent across sectors. Manufacturing has
become the most dispersed, due to the dominant process of the relocation of factories
from the inner city to the suburbs as well as the development of large-scale industrial
zones in the suburbs. And the general dispersion of services and trade-related activity is
still not evident in Beijing. Although the suburbanization of Beijing has been more driven
by market forces recently (Feng et. al., 2008), the spatial form of decentralization has not
changed greatly and the extent of decentralization is still limited.
Furthermore, the formation of employment subcenters in the suburban Beijing can
not be attributed to pure agglomeration economies, but a result of diverse processes, such
as the planning efforts to create a scattering layout of the city as discussed in Chapter 3.
The development of large-scale suburban industrial parks caused the formation of several
subcenters in the near suburbs, while satellite towns that were planned and developed as
independent centers from the central city have procured agglomeration economies of a
sufficient force to attract various types of activity and formed the subcenters with more
comprehensive functions in the outer suburbs. However, the forces and process involved
with the emergence of the subcenters still need to be studied more closely.

139
An interesting finding of our study is the similar distribution patterns of both
population and employment in the Beijing metropolitan area, with the coincidence of the
population and employment subcenters in space. However, it is not very surprising if
considering the spatial distributions of both population and employment in the Beijing
metropolitan area are highly associated with the planned development scheme of the city.
In some way, this may imply a “jobs-housing balanced” urban structure of Beijing.
Although jobs-housing balance has not been stated as a planning objective in the master
plan of Beijing, it can be implicitly achieved through the planning and development
process. The balance of jobs and housing in Beijing is in part as a legacy of the project-
led urban development process of the centrally-planned economy before the reforms in
China, with state enterprises providing services, facilities and housing to their employees
directly, and therefore it is generally not quite different where people worked from where
they lived. Although the establishment of the land and housing market after the reforms
has improved the residential mobility and changed the urban development process
fundamentally, jobs-housing balance has not been broken completely, as the
comprehensive planning through which the municipality managed urban development in
the post-reform era tended to concentrate population and industries into planned urban
centers through the control of land use, according to the scattering layout principle.
Jobs-housing balance has been considered as a controversial planning tool to
solve traffic issues in the planning literature (Giuliano, 1991; Levine, 1998). Therefore, it
is interesting to see how the balanced spatial structure of Beijing relates to the urban

140
commuting pattern. We will further investigate the evidence of jobs-housing balance in
the Beijing metropolitan area and its relationship with urban commuting in the following
chapter.

141
CHAPTER 5.
JOBS-HOUSING BALANCE AND URBAN COMMUTING
IN METROPOLITAN BEIJING

5.1 INTRODUCTION
Commuting, the movement between place of residence and workplace, is highly
related to the urban spatial structure reflected by the spatial distributions of people and
jobs. Research on urban commuting patterns helps understand the nature of urban spatial
structure, and meanwhile, urban form (land use patterns) strongly affects commuting
patterns. The relationship between spatial structure and urban commuting has become a
critical issue for urban studies and a major concern of urban planners and policy-makers.
However, this relationship is still not well understood despite a large volume related
research.
Many planners hold the view that land use patterns affect commuting behavior. In
planning practice, approaches to transportation issues through land use have been
proposed. Notable among these is the so-called “jobs-housing balance” strategy that aims
to promote spatial matches between workplace and affordable housing through patterns
of concentrated multi-use developments, which is supposed to reduce excess commuting,
and consequently to mitigate congestion and related environmental problems (Cervero,
1989; O’Kelly and Lee, 2005). Though it is still not clear how it relates to commuting
patterns or its effectiveness in reducing auto-dependency, congestion and air pollution,

142
jobs-housing balance as a planning tool has been accepted and applied by urban planners
and policy-makers (Giuliano, 1991; Peng, 1997). For example, Nowlan and Stewart
(1991) show a case of Toronto where planning interventions that improved the jobs-
housing balance have yielded demonstrable transportation benefits. And, Cervero (1996)
also reports the case that regions in California set jobs-housing balance targets in the
1980s, and he notes that due to growing skepticism over the effectiveness of policies
aimed at jobs-housing balance, enforcement of such targets has been further abandoned
later.
The viability of jobs-housing balance policy to transportation problems has been
questioned since it was first proposed (Giuliano, 1991; Giuliano and Small, 1993), while
more recently several studies do have provided moderate support for a relationship
between jobs-housing balance and commuting (e.g. Peng, 1997; Levinson, 1998; Shen,
2000; Sultana, 2002). Nevertheless, the controversy over jobs-housing balance and its
impacts on urban commuting has been far from resolved. Previous research has examined
the issues mostly based on experiences drawn from North American metropolitan areas.
In this chapter, we aim to supplement this strand of literature with research from a
developing and transition economy context. In our view, the Beijing case deserves
attention because rapid suburbanization and spatial restructuring have reshaped the
spatial form and commuting patterns of the city, and the pattern and process are
fundamentally different from those in North America due to different culture, institutions
and urbanization and planning histories.

143
A jobs-housing balance occurs when the number of workers residing in an area
approximates the number of jobs there. Therefore, an urban pattern that is considered
balanced is characterized by the fabric with population and employment approximately
equally distributed. The previous chapters have shown that the spatial structure of the
Beijing metropolitan area reflected by the distribution patterns of population and
employment can be viewed as approximately balanced in the way that both population
and employment centers of the city coincide with each other in space. The balanced
nature of the spatial structure of Beijing is partly attributed to the work-unit model of
urban development that characterizes most socialist cities before the reforms, which tends
to organize housing and service facilities along with production activities in the work-unit
compound. This induces an urban pattern with a homogeneous distribution of social
classes and the non-existence of spatial separation between workplace and residence (Wu,
1998b). After the market reforms in China, especially after the establishment of the land
and housing markets, the pre-existing workplace-residence tie in the city has been broken,
due to the decline of the work-unit system and resulting improved mobility of residents .
And meanwhile, the municipality has gained more political and fiscal power as well as
incentives to implement comprehensive urban development through allocating urban land
to different uses guided by the city plan. Therefore, land use planning has been
emphasized and employed to control urban development and implemented by physically
delineating land uses and balancing distributions of population and jobs in specific areas
in the city. Although the jobs-housing balance is not a clearly stated planning target in the

144
master plan of Beijing, it has been implicitly achieved through physically shaping the
spatial form of the city according to the scattering layout principle. Under this principle,
concentrated residential and industrial developments in the centers outside the central
area (such as in the satellite towns) were planned and developed to avoid the undesirable
sprawl of the central city, which has led to the spatial match between the concentrations
of population and jobs in the metropolitan area.
Planning through the control of land use has been taken as an important way in
Beijing to meet various challenges that the city is facing, such as accommodating rapid
urban growth as well as mitigating the deterioration of traffic conditions and protecting
cultural, historical, environmental and social capital. However, planning practices that put
great emphasis on shaping spatial forms of cities have been criticized for the lack of
consideration of integrating land use and transportation systems (Song et al., 2006), and
the physical planning approach is considered not sufficient as an effective way to solve
serious transportation problems nowadays in Beijing. Therefore, the debate over the
effectiveness of land use planning to transportation problems is also evident in the
planning of Beijing. In this chapter, we wish to investigate further the spatial structure of
the Beijing metropolitan area by analyzing detailed individual-level survey data
pertaining to commuting patterns. Two questions are raised. First, does commuting
patterns reveal a balanced urban structure of Beijing? Second, does urban spatial
structure, in terms of the jobs-housing balance, have an important effect on commute
duration, and what are the main factors that explain variations in commute time when

145
both socioeconomic characteristics of commuters and urban structure are considered?
The objective of this analysis is to understand better the nature of urban spatial structure
of Beijing and its impacts on urban commuting and to provide implications for planning
practices in Beijing.

5.2 DATA
Information on commuting patterns is not collected in the National Census in
China, so we are unable to use such regular census data as the CTPP (Census
Transportation Planning Package) in the U.S. in our analysis for the Beijing metropolitan
area. Our study in this chapter is mainly based on a survey of commuting behavior of
urban residents undertaken in Beijing in 2005. As the focus of the survey is on intra-
urban commuting, only the central urbanized area is selected as the survey area, and the
peripheral towns and townships that are mostly rural are generally excluded (Figure 5.1).
The survey covered 129 out of 232 subdistricts that comprise the Beijing metropolitan
area, covering an area of 1492.62 square kilometers that accounts for 16.4% of the total
area of metropolitan Beijing. The survey area had a population of 8,512,872 in 2000 and
7,397,881 jobs in 2001, accounting for 70.9% of the population and 82.2% of the jobs of
the Beijing metropolitan area respectively. Overall, there is a higher percentage of jobs
located within the survey area than that of population, while such a discrepancy is normal
as the central city usually contains more jobs than population, and workers may commute
from nearby rural areas.

146
Figure 5.1 The Survey Area

The survey had a relatively large sample size and collected information on a wide
range of topics, such as commute time and mode, subdistricts of residence and workplace,
and attributes of commuters and their households, etc. Excepting the traffic flow into and
out of the periphery, it is an ideal data source for our analysis. A total of 5225
questionnaires with completed information on the origin and destination (O-D) of
commuting trips were collected. For convenience, quota sampling techniques were
employed in selecting households and respondents. In quota sampling, the population is
first segmented into mutually exclusive sub-groups, and then the subjects are selected
from each segment based on a specified proportion. In our sample, the population was

147
stratified by subdistricts, and the respondents were selected from each subdistrict based
on an ad hoc quota that was proportional to its population size, so that the spatial
distribution of the sample approximated that of the population in the survey area. The
quota sample is a non-probability sample, therefore it may not provide an accurate
representation of the population in the survey area, but this is the data of best quality that
we can acquire so far.
Table 5.1 Summary of Respondents’ Socioeconomic and Commuting Attributes
Commute Time <=10 10-30 30-60 >60
(Minutes) 13.6% 42.0% 34.3% 10.1%
Transportation Mode Walk Bicycle Public Transit Drive
12.3% 22.6% 52.5% 12.6%
Age <30 30-39 40-49 >=50
(Years) 43.0% 23.7% 22.7% 10.5%
Gender Male Female
50.1% 49.9%
Household Size 1 2 3 4 and more
(Persons) 18.6% 16.6% 53.4% 11.5%
Education Junior High Senior High Undergraduate Graduate
7.0% 26.9% 60.3% 5.8%
Occupation Senior
Management
Middle
Management
Staff/Clerk
14.1% 25.7% 60.1%
Monthly Household
Income
<3000 3000-4999 5000-9999 >=10000
(RMB, Yuan) 25.8% 38.1% 28.1% 8.1%

The disaggregate individual-level data collected through the survey can be used to
test the relationship between urban structure and commuting patterns in the Beijing
metropolitan area. Table 5.1 summarizes the socioeconomic and commuting
characteristics of the respondents. It is shown that the sample is slightly skewed towards

148
highly educated workers, with 66.1% of the respondents with undergraduate or graduate
degrees. The total average commute time is 37.4 minutes in our sample, with over half of
the respondents spending less than half an hour on commuting. The majority of the
respondents are transit users, with 52.5% of the respondents using public transit as the
primary mode of commute, which includes bus, subway, light rail, company shuttle and
taxi. Non-motorized commute modes such as walking or bicycling are also important
ways of commuting in Beijing, with 12.3% and 22.6% of the respondents commuting by
walking and bicycles respectively. Despite the increase of the auto-ownership and the
auto-dependency in Beijing, the survey shows that only 12.6% of the respondents drove
to work. This result is partly due to the fact that we chose the central urbanized area as
the survey area and longer commutes using cars from the outer area to the central city
were not included in our sample.

5.3 EVIDENCE OF JOBS-HOUSING BALANCE
The previous chapters have revealed a balanced urban structure of metropolitan
Beijing with both the population and employment approximately equally distributed
throughout the metropolitan area. In this section, we will further investigate the
distributions of people and jobs and test the jobs-housing balance hypothesis in the
Beijing metropolitan area. The jobs-housing balance refers to the spatial relationship
between the number of jobs and housing units within a given area (Peng, 1997). A simple
and most often used measure for jobs-housing balance is the jobs-housing ratio,

149
calculated as the ratio of jobs to housing units in an area (Sultana, 2002), or in most cases
approximated by the ratio of jobs to resident workers or households (Cervero, 1989,
1996). In our study, due to the lack of the data on the number of housing units or resident
workers at the subdistrict level in the study area, jobs-housing balance is approximately
measured by the employment-population ratio (E/P ratio), which is calculated based on
the subdistrict-level population and employment data used in the previous chapters.
However, the population data drawn from the 2000 population census of Beijing are not
directly comparable to the employment data generated from the 2001 establishment
census in which self-employed workers are excluded. Therefore, the employment-
population ratio here is calculated as the ratio of the percentage of the employment in the
subdistrict to the percentage of the population, instead of using the absolute number of
people and jobs.
Another concern about measuring jobs-housing balance is the geographical scale
of analysis. As Cervero (1996) argued, the size of spatial units for measuring jobs-
housing balance does matter, and the larger the size, the more likely the balance. For
example, regions are balanced by definition as they are identified as economically self-
contained units (Giuliano, 1991). Therefore, Sultana (2002) argued that studies that
measured jobs-housing balance at the macro level are problematic. At the micro level,
jobs-housing balance is usually measured using such spatial units as census tracts or
TAZs. However, it may be misleading to consider jobs and housing balanced only when
the residents live and work in the same such small neighborhood as the census tract or

150
TAZ. And it is not reasonable to consider residents working in nearby neighborhoods as
imbalanced since they may commute only a short distance to their jobs. Therefore, most
previous research, e.g. Peng (1997), suggested that it is more appropriate to measure
jobs-housing balance at the meso-level, within a reasonable commute distance from a
given job or residential site. To define a reasonable commute range for measuring jobs-
housing balance in our case, we calculate the average straight-line distance between the
centroids of the subdistricts of home-site and workplace in our sample. It should be noted
that it is more reasonable to use the average home-to-work trip length here but
unfortunately commute distance data are not available in our dataset. Following Peng’s
methodology (1997), we employ floating catchment areas and measure the employment-
population ratio at a 9.5-kilometer buffer for each subdistrict.
The calculated E/P ratios are shown in Figure 5.2, compared with the identified
urban structure of metropolitan Beijing. The results also reveal a centralized urban
pattern, consistent with the finding of the previous chapters. Subdistricts within the
central city are generally job-rich and peripheral subdistricts tend to have population
surpluses. We assume the subdistricts with an E/P ratio within the range of 0.9 to 1.1 as
balanced areas. It turns out that these balanced areas are mainly located in the suburbs
adjacent to the central city, corresponding to the emerged near suburban subcenters.
Although the subcenter subdistricts in the outer suburbs are not shown to be as strongly
balanced as those in the near suburbs, they are still more balanced than surrounding non-
center areas in general. These results have supported the idea of Giuliano (1991) that

151
jobs-housing balance occurs as part of the urban development process. With the rapid
urban growth of Beijing, the decentralization of both population and employment has
expanded the scale of the central agglomeration, and both people and jobs have relocated
into the near suburbs where infrastructure is generally well developed and land is much
cheaper and more available, which tends to make those areas balanced with people and
jobs. Meanwhile, the central areas are shown to be unbalanced with more jobs, and the
peripheral suburbs are also unbalanced with more people. This reflects the fact that jobs
are not generally decentralized per se in the Beijing metropolitan area, and employers still
prefer the locations within the central city for the agglomeration economy benefits
inherent to the areas. The outer suburbs generally have less job opportunities, and jobs
tend to concentrate into the subcenters to take advantage of agglomeration economies.

152
Figure 5.2 Employment-Population Ratios and Urban Subcenters

153
Table 5.2 The Distribution of E/P Ratios at Different Levels
Percentage
E/P Ratios
Number of
Subdistricts Population Employment Area
E/P <= 0.7 64 17.95 9.51 65.58
0.7 < E/P <= 0.9 42 11.18 8.72 19.02
0.9 < E/P <= 1.1 25 11.17 8.13 7.52
1.1 < E/P <= 1.3 59 31.57 39.85 5.91
E/P > 1.3 42 28.12 33.78 1.98

Table 5.2 summarizes the distribution of the E/P ratios at different levels. It is
shown that 10.8% of the subdistricts are strongly balanced (E/P ratio is 0.9 to 1.1) and
43.5% of the subdistricts are moderately balanced (E/P ratio is 0.7 to 0.9 or 1.1 to 1.3),
which account for 7.52% and 24.93% of the total area of metropolitan Beijing
respectively. It is interesting to note that the most job-rich subdistricts (E/P ratio is above
1.3) account for both 33.78% of the total employment and 28.12% of the total population
of the Beijing metropolitan area, while the most population-rich subdistricts (E/P ratio is
below 0.7) account for only 9.51% of the total employment and 17.95% of the total
population of the Beijing metropolitan area. Comparing the population distribution with
the employment distribution at different E/P ratio levels, we can find that population is
more evenly distributed across different levels. This implies that people are more
dispersed in distribution than jobs, and the jobs-housing imbalance is mainly attributed to
the strong concentration of employment within the central areas, which is also revealed
by the fact that the job-rich subdistricts (E/P ratio is above 1.1) only account for 7.89% of
the total area of metropolitan Beijing.

154
Figure 5.3 Loess Curves of Urban Densities of People and Jobs

Another way to examine the relationship between the spatial distributions of
population and employment is to observe the density curves. The urban densities of
people and jobs are plotted against the distances to the city center, and the Loess curves
are fitted in Figure 5.3 (The smoothing parameter is selected as 0.2). As the absolute
numbers of population and employment are not comparable to each other, we calculate
the density as the ratio of the percentage of the population or employment of each
subdistrict to the percentage of the area of the subdistrict. A higher density of jobs and a
lower density of people are observed in and near the CBD, which reflects the center
preference of employers, and suggests that employment is more concentrated than
population in and near the CBD. Meanwhile, the density of jobs peaks at the CBD and
then decreases further as we go farther from the CBD, while the density of people peaks
at the location near the CBD and decreases in a smoother way from the center to the

155
periphery. Some local peak is also observed along with the density curve of population
that might reflect the subcentering of people in the suburbs. Overall, the urban pattern of
the Beijing metropolitan area is still highly centralized by nature, especially from the
perspective of the employment distribution, although the decentralization of both people
and jobs is evident, and jobs-housing balance occurs manly in the near suburbs with the
expansion of the central city, which is clearly a result of the urban development process.

5.4 COMMUTING PATTERNS AND URBAN STRUCTURE
The previous analysis above investigated the spatial structure of the whole
metropolitan area of Beijing, and has shown the dominance of the central agglomeration
besides the emergence of the suburban subcenters. The spatial extent of the central
agglomeration in the Beijing metropolitan area is shown to be quite large, so there is no
reason to believe it is monocentric by nature. Therefore, it is interesting to further
investigate the spatial structure within the central city. To examine urban structure at
different spatial scales helps better understand the nature of the urban form of
metropolitan Beijing. In this section, we will restrict our analysis within the survey area,
the spatial extent of which is approximately equivalent to that of the central city. And we
will analyze the spatial structure based on both the distributions of population and
employment as well as the commuting pattern.
In Chapters 3 and 4, urban spatial structure was identified via the locations of
employees and residents. Urban spatial structure can also be identified through examining

156
urban commuting patterns. For example, if we know the inflows to all destinations
(workplaces) and the outflows from all origins (residences), then the O-D pattern can be
derived to show where most people and jobs are located, reflecting the spatial
relationship between the distributions of population and employment. Moreover,
analyzing the commuting pattern will provide more direct evidences for testing the jobs-
housing balance hypothesis in central Beijing. One simple way to examine the
commuting pattern is to look at the work trip distribution between the central and the
peripheral areas. Table 5.3 shows the commuting distribution pattern in the survey area.
We divide the survey area into three zones from the urban core to the periphery and the
work trip flows between these zones are summarized in the table.
Table 5.3 Commuting Distribution Pattern in the Survey Area
Destination
Origin
Zone 1
(Core)
Zone 2
(Intermediate)
Zone 3
(Periphery)
Total
Zone 1 (Core)
671
(12.84%)
752
(14.39%)
104
(1.99%)
1527
(29.22%)
Zone 2 (Intermediate)
433
(8.29%)
1875
(35.89%)
270
(5.17%)
2578
(49.34%)
Zone 3 (Periphery)
213
(4.08%)
653
(12.50%)
254
(4.86%)
1120
(21.44%)
Total
1317
(25.21%)
3280
(62.78%)
628
(12.02%)
5225
(100.00%)
Note: Zone 1 comprises the core areas within the inner city; Zone 2 comprises the
intermediate areas outside the inner city while inside the 5
th
ring road; Zone 3 comprises
the peripheral areas outside the 5
th
ring road in the survey area. Please refer to Figure 5.1
for the spatial extent of each zone.

With the suburbanization of both population and employment in metropolitan
Beijing, the urban core (or the inner city) is no longer dominant any more, with only less

157
than 30% of the total work trips connected to the core area. While almost and over one
half of the total outgoing and incoming work trips are connected to the central area
outside the urban core. It is interesting to note that the outgoing work trips are distributed
more evenly across the three zones than the incoming work trips which tend to
concentrate in the intermediate zone. This reflects the fact that people are more
decentralized than jobs in the Beijing metropolitan area, and in other words, people seem
to have moved relatively farther away from the core than jobs. Moreover, a higher
percentage of the intra-zonal and a lower percentage of the inter-zonal commuting flows
are shown in the table, with over one half of the total work trips beginning and ending in
the same zones. Among the three zones, the intermediate zone that is the central area
outside the urban core is the most self-contained, with 35.89% of the total work trips in
the survey area beginning and ending in the zone. The inter-zonal commuting is
obviously dominated by the work trips from the core or the periphery to the intermediate
zone, and the work trips between Zone 1 and Zone 2 and between Zone 2 and Zone 3 are
the most common inter-zonal commutes, while the longer commutes between Zone 1 and
Zone 3 are relatively rare. Clearly, the commuting distribution pattern shows that the
most dominant area in commuting as both the origin and the destination is not the core
area but the central area outside the core. This implies that both people and jobs have
moved outside the inner city with the suburbanization in metropolitan Beijing, but they
tend to reconcentrate in the near suburbs adjacent to the urban core.

158
Figure 5.4 Spatial Distributions of Commuting Inflows and Outflows

159
Figure 5.5 Contour Maps for the Population and Employment Density Surfaces

(a) Population Density Surface

(b) Employment Density Surface

160
To further investigate the spatial structure within the central city, we generate the
thematic maps that depict the distributions of the work trip inflows to all destinations and
the work trip outflows from all origins, and the contour maps that depict the density
surfaces of population and employment in the survey area. The thematic maps are
generated based on the O-D information contained in the survey dataset, and the contour
maps are drawn using the Surfer software package based on the subdistrict-level
aggregate population and employment density data. Several interpolation methods can be
used in Surfer to create the density surface. We choose the Kriging method, which is one
of the most commonly used geostatistical techniques to generate interpolated surfaces.
Figure 5.4 shows the commuting O-D pattern in the survey area. As the quota
sampling method has been used to create the survey sample, the distribution pattern of
the origins should conform to the overall distribution pattern of the population. It is
shown that the origins where the work trips begin are more scattered in the suburbs, while
the destinations where the work trips end are more concentrated around the urban core.
Three job centers can be identified in the western, eastern and northwestern suburbs
based on the distribution pattern of the destinations. Similar patterns are also shown on
the contour maps (Figure 5.5). Analyzing the distributions of population and employment,
we find that people are generally more dispersed than jobs and tend to form several
centers in the near suburbs around the urban core. This is partly a result of the city plan
that promoted the development of the scattered residential groups in the near suburbs.
Furthermore, the three job centers shown in Figure 5.4 are also evident on the contour

161
map as the peaks of the employment density surface. Overall, the urban structures
revealed by the commuting pattern and the distributions of population and employment
are quite similar. People and jobs are shown to tend to concentrate in the suburbs adjacent
to the urban core and form both residential and employment centers within the central
city. And population is shown to be more decentralized than employment, and jobs tend
to be located more near the urban core. The overall distribution patterns of the origins and
destinations are pretty matched, which suggests that the spatial separation between the
residence and workplace may not be so severe in central Beijing. However, the jobs-
housing balance can be more directly tested through examining the major external (inter-
subdistrict) work trip flows, which show that longer commutes do exist (Figure 5.6).

162
Figure 5.6 Major External Commuting Flows and Self-Containment of Subdistricts

Figure 5.6 depicts the external commuting flows with volumes greater than 10
from the origins to the destinations. It is evident that some long distance commutes do
occur, a large portion of which are from the northwestern suburbs to the eastern job-
center areas. The figure also maps the self-containment of each subdistrict that is
measured by the independence index calculated as the number of internal (within
subdistrict) work trips divided by the sum of in and out (external) work trips for each
subdistrict (Cervero, 1996). It is shown that suburban subdistricts are more likely to be

163
self-contained with more residents working locally. However, the independence indexes
of the subdistricts are generally small with the majority of them below 0.1, as the internal
work trips only account for 9.5% of the total work trips in our sample. That means most
of the residents in the central city commute across subdistricts to work. Figure 5.7 shows
the frequency distribution of the work trips with the distances between the subdistricts of
residence and workplace. It is shown that there are over 40% of the work trips with the
O-D distances above the average (9.5 kilometers). This suggests long commutes are
possible even when the spatial matches between residences (origins) and jobs
(destinations) exist, which may result from the incompatibility of the mix of jobs and
housing. Therefore, the relationship between commuting and the distributions of
population and employment is ambiguous, and we will further examine it in more detail
in the following section.

164
Figure 5.7 Cumulative Percentage of Commuting Flows with Distances between
Subdistricts of Residence and Workplace
0
20
40
60
80
100
0 5 10 15 20 25 30 35 40 45
Distance between Subdistricts of Residence and Workplace (km)
Cumulative Percent of Work Trip Flow

5.5 JOBS-HOUSING BALANCE AND COMMUTING TIME
5.5.1 E/P Ratio and Mean Commuting Time
The relationship between jobs-housing balance and commuting time is possible as
the concept of jobs-housing balance implicitly assumes that residents would work as
close to their homes as possible and workers would live as close to their jobs as possible.
Therefore, residents that live in a job-rich area tend to have shorter commuting times as
they should be able to find jobs locally, and meanwhile workers that work in a housing-
rich area also tend to have shorter commuting times as they are more likely to live nearby.
So the hypothesized relationship between jobs-housing balance and commuting time is

165
that mean commuting times by residents should be negatively related to jobs-housing
ratios (that are E/P ratios in our case) while mean commuting times by workers should be
positively related to jobs-housing ratios.
Figure 5.8 E/P Ratio and Mean Commuting Time
R
2
= 0.0142
R
2
= 0.0003
0
20
40
60
80
100
120
0.5 0.7 0.9 1.1 1.3 1.5
E/P Ratio
Mean Commute Time (Minutes)
by Residents
by Workers
Regression Line (by Residents)
Regression Line (by Workers)

To visualize the relationship between mean commuting times and E/P ratios, we
first calculate the mean commuting time of residents or workers in each subdistrict, and
then plot them against the E/P ratios of the subdistricts. Figure 5.8 shows that the
relationship between commuting times and E/P ratios is quite weak in our case. However,
the regression lines give the evidence that as the E/P ratio increases, the mean commuting
time of workers increases slightly while that of residents decreases. In other words, the
job-rich subdistricts have longer commuting times for workers employed there and

166
shorter commuting times for residents living there. On the contrary, the housing-rich
subdistricts have longer commuting times for local residents while shorter commuting
times for local workers. So the relationship is consistent with the hypothesis, but is not
significant as indicated by the low values of R
2
.

5.5.2 Model Specifications
To examine further the relationship between jobs-housing balance and commuting
time, and the determinants of work trip duration in the survey area, we build a variety of
regression models. The socioeconomic attributes of commuters are included as control
variables in the models. Previous studies have shown that personal and household
characteristics affect work-trip duration, including gender, age, education, occupational
status, income, household size and composition, etc. (Punpuing, 1993; Shen, 2000; Wang,
2001; Vandersmissen et al., 2003). These characteristics also affect residential location
and job location. Therefore, it is appropriate to hold constant their influences on
commuting time to measure the specific effect of land use pattern on commuting.
In the U.S, the literature on the relations between commuting and socioeconomic
factors is overwhelming. For example, a large amount of studies have shown that women
and minority workers may have different commuting behavior (Wang, 2001). Empirical
evidences that show women tend to engage in shorter work commutes than men are
persistent in the literature (Turner and Niemeier, 1997). This gender differential in
commuting may be resulted from different labor-force characteristics of male and female

167
workers as well as gender differences in the household division of labor. For example,
women tend to earn less, have different educational backgrounds, work in different
occupations, and have greater household and child-care responsibilities. The relationship
of age with commuting time is not as widely tested as that of gender in the literature.
Punpuing (1993) suggested that younger people tend to have shorter commutes as they
have fewer costs of moving and therefore can adjust their place of residence close to their
place of work. Levinson (1998) found that the relationship between age and work trip
duration is nonlinear, with younger and older people having shorter commutes than
middle-aged people, however, Vandersmissen et al. (2003) found the nonlinear effect is
quite weak in their study. The effect of educational attainment or occupational status on
commuting is based on the argument related to levels of skills and job search
(Vandersmissen et al., 2003). It has been shown that more highly-educated people or
those with higher occupational status are more likely to have longer commutes, as they
tend to conduct more spatially extensive job searches (Simpson, 1987). Household size
and composition also matter. It has been suggested that like young people, single-person
households face less constraints of family factors in their residential location choices, so
they are more likely to reside close to their workplace and have shorter commutes
(Punpuing, 1993; Vandersmissen et al., 2003). While multiple-workers households tend
to have longer commutes partly due to the difficulty to optimize individuals’ commutes
between different workplaces (Wang, 2001), and the commuting patterns of married-
couples households are also affected by the work status of spouses and the numbers of

168
their school-age children (Punpuing, 1993). At last, household income affects residential
location choice and thus commuting behavior (Wang, 2001). The standard urban model
has suggested that lower housing prices and higher wages are associated with longer
commutes. All these evidences in the literature are mostly drawn from experiences in
developed countries, while we will test whether these relationships also hold in a
developing and transition context through our study.
Besides the personal and household characteristic variables, we also include three
other variables related to the housing characteristics of commuters in our models. The
first is housing area. The suburbanization theory has suggested that the preference of
people for lower-density living environments and more living space has relocated them to
the suburbs, resulting in longer commutes. The second variable indicates home
ownership status. It has been suggested that home ownership status may affect relocation
costs and thus renters are more likely to change their residence to minimize commuting
when they change their jobs (Wang, 2001). The third variable is peculiar in the Chinese
city context, which indicates whether housing is provided by employers. This variable is
used to test the effect of the work-unit system on urban commuting. The work-unit
system tends to organize housing and production activities together in the work-unit
compound, which induces shorter commutes in view of the non-separation between
workplace and residence. Although the system has broken down since the reforms in
Chinese cities, it is interesting to see whether it still has remaining effects on urban
commuting.

169
Besides the characteristics of commuters, mode of transportation has the most
direct and evident impact on commuting time. Previous studies have shown that going to
work by car takes, on average, much less time than taking the public transit
(Vandersmissen et al., 2003). So in our models, we include dummy variables to indicate
different transport modes taken by commuters. The E/P ratio is used to measure urban
structure and is the factor of most interest. In some regressions, we replace the variable of
E/P ratio with the variable of number of bus-stops per capita in each subdistrict. Although
various commuting modes are used by workers in our sample, going to work by bus is
still the most dominant, which accounts for 38.6% of the total work trips. So the number
of bus-stops per capita here is used to proxy the mobility of the area. The purpose of the
substitution is to find whether it is mobility or accessibility to jobs that matters to
commuting time, which sheds light on the planning argument whether land use planning
is effective to solve transportation problems or these problems are better addressed in a
more direct way.

5.5.3 Regression Results
Table 5.4 reports the regression results. From a comparison of all the models, it is
obvious that the accessibility to jobs measured by the employment-population ratio (E/P
ratio) has no explanatory power to the variation in commuting time. The mobility
variable measured by the number of bus-stops per capita is significant as a factor to affect
commuting time, but its explanatory power is also quite low. The variation in commuting

170
time is mostly explained by travel modes, while the socioeconomic variables also have a
modest explanatory power. The results suggest that commuting duration is more directly
related to transportation modes and the mobility of the area.
The results show that the transportation mode variables explain about 25% of the
variation in commuting time. And once after controlling for transportation mode choices,
most of the socioeconomic variables and the mobility variable are not significant any
more. This is because transportation mode choices are highly related to the
socioeconomic status of commuters and the mobility of the area, once they are all
included in the model, the effects of the socioeconomic factors and the mobility are then
dominated by the transportation mode variables. The standardized coefficients of the
transportation mode dummies also provide the evidence that besides commuting by
walking or bicycling, commuting by cars takes much less time than by such public
transits as buses or subways. And commuting by subways or light-rails is also faster than
by buses. This is consistent with the evidences from the previous studies.

171
Table 5.4 Regression Results
Model 1 Model 2 Model 3
Std.
Beta
t Sig.
Std.
Beta
t Sig.
Std.
Beta
t Sig.
Accessibility to jobs (E/P ratio) 0.005 0.358 0.720 -0.003 -0.242 0.809 -0.002 -0.172 0.864
Mobility
(number of bus-stops per capita)

Transportation mode:
Bus

0.530 39.019 0.000

Car 0.187 14.1260.000
Subway 0.359 28.1240.000
Personal characteristics:
Age 1 (<= 30 years old)

0.058 3.705 0.000
Age 2 (>= 50 years old) -0.051 -3.429 0.001
Female 0.0201.4030.161
Education (college & graduate =1) 0.083 5.470 0.000
Occupation (management =1) -0.005 -0.337 0.736
Household characteristics:
Income

-0.033 -2.153 0.031
Household size (range =1~5) -0.008 -0.540 0.589
Housing characteristics:
Housing area

0.016 1.022 0.307
Housing status
(housing provided by employers =1)

-0.047 -3.028 0.002
Renters -0.093-5.8990.000
Adjusted R
2
0.000 0.259 0.025

172
Table 5.4, Continued
Model 4 Model 5 Model 6
Std.
Beta
t Sig.
Std.
Beta
t Sig.
Std.
Beta
t Sig.
Accessibility to jobs (E/P ratio) -0.005 -0.383 0.702 -0.016 -0.949 0.343
Mobility
(number of bus-stops per capita)

-0.039 -2.392 0.017 -0.031 -2.225 0.026
Transportation mode:
Bus 0.524 37.701 0.000

Car 0.18813.2840.000
Subway 0.35427.3060.000
Personal characteristics:
Age 1 (<= 30 years old) 0.021 1.551 0.121

Age 2 (>= 50 years old) 0.001 0.060 0.952
Female 0.0131.0460.296
Education (college & graduate =1) 0.0191.3950.163
Occupation (management =1) 0.0020.1550.876
Household characteristics:
Income -0.046-3.3990.001

Household size (range =1~5) 0.000 -0.023 0.981
Housing characteristics:
Housing area -0.001 -0.105 0.916

Housing status
(housing provided by employers =1)
-0.031 -2.276 0.023

Renters -0.072-5.2610.000
Adjusted R
2
0.264 0.001 0.001

173
Table 5.4, Continued
Model 7 Model 8 Model 9
Std.
Beta
t Sig.
Std.
Beta
t Sig.
Std.
Beta
t Sig.
Accessibility to jobs (E/P ratio) -0.024 -1.494 0.135
Mobility
(number of bus-stops per capita)
-0.013 -1.057 0.291 -0.029 -2.100 0.036 -0.042 -2.571 0.010
Transportation mode:
Bus 0.529 38.954 0.000

Car 0.18614.1040.000
Subway 0.35928.1100.000
Personal characteristics:
Age 1 (<= 30 years old)

0.057 3.639 0.000 0.058 3.698 0.000
Age 2 (>= 50 years old) -0.049 -3.349 0.001 -0.049 -3.356 0.001
Female 0.019 1.3960.1630.0191.3950.163
Education (college & graduate =1) 0.084 5.499 0.000 0.084 5.520 0.000
Occupation (management =1) -0.006 -0.377 0.707 -0.005 -0.344 0.731
Household characteristics:
Income

-0.034 -2.211 0.027 -0.034 -2.197 0.028
Household size (range =1~5) -0.009 -0.594 0.552 -0.009 -0.569 0.569
Housing characteristics:
Housing area

0.017 1.085 0.278 0.018 1.149 0.251
Housing status
(housing provided by employers =1)

-0.047 -3.067 0.002 -0.046 -2.979 0.003
Renters -0.092 -5.8780.000-0.093-5.9120.000
Adjusted R
2
0.259 0.026 0.026

174
Further insights into commuting behaviors can be gained by examining the
coefficients of the socioeconomic variables. Overall, they are highly consistent through
the models. First, age is highly significant in explaining the variation in commuting time.
Contrary to the evidence that younger people tend to have shorter commutes in the
Western literature, we find that younger people are associated with longer commutes in
our sample. This may be true as in Beijing younger people tend to live farther away from
the urban core in the newly-developed suburban residential areas where housing is much
cheaper and more available while jobs are mostly located around the urban core, so this
may result in longer commutes. In addition, older people tend to live nearby their
workplaces, resulting in shorter commutes, simply because most of them generally work
and live in the work-unit compound.
Second, although the empirical evidence that women tend to commute shorter
distances and times than men is quite robust in the Western literature, we do not find that
evidence in our case, as the coefficient of the female dummy is not significant. The result
may be attributed to the fact that in China the labor-force characteristics of women are
not quite different from men, and the work participation rates of women are really high,
which is also validated by the fact that approximately half of male and female workers
are selected randomly in our sample. Besides, two-earner households are quite common
in China, and although women do tend to have more household and child-care
responsibilities, they also spend as much time in work activities as men. So the household
responsibility hypothesis may not hold in our context. Furthermore, the household size

175
variable is also not significant in our models. The effect of household size and
composition on commuting duration is not clear in our case, and tends to be taken over
by other socioeconomic factors such as age and income in our models.
Third, educational attainment has a significant effect on commuting time and the
results conform to the common evidence that highly educated people tend to spend more
time commuting. This may reflect that individuals with a bachelor’s or graduate degree
tend to have more extensive social network and job search. Moreover, the effect of
occupational status on commuting is not significant, and may be taken over by the
education variable, as the two are highly correlated.
Fourth, higher household income is associated with shorter commutes in our
models. This relationship is not consistent with the prediction of the standard urban
model. This result may be due to that household income also affects transportation mode
choices, and higher income households tend to commute by cars, resulting in shorter
commuting times.
Fifth, the effects of housing characteristic variables on commuting time are pretty
much as would be expected, except that housing area is not significant in explaining the
variation in commuting time. It is interesting to note that although the work-unit system
has broken down since the land and housing reforms, its effect on urban commuting
remains strong in Beijing, and this suggests that urban structure is determined by the
historical urban development process and because of the historical path-dependence of
urban development, the old structure tends to have a longstanding impact on urban

176
activities. In addition, home ownership status also matters, and it is shown that renters do
tend to have shorter commutes because they can easily adjust their place of residence
close to their workplace.
Overall, some implications are suggested by the empirical findings. The results
imply that balancing people and jobs by configuring land use patterns seems not quite
relevant to shortening commuting durations in our case. But this is more as a conjecture
than a conclusive statement, since our study is only based on a single survey that may not
be representative. Besides, more specific investigations into the land use planning process
and its relationship with urban commuting are still needed to draw the conclusions. The
results also reveal different commuting behaviors in Beijing from those findings based on
the experience of North American cities, and the peculiarities can be by and large
explained with the specific context of Chinese cities. It also shows that the work-unit
system does have an impact on the urban structure of Beijing and results in shorter
commutes. However, the relatively low explanatory power of the models also implies
some other factors may be more important than those included in explaining the variation
in commuting time, which need to be examined further more carefully.

5.6 SUMMARY
Following the previous chapters, this chapter further investigated the relationship
between the distributions of people and jobs in the Beijing metropolitan area and
searched for the evidence of jobs-housing balance in Beijing. The relationship between

177
the distributions of population and employment was studied further through the
employment-population ratio, which approximately measured jobs-housing balance in the
Beijing metropolitan area. The distribution pattern of the E/P ratio has revealed that the
urban structure of Beijing is by and large centralized, with the central city job-rich while
the periphery population-rich. The suburban areas adjacent to the central areas are the
most balanced with people and jobs, corresponding to the emerged subcenters in the near
suburbs. The result has largely conformed to the argument that jobs-housing balance
occurs as part of the urban development process, and the decentralization of both
population and employment in the Beijing metropolitan area from the inner city to the
near suburbs has induced jobs-housing balance in the near suburbs. Employment has
been shown to be more concentrated in and near the central areas than population in
metropolitan Beijing, resulting in the jobs-housing imbalance in both the central areas
and the periphery.
An analysis of urban spatial structure at a smaller spatial scale based on both the
distributions of population and employment and the commuting pattern generated using
the survey data has revealed the more detailed spatial structure within the central
agglomeration of metropolitan Beijing. It has been shown that with the suburbanization
in the Beijing metropolitan area, both people and jobs have moved outside the inner city
but they tend to reconcentrate in the near suburbs adjacent to the core area, with most of
the work trips connected to those suburban areas. Besides, people seem to be more
decentralized than jobs and jobs tend to locate more near the urban core. Both residential

178
and job centers are evident within the central agglomeration, mainly located around the
core area. Further analysis on commuting flows has revealed that long commutes are
common in our sample, even though the overall distribution patterns of residences and
jobs seem to be pretty matched.
At last, we formally tested the relationship between jobs-housing balance and
urban commuting through the regression analysis. Empirical findings are broadly
consistent with the longstanding criticism of the Chinese comprehensive planning process
that places main emphasis on land-use control and neglects the importance of the
integration of land use and transportation. The results have shown that balancing people
and jobs by configuring land use patterns seems not quite relevant to shortening
commuting durations in our case, so it is more promising to integrate transportation
planning and land use planning to address the transportation problems in Beijing.

179
CHAPTER 6.
CONCLUSION

This dissertation studies the spatial distributions of population and employment in
one of the largest Chinese cities, Beijing, during the post-reform era. Complementary to
the existing empirical literature of urban spatial evolution, our findings provide evidence
regarding the pattern and process of urban decentralization and restructuring from a
developing and transitional economy context, and offer further understanding of the
spatial organization of contemporary urban areas that departs from the North American or
European experience. Our study focuses on several questions regarding the spatial
structure of metropolitan Beijing that have largely not been addressed in the literature so
far, and aims to understand better the changing nature of the urban form of Beijing during
the post-reform era, and provide useful implications for the urban planning of Beijing.
The major findings of our study are summarized as followed. First, similar to the
findings of previous studies, our study also finds a trend toward decentralization of both
population and employment in the Beijing metropolitan area in the post-reform era. The
spatial pattern of today’s Beijing is becoming more alike to large Western cities, with the
compact urban form in the pre-reform era replaced by a more dispersed and polycentric
spatial pattern. The overall trend of urban expansion and decentralization characterized
by the relocation of urban residents and industries from the inner city to the suburbs as

180
well as the rapid suburban growth is also broadly analogous to the suburbanization in the
West since the World War II.
Different from the suburbanization in the U.S., however, the process of
decentralization in the Beijing metropolitan area started with the dispersion of both
population and employment simultaneously. Therefore, not like in the U.S. where
suburbanization was first driven by the residential relocation, the suburbanization of
Beijing has witnessed the redistribution of both residents and industries since the 1980s.
Although both people and jobs have decentralized in the Beijing metropolitan area, jobs
tend to more concentrate in the central city, and employment has been shown to be less
decentralized than population. Overall, the suburbanization of Beijing is still in its initial
stage, with people and manufacturing activity moving out of the inner city. This was first
driven by the government-led redevelopment of the inner city that relocated residents and
factories to the suburbs, and further by the large scale development of industrial parks,
development zones and residential communities in the near suburbs. Meanwhile, the
dispersion of services and trade related industries is not evident in the Beijing
metropolitan area, and it has been shown that higher-order functions such as producer
services and arts, sports and broadcasting activities remain highly concentrated in the
CBD to take advantage of the agglomeration economies inherent to that location.
Although, as argued by Feng et al. (2008), the differences between the suburbanization of
Beijing and other Western cities may be largely attributed to the different stages of
suburbanization rather than the dichotomy of market and planned economies, the driving

181
forces and the process involved still need be understood with reference to the peculiar
Chinese context, and the similar factors that caused the suburbanization in the West have
taken their effects on the suburbanization of Beijing in a totally different context.
Compared with the decentralization of large Western metropolitan areas, the
extent of the decentralization of metropolitan Beijing is quite limited. We show that both
people and jobs that moved out of the inner city tend to re-concentrate in the near suburbs
adjacent to the central area instead of dispersing throughout the metropolitan area. The
rapid growth of the near suburbs has expedited the expansion of the central city, with a
larger central agglomeration emerged dominating the whole metropolitan area. In this
broad sense, the spatial pattern of the Beijing metropolitan area is still highly centralized.
From the perspective of employment distribution, the centralized feature of the urban
structure is more phenomenal. The central agglomeration contains over 70% of
employment in the Beijing metropolitan area, and plays a key role in describing the
location of the region’s jobs. And its spatial extent is quite large. Therefore, although the
decentralization has occurred in the Beijing metropolitan area, its spatial structure can be
largely characterized by monocentricity.
Even though Beijing has been largely regarded as a monocentric city, our study
provides the evidence that significant population and employment subcenters do have
emerged in the suburbs of Beijing. It has been shown that population subcenters have
began to emerge in the near suburbs of Beijing since 1982, and the number of subcenters
has increased over time. The existence of the population subcenters has been shown to be

182
persistent through the years, and the conurbation process with the subcenters first
emerged near the central area further incorporated into the central agglomeration with its
continual spatial expansion is also evident. Suburban employment subcenters have also
been identified in the Beijing metropolitan area, however, the number of subcenters is far
less than that in some American metropolitan areas of equivalent size, and only about 6%
of total jobs have clustered into the subcenters. Overall, the subcentering phenomenon in
the Beijing metropolitan area is far less advanced than in large American metropolitan
areas.
The emergence of subcenters in the suburbs of Beijing can be explained by the
planning effort to promote the dispersed and scattering spatial development in the city.
For instance, the population subcenters identified are mostly associated with the planning
and development of several peripheral residential groups in the near suburbs and the
satellite towns in the outer suburbs. And the formation of employment subcenters can
neither be solely attributed to pure agglomeration economies, while the development of
large-scale suburban industrial parks caused the formation of several subcenters in the
near suburbs, and the satellite towns that were planned and developed as independent
centers from the central city formed the subcenters with more comprehensive functions in
the outer suburbs. Generally, the pattern of the subcenters in the Beijing metropolitan
area is highly adherent to the development scheme of the city, with subcenters emerged
mainly along the ring and radial transport axes, especially along the east-west corridor of
the city, and mostly in the peripheral industrial parks and residential groups or the

183
satellite towns. So the polycentricity emerged in the Beijing metropolitan area is very
different by nature from that observed in Western cities, and has different origins, which
can be referred to as the so called “planned polycentricity”.
An interesting finding of our study is the similar distribution patterns of both
population and employment in the Beijing metropolitan area, with the coincidence of the
population and employment subcenters in space. However, it is not very surprising if
considering the spatial distributions of both population and employment in the Beijing
metropolitan area are highly associated with the planned development scheme of the city.
The relationship between the distributions of people and jobs in the Beijing metropolitan
area is then studied through the employment-population ratio, which approximately
measures jobs-housing balance in the Beijing metropolitan area. The distribution pattern
of the E/P ratio has also revealed that the urban structure of Beijing is by and large
centralized, with the central city job-rich while the periphery population-rich. The
suburban areas adjacent to the central areas are the most balanced with people and jobs,
corresponding to the emerged subcenters in the near suburbs. The result might be used to
argue that jobs-housing balance occurs as part of the urban development process, and the
decentralization of both population and employment in the Beijing metropolitan area
from the inner city to the near suburbs has induced jobs-housing balance in the near
suburbs. Employment has been shown to be more concentrated in and near the central
areas than population in metropolitan Beijing, resulting in the jobs-housing imbalance in
both the central areas and the periphery.

184
At last, the relationship between jobs-housing balance and urban commuting is
tested through regression analysis. The results show that balancing people and jobs by
configuring land use patterns seems not quite relevant to shortening commuting durations
in our case, so this may suggest that it is more promising to integrate transportation
planning and land use planning to address the transportation problems in Beijing. But,
since our study is only based on a single survey that may not be representative, more
detailed investigations into the land use planning process and its relationship with urban
commuting are still needed to draw the specific conclusions. This calls for further
empirical work in the future. Our results also reveal different commuting behaviors in
Beijing from those findings based on the experience of North American cities, and the
peculiarities can be by and large explained with the specific context of Chinese cities.
On the whole, this dissertation is just a start of the detailed empirical investigation
to the spatial structure of the Beijing metropolitan area, and established some stylized
facts that further empirical studies can be based on. This study can be further extended
along several aspects in the future. First, the detailed spatial structure within the central
agglomeration or the central city of Beijing should be examined more closely. Second,
the impacts of the subcenters identified in this study on the land use pattern, housing
development and urban transportation organization in the Beijing metropolitan area can
be explored further. Third, disaggregate population and employment data across
occupations of worker residents, employment sectors and other socio-economic attributes

185
can be used to examine the socio-spatial differentiation and stratification in the Beijing
metropolitan area.

186
REFERENCES

Ajo, R. (1965): “On the structure of population in London’s field,” Acta Geographica, 18,
1-17.
Alonso, W. (1964): Location and Land Use: Toward a General Theory of Land Rent.
Cambridge, MA: Harvard University Press.
Alperovich, G. (1995): “The effectiveness of spline urban density functions: An
empirical investigation,” Urban Studies, 32 (9), 1537-1548.
Anas, A., R. Arnott and K.A. Small (1998): “Urban spatial structure,” Journal of
Economic Literature, 36 (3), 1426-1464.
Anderson, J. (1982): “Cubic spline urban density functions,” Journal of Urban Economics,
12 (2), 155-167.
Anderson, J. (1985): “The changing structure of a city: temporal changes in cubic spline
urban density patterns,” Journal of Regional Science, 25 (3), 413-425.
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.
Anselin, L. (1995): “Local indicators of spatial association – LISA,” Geographical
Analysis, 27 (2), 93-115.
Anselin, L. (1996): “The Moran scatterplot as an ESDA tool to assess local instability in
spatial association,” In: M. Fisher, H.J. Scholten, and D. Unwin (eds), Spatial
Analytical Perspectives on GIS, London: Taylor & Francis.
Anselin, L. (2003): GeoDa 0.9 User’s Guide, Urbana-Champaign, IL: Spatial Analysis
Laboratory, University of Illinois.
Baumont, C., C. Ertur, and J. Le Gallo (2004): “Spatial analysis of employment and
population density: The case of the agglomeration of Dijon 1999,” Geographical
Analysis, 36 (2), 146-176.
Beckmann, M. (1976): “Spatial equilibrium in a dispersed city,” In: Y.Y. Papageorgiou
(eds.), Mathematical Land Use Theory, Lexington: Lexington Books.

187
Bertaud, A., and B. Renaud (1997): “Socialist cities without land markets,” Journal of
Urban Economics, 41 (1), 137-151.
Boots, B.N., and A. Getis (1988): Point Pattern Analysis, Newbury Park, CA: Sage
Publications
Cervero, R. (1989): “Jobs-housing balancing and regional mobility,” Journal of the
American Planning Association, 55 (2), 136-150.
Cervero, R. (1995): “Planned communities, self-containment and commuting: A cross-
national perspective,” Urban Studies, 32 (7), 1135-1161.
Cervero, R. (1996): “Jobs-housing balance revisited: Trends and impacts in the San
Francisco Bay Area,” Journal of the American Planning Association, 62 (4), 492-
510.
Cervero, R., and Kang-Li Wu (1998): “Sub-centering and commuting: Evidence from the
San Francisco Bay area, 1980-90,” Urban Studies, 35 (7), 1059-1076.
Champion, A.G. (1976): “Evolving patterns of population distribution in England and
Wales, 1951-71,” Transactions of the Institute of British Geographers, 1 (4), 401-
420.
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.
Clark, C. (1951): “Urban population densities,” Journal of Royal Statistical Society, 114,
490-496.
Clark, W.A.V., and M. Kuijpers-Linde (1994): “Commuting in restructuring urban
regions,” Urban Studies, 31 (3), 465-483.
Cleveland, W.S. (1979): “Robust locally weighted regression and smoothing
scatterplots,” Journal of the American Statistical Association, 74, 829-836.
Cleveland, W.S., and E. Grosse (1991): “Computational methods for local regression,”
Statistics and Computing, 1 (1), 47-62.
Cleveland, W.S., S.J. Devlin, and E. Grosse (1988): “Regression by local fitting:
Methods, properties, and computational algorithms,” Journal of Econometrics, 37
(1), 87-114.

188
Coffey, W.J. and R.G. Shearmur (2001): “The identification of employment centers in
Canadian metropolitan areas: The example of Montreal, 1996,” Canadian
Geographer, 45 (3), 371-386.
Cohen, M.A. (1996): “The hypothesis of urban convergence: Are cities in the North and
South becoming more alike in the age of globalization?,” In: M. Cohen, B. Ruble
and J. Tulchin (eds), Preparing for the Urban Future: Global Pressures and Local
Forces, Princeton, NJ: Woodrow Wilson Centre Press.
Cuthbert, A.L., and W.P. Anderson (2002): “Using spatial statistics to examine the
pattern of urban land development in Halifax-Dartmouth,” Professional
Geographer, 54 (4), 521-532.
De Smith, M.J., M.F. Goodchild, and P.A. Longley (2008): Geospatial Analysis: A
Comprehensive Guide to Principles, Techniques and Software Tools (Second
Edition), Leicester, UK: Matador.
Deng, F., and Y. Huang (2004): “Uneven land reform and urban sprawl in China: the
case of Beijing,” Progress in Planning, 61 (3), 211-236.
Dick, H.W., and P.J. Rimmer (1998): “Beyond the third world city: The new urban
geography of south-east Asia,” Urban Studies, 35 (12), 2303-2322.
Diggle, P.J., and A.G. Chetwynd (1991): “Second-order analysis of spatial clustering for
inhomogeneous populations,” Biometrics, 47 (3), 1155-1163.
Fan, J., and W. Taubmann (2002): “Migrant enclaves in large Chinese cities,” In: J.R.
Logan (ed.), The New Chinese City: Globalization and Market Reform, Oxford,
UK: Blackwell.
Feng, J., and Y. Zhou (2005): “Suburbanization and the changes of urban internal spatial
structure in Hangzhou, China,” Urban Geography, 26 (2), 107-136.
Feng, J., Y. Zhou, and F. Wu (2008): “New trends of suburbanization in Beijing since
1990: From government-led to market-oriented,” Regional Studies, 42 (1), 83-99.
Feser, E.J., and S.H. Sweeney (2000): “A test for the coincident economic and spatial
clustering of business enterprises,” Journal of Geographical Systems, 2 (4), 349-
374.
Frankena, M.W. (1978): “A bias in estimating urban population density functions,”
Journal of Urban Economics, 5 (1), 35-45.

189
Freestone, R., and P. Murphy (1998): “Metropolitan restructuring and suburban
employment centers: Cross-cultural perspectives on the Australian experience,”
Journal of the American Planning Association, 64 (3), 286-297.
Fujita, M., and H. Ogawa (1982): “Multiple equilibria and structural transition of
nonmonocentric urban configurations,” Regional Science and Urban Economics,
12 (2), 161-196.
Fujita, M., Jacques-Francois Thisse, and Y. Zenou (1997): “On the endogenous formation
of secondary employment centers in a city,” Journal of Urban Economics, 41 (3),
337-357.
Fulton, W. (1996): “Are edge cities losing their edge?,” Planning, 62 (5), 4-7.
Garreau, J. (1991): Edge city. New York: Doubleday.
Getis, A. (1983): “Second-order analysis of point patterns: the case of Chicago as a multi-
center urban region,” Professional Geographer, 35 (1), 73-80.
Giuliano, G. (1991): “Is jobs-housing balance a transportation issue?,” Transportation
Research Record, 1305, 305-312.
Giuliano, G., and C. Redfearn (2005): “Not all sprawl: Evolution of employment
concentrations in Los Angeles, 1980-2000,” Paper Presented at ERSA Conference,
Amsterdam, Netherlands.
Giuliano, G., and K. Small (1991): “Subcenters in the Los Angeles region,” Regional
Science and Urban Economics, 21 (2), 163-182.
Giuliano, G., and K.A. Small (1993): “Is the journey to work explained by urban
structure?,” Urban Studies, 30 (9), 1485-1500.
Glaeser, E.L., and M.E. Kahn (2001): “Decentralized employment and the transformation
of the American city,” NBER Working Paper No. 8117.
Glaeser, E.L., and M.E. Kahn (2003): “Sprawl and Urban Growth,” NBER Working
Paper No. 9733.
Gordon, P., A. Kumar, and H.W. Richardson (1989): “Congestion, changing metropolitan
structure, and city size in the United States,” International Regional Science
Review, 12 (1), 45-56.

190
Gordon, P., and H.W. Richardson (1996): “Beyond polycentricity: The dispersed
metropolis, Los Angeles, 1970-1990,” Journal of the American Planning
Association, 62 (3), 289-295.
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 (2), 161-173.
Gordon, P., H.W. Richardson, and M. Jun (1991): “The commuting paradox: evidence
from the top twenty,” Journal of the American Planning Association, 57 (4), 416-
420.
Griffith, D. (1981a): “Modelling urban population density in a multi-centered city,”
Journal of Urban Economics, 9 (3), 298-310.
Griffith, D. (1981b): “Evaluating the transformation from a monocentric to a polycentric
city,” Professional Geographer, 33 (2), 189-196.
Griffith, D., and D.W. Wong (2007): “Modeling population density across major US
cities: a polycentric spatial regression approach,” Journal of Geographical
Systems, 9 (1), 53-75.
Gu, C., and J. Shen (2003): “Transformation of urban socio-spatial structure in socialist
market economies: the case of Beijing,” Habitat International, 27 (1), 107-122.
Han, S. (2004): “Spatial structure of residential property-value distribution in Beijing and
Jakarta,” Environment and Planning A, 36 (7), 1259-1284.
Helsley, R.W., and A.M. Sullivan (1991): “Urban subcenter formation,” Regional
Science and Urban Economics, 21 (2), 255-275.
Henderson, J.V., and A. Mitra (1996): “The new urban landscape: Developers and edge
cities,” Regional Science and Urban Economics, 26 (6), 613-643.
Horner, M.W. (2004): “Spatial dimensions of urban commuting: A review of major
issues and their implications for future geographic research,” The Professional
Geographer, 56 (2), 160-173.
Huang, Y. (2004): “Urban spatial pattern and infrastructure in Beijing,” Land Lines, 16
(4), 1–5.

191
Huang, Y. (2005): “From work-unit compounds to gated communities: housing
inequality and residential segregation in transitional Beijing,” In: L.J.C. Ma, and F.
Wu (eds), Restructuring the Chinese City, London: Routledge.
Hurvich, C.M., J.S. Simonoff, and C.L. Tsai (1998): “Smoothing parameter selection in
nonparametric regression using an improved Akaike information criterion,”
Journal of the Royal Statistical Society B, 60 (2), 271-293.
Jun, Myung-jin, and Seong-kyu Ha (2002): “Evolution of employment centers in Seoul,”
Review of Urban & Regional Development Studies, 14 (2), 117-132.
Kain, J. (1968): “Housing segregation, Negro employment, and metropolitan
decentralization,” Quarterly Journal of Economics, 82 (2), 175-197.
Kau, J.B., and C.F. Lee (1976): “The functional form in estimating the density gradient:
An empirical investigation,” Journal of the American Statistical Association, 71,
326-327.
Kloosterman, R.C., and S. Musterd (2001): “The polycentric urban region: Towards a
research agenda,” Urban Studies, 38 (4), 623-633.
Lahiri, K., and R. Numrich (1983): “An econometric study on the dynamics of urban
spatial structure,” Journal of Urban Economics, 14 (1), 55-79.
Lang, R.E., and J. LeFurgy (2003): “Edgeless cities: Examining the noncentered
metropolis,” Housing Policy Debate, 14 (3), 427-460.
Lee, B. (2007): “‘Edge’ or ‘edgeless cities’?: Urban spatial structure in US metropolitan
areas, 1980 to 2000,” Journal of Regional Science, 47 (3), 479-515.
Lefever, D.W. (1926): “Measuring geographic concentration by means of the standard
deviational ellipse,” The American Journal of Sociology, 32 (1), 88-94.
Lei, Q. (2007): “Socio-economic forces behind sprawl and compactness in Beijing,”
ENHR 2007 International Conference ‘Sustainable Urban Areas’, Rotterdam.
Levine, J. (1998): “Rethinking accessibility and jobs-housing balance,” Journal of the
American Planning Association, 64 (2), 133-149.
Levine, N. (2007): CrimeStat: A Spatial Statistics Program for the Analysis of Crime
Incident Locations (v 3.1), Houston, TX: Ned Levine & Associates and
Washington, DC: the National Institute of Justice.

192
Levinson, D.M. (1998): “Accessibility and the journey to work,” Journal of Transport
Geography, 6 (1), 11–21.
Li, S.M., and Y.M. Siu (2001): “Residential mobility and urban restructuring under
market transition: A study of Guangzhou, China,” Professional Geographer, 53
(2), 219-229.
Li, Z., and F. Wu (2006): “Socio-spatial differentiation and residential inequalities in
Shanghai: A case study of three neighbourhoods,” Housing Studies, 21 (5), 695 -
717.
Lin, G.C.S. (2002): “The growth and structural change of Chinese cities: a contextual and
geographic analysis,” Cities, 19 (5), 299-316.
Lin, G.C.S. (2004a): “A theme issue on planning for China’s large cities in the era of
globalization,” Progress in Planning, 61 (3), 137.
Lin, G.C.S. (2004b): “Toward a post-socialist city? Economic tertiarization and urban
reformation in the Guangzhou Metropolis, China,” Eurasian Geography and
Economics, 45 (1), 18-44.
Lin, G.C.S. (2007a): “Chinese urbanism in question: State, society, and the reproduction
of urban spaces,” Urban Geography, 28 (1), 7-29.
Lin, G.C.S. (2007b): “Reproducing spaces of Chinese urbanisation: New city-based and
land-centred urban transformation,” Urban Studies, 44 (9), 1827-1855.
Lin, G.C.S., and Y.H.D. Wei (2002): “China’s restless urban landscapes 1: new
challenges for theoretical reconstruction,” Environment and Planning A, 34 (9),
1535-1544.
Liu, X., and W. Liang (1997): “Zhejiangeun: social and spatial implications of informal
urbanization on the periphery of Beijing,” Cities, 14 (2), 95-108.
Logan, J.R., ed. (2002): The New Chinese City: Globalization and Market Reform,
Oxford, UK: Blackwell.
Luo, J., and Y.D. Wei (2006): “Population distribution and spatial structure in transitional
Chinese cities: A study of Nanjing,” Eurasian Geography and Economics, 47 (5),
585-603.
Ma, L.J.C. (2002): “Urban transformation in China, 1949-2000: A review and research
agenda,” Environment and Planning A, 34 (9), 1545-1570.

193
Ma, L.J.C. (2004): “Economic reforms, urban spatial restructuring, and planning in
China,” Progress in Planning, 61 (3), 237-260.
Ma, L.J.C. (2005): “Urban administrative restructuring, changing scale relations and local
economic development in China,” Political Geography, 24 (4), 477-497.
Ma, L.J.C., and B. Xiang (1998): “Native place, migration and the emergence of peasant
enclaves in Beijing,” The China Quarterly, 155, 546-581.
Ma, L.J.C., and F. Wu (2005a): “Restructuring the Chinese city: Diverse processes and
reconstituted spaces,” In: L.J.C. Ma, and F. Wu (eds), Restructuring the Chinese
City, London: Routledge.
Ma, L.J.C., and F. Wu, eds. (2005b): Restructuring the Chinese City, London: Routledge.
Ma, L.J.C., and G. Cui (1987): “Administrative changes and urban population in China,”
Annals of the Association of American Geographers, 77 (3), 372-395.
Maoh, H., and P. Kanaroglou (2007): “Geographic clustering of firms and urban form: a
multivariate analysis,” Journal of Geographical Systems, 9 (1), 29-52.
McDonald, J.F. (1987): “The identification of urban employment subcenters,” Journal of
Urban Economics, 21 (2), 242-258.
McDonald, J.F. (1989): “Econometric studies of urban population density: A survey,”
Journal of Urban Economics, 26 (3), 361-385.
McDonald, J.F., and D.P. McMillen (1990): “Employment subcenters and land values in
a polycentric urban areas: The case of Chicago,” Environment and Planning A, 22
(12), 1561-1574.
McDonald, J.F., and P.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 (3), 448-473.
McMillen, D.P. (2004): “Employment densities, spatial autocorrelation, and subcenters in
large metropolitan areas,” Journal of Regional Science, 44 (2), 225-243.
McMillen, D.P. and S.C. Smith (2003): “The number of subcenters in large urban areas,”
Journal of Urban Economics, 53 (3), 321-338.

194
McMillen, D.P., and J.F. McDonald (1997): “A nonparametric analysis of employment
density in a polycentric city,” Journal of Regional Science, 37 (4), 591-612.
Mieszkowski, P., and B. Smith (1991): “Analyzing urban decentralization: The case of
Houston,” Regional Science and Urban Economics, 21 (2), 183-199.
Mieszkowski, P., and E.S. Mills (1993): “The causes of metropolitan suburbanization,”
Journal of Economic Perspectives, 7 (3), 135-147.
Mills, E.S. (1967): “An aggregative model of resource allocation in a metropolitan area,”
American Economic Review, 61, 197-210.
Mills, E.S. (1972): Studies in the Structure of the Urban Economy, Baltimore: Johns
Hopkins University Press.
Mills, E.S., and J.P. Tan (1980): “A comparison of urban population density functions in
developed and developing countries,” Urban Studies, 17 (3), 313-321.
Muth, R. (1969): Cities and Housing. Chicago, Illinois: University of Chicago Press.
Newling, B. (1969): “The spatial variation of urban population densities,” Geographical
Review, 59 (2), 242-252.
Nowlan, D.M., and G. Stewart (1991): “Downtown population growth and commuting
trips: Recent experience in Toronto,” Journal of the American Planning
Association, 57 (2), 165-182.
O’Kelly, M.E., and W. Lee (2005): “Disaggregate journey-to-work data: implications for
excess commuting and jobs-housing balance,” Environment and Planning A, 37
(12), 2233-2252.
Papageorgiou, Y.Y., and D. Pines (1999): An Essay on Urban Economic Theory, Boston:
Kluwer Academic Publishers.
Parr, J.B. (1985a): “A population-density approach to regional spatial structure,” Urban
Studies, 22 (4), 289-303.
Parr J.B. (1985b): “The form of the regional density function,” Regional Studies, 19 (6),
535-546.
Peng, Zhong-Ren (1997): “The jobs-housing balance and urban commuting,” Urban
Studies, 34 (8), 1215-1235.

195
Punpuing, S. (1993): “Correlates of commuting patterns: A case-study of Bangkok,
Thailand,” Urban Studies, 30 (3), 527-546.
Redfearn, C.L. (2007): “The topography of metropolitan employment: Identifying centers
of employment in a polycentric urban area,” Journal of Urban Economics, 61 (3),
519-541.
Riguelle, F., I. Thomas, and A. Verhetsel (2007): “Measuring urban polycentrism: a
European case study and its implications,” Journal of Economic Geography, 7 (2),
193-216.
Ripley, B.D. (1976): “The second-order analysis of stationary point processes,” Journal
of Applied Probability, 13 (2), 255-266.
Rosetti, M., and B. Eversole (1993): “Journey to work trends in the United States and its
metropolitan areas,” Cambridge, MA: John A. Volpe National Transportation
Systems Center.
Sasaki, K., and Se-Il Mun (1996): “A dynamic analysis of multiple-center formation in a
city,” Journal of Urban Economics, 40 (3), 257-278.
Schwanen, T., F.M. Dieleman, and M. Dijst (2003): “Car use in Netherlands daily urban
systems: Does polycentrism result in lower commute times?,” Urban Geography,
24 (5), 410-430.
Shearmur, R., W. Coffey, C. Dubé, and R. Barbonne (2007): “Intrametropolitan
employment structure: Polycentricity, scatteration, dispersal and chaos in Toronto,
Montreal and Vancouver, 1996-2001,” Urban Studies, 44 (9), 1713-1738.
Shen, Q. (2000): “Spatial and social dimensions of commuting,” Journal of the American
Planning Association, 66 (1), 68-82.
Silverman, B.W. (1986): Density Estimation for Statistics and Data Analysis, London:
Chapman and Hall.
Simpson, W. (1987): “Workplace location, residential location, and urban commuting,”
Urban Studies, 24 (2), 119-128.
Small, K.A., and S. Song (1994): “Population and employment densities: Structure and
change,” Journal of Urban Economics, 36 (3), 292-313.

196
Sohn, J. (2002): Spatial Econometric Modeling of Urban Spatial Structure of Chicago
and Seoul in the 1990s: Agglomeration Economies, Information Technology and
Commuting, Ph.D. Dissertation, University of Illinois at Urbana-Champaign.
Song, S. (1994): “Modelling worker residence distribution in Los Angeles region,” Urban
Studies, 31 (9), 1533-1544.
Song, Y., C. Ding, and G. Knaapd (2006): “Envisioning Beijing 2020 through sketches of
urban scenarios,” Habitat International, 30 (4), 1018-1034.
Sultana, S. (2002): “Job/housing imbalance and commuting time in the Atlanta
metropolitan area: exploration of causes of longer commuting time,” Urban
Geography, 23 (8), 728-749.
Thurstain-Goodwin, M., and D. Unwin (2000): “Defining and delineating the central
areas of towns for statistical monitoring using continuous surface
representations,” Transactions in GIS, 4 (4), 305-317.
Turner, T., and D. Niemeier (1997): “Travel to work and household responsibility: new
evidence,” Transportation, 24 (4), 397-419.
Vandersmissen, Marie-Hélène, P. Villeneuve, and M. Thériault (2003): “Analyzing
changes in urban form and commuting time,” The Professional Geographer, 55
(4), 446-463.
Wachs, M., B.D. Taylor, N. Levine, and P. Ong (1993): “The changing commute: A case
study of the jobs-housing relationship over time,” Urban Studies, 30 (10), 1711-
1729.
Wang, F. (2001): “Explaining intraurban variations of commuting by job proximity and
workers’ characteristics,” Environment and Planning B, 28 (2), 169-182.
Wang, F., and Y. Zhou (1999): “Modeling urban population densities in Beijing 1982-90:
Suburbanization and its causes,” Urban Studies, 36 (2): 271-288.
Wang, S., and Y. Zhang (2005): “The new retail economy of Shanghai,” Growth and
Change, 36 (1), 41-73.
Wei, Y.D. (2005): “Planning Chinese cities: The limits of transitional institutions,” Urban
Geography, 26 (3), 200-221.

197
Wei, Y.H.D., and G.C.S. Lin (2002): “China’s restless urban landscapes 2: socialist state,
globalization, and urban change,” Environment and Planning A, 34 (10), 1721-
1724.
Wheaton, W.C. (1974): “A comparative static analysis of urban spatial structure,” Journal
of Economic Theory, 9 (2), 223-237.
White, K. (1980): “A heteroscedasticity-consistent covariance matrix estimator and a
direct test for heteroskedasticity,” Econometrica, 48 (4), 817-838.
White, M.J. (1976): “Firm suburbanization and urban subcenters,” Journal of Urban
Economics, 3 (4), 323-343.
White, M.J. (1986): “Sex differences in urban commuting patterns,” The American
Economic Review, 76 (2), 368-372.
Wong, D.W.S., and J. Lee (2005): Statistical Analysis with ArcView and ArcGIS, New
York: Wiley.
Wu, F. (1997): “Urban restructuring in China’s emerging market economy: Towards a
framework for analysis,” International Journal of Urban and Regional Research,
21 (4), 640-663.
Wu, F. (1998a): “Polycentric urban development and land-use change in a transitional
economy: the case of Guangzhou,” Environment and Planning A, 30 (6), 1077-
1100.
Wu, F. (1998b): “The new structure of building provision and the transformation of the
urban landscape in Metropolitan Guangzhou, China,” Urban Studies, 35 (2), 259-
284.
Wu, F. (2001): “China’s recent urban development in the process of land and housing
marketisation and economic globalisation,” Habitat International, 25 (3), 273-289.
Wu, F. (2005): “The city of transition and the transition of cities,” Urban Geography, 26
(2), 100-106.
Wu, F., and A.G.O. Yeh (1999): “Urban spatial structure in a transitional economy: the
case of Guangzhou, China,” Journal of the American Planning Association, 65 (4),
377-394.
Wu, F., and Z. Li (2005): “Sociospatial differentiation: processes and spaces in
subdistricts of shanghai,” Urban Geography, 26 (2), 137-166.

198
Wu, W. (2005): “Migrant residential distribution and metropolitan spatial development in
Shanghai,” In: L.J.C. Ma, and F. Wu (eds), Restructuring the Chinese City,
London: Routledge.
Wu, W. (2008): “Migrant settlement and spatial distribution in Metropolitan Shanghai,”
The Professional Geographer, 60 (1), 101-120.
Xie, Y., C. Fang, G.C.S. Lin, H. Gong, and B. Qiao (2007): “Tempo-spatial patterns of
land use changes and urban development in globalizing China: A study of
Beijing,” Sensors, 7 (11), 2881-2906.
Yang, F.F. (2004): “Services and metropolitan development in China: the case of
Guangzhou,” Progress in Planning, 61 (3), 181-209.
Yeh, A.G.O., and F. Wu (1999): “The transformation of the urban planning system in
China from a centrally-planned to transitional economy,” Progress in Planning, 51
(3), 167-252.
Yin, H., X. Shen, and Z. Zhao (2005): “Industrial restructuring and urban spatial
transformation in Xi’an,” In: L.J.C. Ma, and F. Wu (eds), Restructuring the
Chinese City, London: Routledge.
Yuill, R.S. (1971): “The standard deviational ellipse: An updated tool for spatial
description,” Geografiska Annaler, Series B, 53 (1), 28-39.
Zhang, L. (2001): Strangers in the City: Reconfigurations of Space, Power, and Social
Networks within China’s Floating Population, Stanford: Stanford University Press.
Zhang, L. (2005): “Migrant enclaves and impacts of redevelopment policy in Chinese
cities,” In: L.J.C. Ma, and F. Wu (eds), Restructuring the Chinese City, London:
Routledge.
Zhang, L., and S.X.B. Zhao (1998): “Re-examining China’s “urban” concept and the
level of urbanization,” China Quarterly, 154, 330-381.
Zheng, S., and M.E. Kahn (2008): “Land and residential property markets in a booming
economy: New evidence from Beijing,” Journal of Urban Economics, 63 (2), 743-
757.

199
Zhou, Y. (1997): “On the suburbanization of Beijing,” Chinese Geographical Science, 7
(3), 208-219.
Zhou, Y., and L.J.C. Ma (2000): “Economic restructuring and suburbanization in China,”
Urban Geography, 21 (3), 205-236.

Abstract (if available)

Abstract From a comparative international perspective, this dissertation explores the spatial distributions of population and employment in the Beijing metropolitan area in the post-reform era. This study aims to extend the literature on urban spatial structure, with special reference to the pattern and process of urban decentralization and restructuring from a developing and transitional economy context, and to offer further understanding of the spatial organization of contemporary urban areas that departs from the North American or European experience.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Urban spatial transformation and job accessibility: spatial mismatch hypothesis revisited

PDF

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

PDF

Unraveling decentralization of warehousing and distribution centers: three essays

PDF

Beyond spatial mismatch: immigrant employment in urban America

PDF

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

PDF

The long-term impact of COVID-19 on commute, employment, housing, and environment in the post-pandemic era

PDF

The demand for reliable travel: evidence from Los Angeles, and implications for public transit policy

PDF

Spatial and temporal expenditure-pricing equity of rail transit fare policies

PDF

Essays on economic modeling: spatial-temporal extensions and verification

PDF

The development implications of China’s Belt and Road Initiative for Russia, Kazakhstan and Belarus

PDF

The impact of demographic shifts on automobile travel in the United States: three empirical essays

PDF

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

PDF

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

PDF

Social construction of the experience economy: the spatial ecology of outdoor advertising in Los Angeles

PDF

Distribution and correlates of feral cat trapping permits in Los Angeles, California

PDF

The crisis of potable water in Mexico City: institutional factors and water property rights as conditions for creating adequate metropolitan water governance

PDF

An empirical analysis of the relationship between real estate investment and regional economy in China

Asset Metadata

Creator Sun, Tieshan (author)

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

School School of Policy, Planning, and Development

Degree Doctor of Philosophy

Degree Program Planning

Publication Date 04/29/2009

Defense Date 03/26/2009

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

Tag employment distribution,OAI-PMH Harvest,population distribution,urban spatial structure

Place Name Beijing (city or populated place), China (countries)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Redfearn, Christian L. (committee chair), Giuliano, Genevieve (committee member), Moore, James Elliott, II (committee member)

Creator Email tieshans@usc.edu,tieshansun@hotmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2156

Unique identifier UC1313267

Identifier etd-Sun-2848 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-229096 (legacy record id),usctheses-m2156 (legacy record id)

Legacy Identifier etd-Sun-2848.pdf

Dmrecord 229096

Document Type Dissertation

Rights Sun, Tieshan

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