University of Southern California Dissertations and Theses

Essays on congestion, agglomeration, and urban spatial structure

(USC Thesis Other)

Essays on congestion, agglomeration, and urban spatial structure

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

December 2014

ESSAYS ON CONGESTION, AGGLOMERATION, AND URBAN SPATIAL
STRUCTURE
By

Yuting Hou

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
(POLICY, PLANNING AND DEVELOPMENT)
Copyright 2014 Yuting Hou
ii

DEDICATION

To my parents

iii
ACKNOWLEDGEMENTS
I would like to express my deepest gratitude to my advisors Professors Christian
Redfearn and Genevieve Giuliano. I wish to thank Professor Redfearn for cultivating my interest
in urban economic research, pushing me hard to search for engaging research topic and tutoring
me to turn those interesting but vague ideas into this dissertation. I wish to thank Professor
Giuliano for helping me to improve my research skills, guiding me through the hazy parts of my
dissertation, and encouraging me to be meticulous and precise. Their critical perspectives and
invaluable suggestions have benefited me throughout my doctoral study and would benefit me
for the rest of my academic life.
My heartfelt thanks to my dissertation committee members Professors Marlon Boarnet
and James Moore II for their careful reading of my dissertation and their constructive critiques
and suggestions. I appreciate the valuable time from them to talk with me regarding my
dissertation research and help me think critically about my work.
In addition, I am also grateful to my cohorts and friends at USC. With their help and
supports, the long journey of doctoral study has become meaningful and interesting.
Finally, I want to thanks my parents. Without their endless love, I would not have been
able to go this far. I would like to dedicate this dissertation to them.

iv
TABLE OF CONTENTS

ACKNOWLEDGEMENTS ........................................................................................................ iii
LIST OF TABLES ................................................................................................................................. vii
LIST OF FIGURES .............................................................................................................................. viii
ABSTRACT.. ............................................................................................................................................ ix
CHAPTER 1 INTRODUCTION .......................................................................................................... 1
1 Background and motivation ................................................................................................. 1
2 Research questions and approach ......................................................................................... 3
3 Dissertation structure ............................................................................................................ 5
Chapter 1 References .............................................................................................................. 8
CHAPTER 2 AGGLOMERATION ECONOMIES, ACCESSIBILITY, AND
CONGESTION COSTS .............................................................................................. 10
1 Introduction ........................................................................................................................ 10
2 Agglomeration economies and diseconomies—An overview of theoretical basis and
implications ........................................................................................................................ 12
2.1 Agglomeration economies: How Close is Close Enough ........................................... 12
2.2 Theoretical basis of congestion costs (diseconomies) ................................................. 15
2.3 Implications for metropolitan spatial structure ........................................................... 18
3 Do economies of proximity matter for firm location decisions? ........................................ 20
3.1 Measures of agglomeration economies ....................................................................... 21
3.2 Empirical evidence: the geographic extent of agglomeration benefits ....................... 26
3.3 Summary ..................................................................................................................... 37
4 Does traffic congestion reduce economic efficiency? ........................................................ 39
4.1 Modeling the effects of traffic congestion on the economy ........................................ 40
4.2 The endogeneity of traffic congestion to the economy ............................................... 47
4.3 Empirical results .......................................................................................................... 49
5 Conclusions and extensions ................................................................................................ 51
Chapter 2 References ............................................................................................................ 53
Appendix for Chapter 2 ............................................................................................................. 59
v
CHAPTER 3 TRAFFIC CONGESTION, POLYCENTRICITY, AND INTRA-URBAN
FIRM LOCATION CHOICES ................................................................................. 72
1 Introduction ........................................................................................................................ 72
2 Literature review: Transport, traffic congestion, and firm location ................................... 73
2.1 Theoretical basis .......................................................................................................... 74
2.2 Possible impacts of traffic congestion on firms .......................................................... 77
3 Research approach .............................................................................................................. 78
3.1 The basic model ........................................................................................................... 79
3.2 Definition of choice sets .............................................................................................. 79
3.3 Definition of explanatory variables ............................................................................. 80
4 Data .................................................................................................................................... 87
4.1 Employment data ......................................................................................................... 87
4.2 Transportation network data ........................................................................................ 94
4.3 Land use data ............................................................................................................... 94
4.4 Population data ............................................................................................................ 95
5 Results ................................................................................................................................ 95
5.1 Descriptive results ....................................................................................................... 95
5.2 Estimation results ...................................................................................................... 102
6 Conclusions ...................................................................................................................... 111
Chapter 3 References .......................................................................................................... 115
Appendix for Chapter 3 ........................................................................................................... 120
CHAPTER 4 TRAFFIC CONGESTION, ACCESSIBILITY TO EMPLOYMENT AND
HOUSING PRICES ................................................................................................... 144
1 Introduction ...................................................................................................................... 144
2 Literature review .............................................................................................................. 146
2.1 Theoretical basis ........................................................................................................ 146
2.2 Empirical evidence .................................................................................................... 150
3 Research approach ............................................................................................................ 154
3.1 Basic model development ......................................................................................... 155
3.2 Identification of (regional) congestion effects .......................................................... 159
3.3 Housing market segmentation ................................................................................... 160
vi
3.4 Construction of locational variables .......................................................................... 162
4 Data .................................................................................................................................. 167
4.1 House price data ........................................................................................................ 167
4.2 Employment data ....................................................................................................... 170
4.3 Transportation network data ...................................................................................... 170
4.4 Population data .......................................................................................................... 172
4.5 Land use data ............................................................................................................. 173
5 Results .............................................................................................................................. 173
5.1 Basic model estimation ............................................................................................. 174
5.2 Differential effects of accessibility by income .......................................................... 180
6 Conclusions ...................................................................................................................... 186
Chapter 4 References .......................................................................................................... 189
Appendix for Chapter 4 ........................................................................................................... 193
CHAPTER 5 CONCLUSIONS ......................................................................................................... 200
1 Summary of findings ........................................................................................................ 200
2 Policy implications ........................................................................................................... 203
3 Limitations and future research ........................................................................................ 204
Chapter 5 Reference ............................................................................................................ 206

vii
LIST OF TABLES
Table 2-a1 Basic information of selected empirical studies on firm location choice ................................. 59
Table 2-a2 Summary of empirical studies on firm location choices ........................................................... 62
Table 2-a3 Summary of empirical studies on the negative impacts of congestion on economic efficiency
................................................................................................................................................ 70

Table 3-1 Summary statistics of congestion levels at different locations (12-mile boundary) ................... 99
Table 3-2 Estimation results of firm location choice models, 2001-2005 (1) ........................................... 105
Table 3-3 Estimation results of firm location choice models, 2001-2005 (2) ........................................... 106

Table 3-a1 Spatial Concentration pattern of different industries (2001-2005) ......................................... 120
Table 3-a2 Spatial Concentration pattern of selected NAICS3 industries across centers (2000, 10/10
centers) .................................................................................................................................. 125
Table 3-a3 T-tests on the equality of the means of LQs in center locations ............................................. 130
Table 3-a4 T-tests on the equality of the means of LQs in non-center locations ...................................... 130
Table 3-a5 Correspondences between broad economic sectors and land use types .................................. 131
Table 3-a6 Summary statistics of explanatory variables ........................................................................... 132
Table 3-a7 Summary statistics of percentage of centers employment in NAICS2 sectors by 3-quantiles 134
Table 3-a8 Summary statistics of accessibility measures at different locations (12-mile boundary) ....... 142
Table 3-a9 Correlation Matrix of agglomeration variables, accessibility and congestion variables......... 143

Table 4-1 Results of Two-level spatial model .......................................................................................... 175
Table 4-2 Summary statistics of housing price and Accessibility by quintiles of median household income
.............................................................................................................................................. 180
Table 4-3 Summary statistics of housing price and Accessibility, income clusters defined by quintiles of
poverty rate ........................................................................................................................... 181
Table 4-4 Results of three-level spatial model .......................................................................................... 182

Table 4-a1 Descriptive statistics of housing price and structural characteristics ...................................... 193
Table 4-a2 Descriptive statistics of locational attributes and data sources (n=1909) ............................... 194

viii
LIST OF FIGURES
Figure 3-1 Spatial distribution of employment centers identified in 2000 (10/10 criterion) ...................... 89
Figure 3-2 Spatial concentration pattern of chosen sectors across center and non-center locations ........... 91
Figure 3-3 Spatial concentration pattern of chosen sectors across centers of different density categories . 92
Figure 3-4 Distribution of new establishments across centers of different size groups .............................. 96
Figure 3-5 Distribution of new establishments within centers of different density (jobs per acre, by 3-
quintiles) ................................................................................................................................... 97
Figure 3-6 Distribution of new establishments across centers of different levels of local specialization in
the same 2-digit sectors ............................................................................................................. 98
Figure 3-7 Center size/density vs. regional congestion level (12-mile boundary applied) ....................... 100
Figure 3-8 Center size/density vs. within-center congestion level ........................................................... 100
Figure 3-9 Estimated turning points for the Delay Rate_inner variable ................................................... 108

Figure 3-a1 Spatial pattern of congestion delays (measured by Delay Rate, 12-mile boundary used) ..... 134
Figure 3-a2 Spatial pattern of access to workers, 12-mile boundary ........................................................ 135
Figure 3-a3 Spatial pattern of access to manufacturing sectors (NAICS2=31-33), 12-mile boundary .... 136
Figure 3-a4 Spatial pattern of access to wholesale trade sectors (NAICS2=42), 12-mile boundary ........ 137
Figure 3-a5 Spatial pattern of access to retail trade sectors (NAICS2=44-45), 12-mile boundary .......... 138
Figure 3-a6 Spatial pattern of access to information sectors (NAICS2=51), 12-mile boundary .............. 139
Figure 3-a7 Spatial pattern of access to FI services (NAICS2=52), 12-mile boundary ............................ 140
Figure 3-a8 Spatial pattern of access to professional services (NAICS2=54), 12-mile boundary ............ 141

Figure 4-1 Spatial distribution of selected samples .................................................................................. 169
Figure 4-2 Random component of estimated coefficients for accessibility (v01k ), income clusters defined
by quintiles of median household income ( ) ....................................................... 185
Figure 4-3 Random component of estimated coefficients for accessibility (v01k ), income clusters defined
by quintiles of poverty rate (_ ) ................................................................................. 185

Figure 4-a1 Spatial distribution of Accessibility to employment (AM peak) ........................................... 195
Figure 4-a2 Spatial distribution of Accessibility to employment (Night period) ..................................... 196
Figure 4-a3 Spatial distribution of population density .............................................................................. 197
Figure 4-a4 Spatial distribution of median household income ................................................................. 198
Figure 4-a5 Spatial distribution of poverty rate ........................................................................................ 199

ix
ABSTRACT
While traffic congestion is considered an important source of diseconomies in theories of
agglomeration economies, only a few studies measure the effects congestion costs empirically. My study
seeks to fill in this gap by directly estimating the economic impact of traffic congestion at the intra-
metropolitan scale. Three major questions are addressed in my dissertation: 1) What is our current
knowledge on agglomeration economies and congestion costs, in terms of their effective geographic
extents, the methods for measuring them, and the methods for unraveling their separate effects? 2) Do
congestion-induced travel time delays adversely affect locations' attractiveness to firms when potential
benefits of agglomeration are controlled? 3) Does traffic congestion negatively affect single-family
housing values by inducing commuting delays and reducing locations' accessibility premiums?
To answer these questions, I review prior theoretical and empirical studies on agglomeration and
congestion to summarize the mechanism by which traffic congestion influences the urban economy and
urban spatial structure. Using the 2001–2005 establishment-level data from the National Establishment
Time Series (NETS) dataset and the 2001–2005 disaggregate housing sales data from the DataQuick
dataset, I examine the impacts of traffic congestion on intra-urban business and residential geography
through the lens of firms’ location choices and households' bid-up prices for housing units, respectively.
Specifically, I empirical test whether and to what extent congestion-induced travel time delays are costly
to firms and households in their location decisions within a metropolitan area, how these influences differ
at different geographic extents, and how these influences vary across firms of different industrial sectors
and home buyers of different income groups. Focusing on the Los Angeles region, the results indicate that
metro-wide congestion delays negatively influence firm location and housing price, while local
congestion delays will not impose a drag on otherwise high levels of agglomeration benefits until
reaching a threshold. I also find that firms valuing proximity (or agglomeration benefits), such as those
specialized in the information, finance and insurance and services sectors, are more likely to endure
congestion costs than those production-related firms and retail firms. Moreover, compared with those in
x
lower-income and upper-income groups, households in middle-income neighborhoods are more
responsive to commuting costs and congestion costs. In sum, my study provides new evidence that
congestion costs matter to urban spatial structure by altering the accessibility patterns and inhibiting
additional agglomeration benefits at the intra-metropolitan scale.

1
CHAPTER 1 INTRODUCTION

1 Background and motivation
In the U.S., traffic congestion has been and continues to be a major urban problem. Peak-hour
traffic congestion is undesirable because it induces travel time delays and other loses to various business
activities (Downs 2004). According to the 2010 Urban Mobility Report (UMR) by the Taxes
Transportation Institute (TTI) study, about 63 percent of peak-hour vehicle miles traveled (VMT) is
congested across the 101 urban areas of more than 500,000 people.
Traffic congestion is important because time is costly to individuals and firms (Gordon, Kumar,
and Richardson 1989). The costs of traffic congestion have been extensively studied. For example, many
studies focus on the evaluation of travel delays, such as the widely cited Urban Mobility Report done by
the TTI. To measure the costs of congestion, many researchers impute the economic value of time lost in
travel delays using the regional wage rates, though there is no uniform standard in terms of the
appropriate wage rate that should be used in the evaluation (Small and Verhoef 2007). Traffic congestion
has also been a major concern in urban transportation planning and policy. For example, despite the
problem of political acceptability, congestion pricing is the most studied congestion policies, which aims
to internalize the externalities individual travelers would impose on other travelers (Downs 1992;
Giuliano 1992; Small 1997).
Nonetheless, we have only limited evidence on how traffic congestion impacts the wider urban
economy besides its immediate impacts on the transportation system. Is congestion really a bad thing for
the urban economy? Are congestion-induced travel time delays really costly to individual firms and
households? Do they change their behavior accordingly? As suggested by various theories of
agglomeration, traffic congestion might not always be a bad thing because it is a particular byproduct of
agglomeration economies. Besides its possible costs, traffic congestion can be associated with economic
prosperity and success because those congested routes and the destinations they connect are more likely
2
to be favored and valued by the society (Downs 1992; Taylor 2002). Thus, traffic congestion could be
better understood as "a drag on otherwise high levels of accessibility" (Taylor 2002, 10). To better
evaluate the costs of traffic congestion, we should also look at the associated agglomeration
benefits/accessibility advantages because it is the tradeoff between the two groups of forces that explains
the structure and growth of cities (Richardson 1995).
1
From a policy perspective, this implies that
policies aimed at solving only one aspect of the two offsetting effects may bring unexpected results for
the urban economy (Brinkman 2013; Mondschein, Taylor, and Brumbaugh 2011; Taylor 2002). For
example, efforts to reduce congestion levels at those highly concentrated places would also reduce the
benefits of agglomeration and hence result in loss of productivity advantages, which could cancel out the
gains from reduced congestion costs (Brinkman 2013; Mondschein, Taylor, and Brumbaugh 2011; Taylor
2002).
A few empirical studies have examined the wider impacts of traffic congestion on the urban
economy (e.g. Boarnet 1997; Fernald 1999; Hymel 2009; Sweet 2013). Informed by the role of
transportation infrastructure in urban economic growth, these studies consider traffic congestion as
specific services provided by the transportation infrastructure that would reduce the economic efficiency
of urban areas (e.g. Boarnet 1997; Fernald 1999).
2
While these empirical studies conducted so far mainly
compare congestion levels and urban economic efficiency across different counties or metropolitan areas,
traffic congestion can also impact the intra-urban economy and geography. Urban economic theories
predict that the internal spatial structure of urban areas will be reshaped to adjust to traffic congestion in
the long run, resulting from the accumulated locational responses of firms and households. Thus, the
micro-economic behavior of firms and households could be an important aspect to examine the
determinants of the urban economy and the evolution of urban spatial structure. This also implies that

1
Agglomeration economies and accessibility are related concepts and their measures share some similarity. A
detailed discussion on the differences in the concepts and measures of accessibility and agglomeration is conducted
in the next chapter.
2
A more detailed review of this issue is conducted in the next chapter.
3
traffic congestion policies should also account for congestion's influence on individual firms and
households (Taylor 2002).
My dissertation aims to contribute to the studies on the effects of traffic congestion in two ways.
First, I explore the current knowledge on the links between agglomeration benefits and congestion costs,
which form the basis for the later empirical analyses. To investigate this issue, I review and summarize
empirical estimates of the two offsetting effects, focusing mainly on three issues that are not so often
addressed in prior studies: 1) the geographic extent of the two effects, 2) the measures of the two effects,
3) the methods of unraveling congestion effects from agglomeration benefits. Secondly, I empirically
explore how the tradeoff between congestion costs and agglomeration benefits/accessibility advantages
works within a metropolitan area by looking at the location behavior of individual firms and households. I
also investigate how firms of different industries and households of different income groups respond to
congestion-induced travel time delays. The two empirical analyses are based on disaggregate employment
data and housing sales data, respectively, and comprise the main part of my dissertation. Since urban
spatial structure evolves as a result of the aggregate location decisions of firms, households and other
economic agents, my findings help us to understand how the internal spatial structure and economic
growth of an urban area would be affected and adjusted in response to congestion costs.
2 Research questions and approach
Based on the stated research purposes, my dissertation address a set of specific research questions
as follows:
1) An overall review on the empirical estimates of agglomeration economies and congestion costs
z Why is the tradeoff between agglomeration economies and congestion costs important to the urban
economy? What are the geographic extents of agglomeration economies and congestion costs? In
other words, "how close is close enough" to allow for spatial interactions between economic agents
4
to generate agglomeration benefits? And to what extent are such benefits offset by the associated
negative externalities at different geographic scales?
z How should we measure agglomeration economies? Do different types and sources of agglomeration
economies merit different measures? Do different measures of agglomeration economies yield
different results?
z How should we measure congestion costs? How should we unravel the costs of congestion from the
potential benefits of agglomeration?
2) Traffic congestion and intra-metropolitan firm location choices
z Do congestion-induced travel time delays at the regional scale reduce locations' attractiveness to new
firms by altering the regional accessibility pattern and reducing the metro-wide agglomeration
benefits?
z Do congestion-induced travel time delays within employment centers reduce their attractiveness to
new firms by reducing the efficiency of economic interactions within centers? "How bad does
congestion have to be before the congestion costs begin to outweigh the agglomeration benefits"
within centers (Mondschein, Taylor, and Brumbaugh 2011, 17)?
z Do firms of different industries respond differently to congestion costs in their location choices
among employment centers? Are firms benefiting more from intra-urban agglomeration less sensitive
to congestion costs?
3) Traffic congestion and single-family housing values
z At the regional scale, does traffic congestion negatively affect single-family housing values by
increasing commuting time/costs and reducing locations' accessibility to employment?
5
z Which income group will pay more for accessibility and congestion costs in their residential location
decisions, as reflected by the housing prices? Are higher-income groups more willing to pay for
better accessibility to avoid congestion delays and the associated time and monetary loss?
z At the neighborhood scale, will traffic congestion defined in physical terms such as dense traffic
flows negatively impact single-family housing values?
These research questions are closely related but investigate the effects of traffic congestion from
different perspectives. Thus, I examine the three sets of research questions through three separate essays.
The results of each analysis are summarized and discussed in the ending chapter to imply a more
complete story of congestion effects on the urban economy and the urban spatial structure.
The latter two sets of questions are addressed by focusing on the Los Angeles metropolitan area,
which has been the most congested urban area over the past 30 years, according to the research by the TTI.
The Los Angeles region has been favored by studies on urban spatial structure (e.g. Giuliano et al. 2010;
Giuliano et al. 2011; Gordon and Richardson 1996; Redfearn 2007, 2009) and intra-urban growth (e.g.
Aji 1995; Giuliano et al. 2011; Giuliano and Small 1999). Despite data availability issues, an important
reason is that Los Angeles has long been considered a precursor of polycentricity that the suburbanization
and decentralization pattern of other urban areas in the automobile age resemble that of Los Angeles
(Anas, Arnott, and Small 1998; Richardson 1995). Thus, my results of congestion effects on spatial
patterns within the Los Angeles can also provide implications for congestion effects at the sub-
metropolitan level in general.
3 Dissertation structure
This dissertation is mainly composed of three relatively independent essays, each of which has its
own introduction and summary. Each essay is summarized as follows.
6
In the first essay (in chapter 2), I review the theories and empirics on agglomeration economies
and congestion costs, focusing on the geographic extents at which they are effective, the wide disparate
ways to measure the two effects, and the methods for unraveling congestion effects from agglomeration
benefits. By summarizing and discussing previous studies on agglomeration economies and congestion
costs, this essay aims to provide the theoretical and methodological basis for the latter two empirical
essays and imply about the research gap in the existing literature that is worth exploring.
In the second essay (in chapter 3), I estimate the economic impacts of traffic congestion on
agglomeration through the lens of firms' location decisions. A discrete choice model is applied to examine
the location choices of new businesses within the Los Angeles metropolitan area. Compared with
previous studies on intra-metropolitan firm location, this study explicitly incorporates traffic congestion
as a locational factor and uses employment centers, defined as significantly dense nodes of employment
within a metropolitan area (e.g. Giuliano and Small 1991), as the spatial choice sets to explore the nature
and role of intra-urban agglomerations. Congestion costs are distinguished as two types: congestion
delays to destinations at the regional scale and congestion delays within centers. The estimation results
show that regional congestion delays reduce the probability of employment centers being chosen by firms
in most industrial sectors after controlling for localization effects, urbanization effects, and the centers'
accessibility advantages. Interestingly, the results also indicate a threshold beyond which further increases
in congestion delays inside centers reduce the agglomeration benefits of the centers. There is some
evidence that services and information sectors benefiting more from face-to-face interactions and
agglomeration are more likely to endure congestion delays within the centers. In sum, this study provides
new evidence of the tradeoff between congestion costs and agglomeration benefits at the intra-
metropolitan scale.
In the third essay (in chapter 4), I investigate the relationship between traffic congestion and
accessibility to employment and their effects on single-family housing prices, which imply the residential
location behavior of individual households. A multi-level hedonic price model is used to estimate the
7
marginal price of accessibility while controlling for neighborhood-level congestion externality, other
neighborhood attributes, and the correlation of proximal housing sales. By doing this, it is possible to
identify the impacts of traffic congestion by comparing the implicit price of accessibility between
congested (AM peak) flow and free flow (night flow). The results indicate that the implicit prices of the
two accessibility measures differ significantly. The accessibility measure based on congested travel time
yields a higher marginal price, which suggests that households are more responsive to peak-hour travel
time. The results also suggest that the effects of job access are more valued by home buyers in middle-
income neighborhoods, compared with those in the lowest-income or highest-income neighborhoods. By
contrast, neighborhood-level traffic congestion measured in physical terms has no significant impact on
housing prices.
At last, the final chapter summarizes major findings and discusses how the two main empirical
analyses imply the effects of congestion on urban spatial structure in general. The implications of the
results for transportation and land use planning and congestion mitigation policies are also discussed. I
also present the additional needs and themes for future research.

8
Chapter 1 References
Aji, Maria Astrakianaki 1995. "Intra-metropolitan Productivity Variations of Selected Manufacturing and Business
Service Sectors: What Can We Learn from Los Angeles?" Urban Studies 32 (7):1081–1096.
Anas, Alex, Richard Arnott, and Kenneth A. Small. 1998. "Urban Spatial Structure." Journal of Economic Literature
36 (3):1426–1464.
Boarnet, Marlon G. 1997. "Infrastructure Services and the Productivity of Public Capital: the case of streets and
highways." National Tax Journal 50:39–58.
Brinkman, Jeffrey. 2013. "Congestion, Agglomeration, and the Structure of Cities [online]." Philadelphia: Federal
Reserve Banks of Philadelphia [cited 16 September 2013]. Available from:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2272049
Downs, Anthony. 1992. Stuck in Traffic: Coping with Peak-Hour Traffic Congestion. Washington, D.C.: Brookings
Institution Press.
Fernald, John G. 1999. "Roads to Prosperity? Assessing the Link between Public Capital and Productivity."
American Economic Review 89 (3):619–638.
Giuliano, Genevieve. 1992. "An Assessment of the Political Acceptability of Congestion Pricing." Transportation 19
(4):335–358.
Giuliano, Genevieve, Christian Redfearn, Ajay Agarwal, and Sylvia He. 2011. "Network Accessibility and
Employment Centers." Urban Studies 49 (1):77–95.
Giuliano, Genevieve, Peter Gordon, Qisheng Pan, and JiYoung Park. 2010. "Accessibility and Residential Land
Values: some tests with new measures." Urban studies 47 (14):3103–3130.
Giuliano, Genevieve, and Kenneth A. Small. 1991. "Subcenters in the Los Angeles region." Regional Science and
Urban Economics 21 (2):163–182.
Giuliano, Genevieve, and Kenneth A Small. 1999. "The Determinants of Growth of Employment Subcenters."
Journal of Transport Geography 7 (3):189–201.
Gordon, Peter, Ajay Kumar, and Harry W Richardson. 1989. "Congestion, Changing Metropolitan Structure, and
City Size in the United States. " International Regional Science Review 12 (1):45–56.
Gordon, Peter, and Harry W. Richardson. 1996. "Beyond Polycentricity: The Dispersed Metropolis, Los Angeles,
1970–1990. " Journal of the American Planning Association 62 (3):289–295.
Hymel, Kent. 2009. "Does Traffic Congestion Reduce Employment Growth?" Journal of Urban Economics 65 (2):
127–135.
Mondschein, A., Brian D. Taylor, and Stephen Brumbaugh. 2011. "Congestion and Accessibility: What's the
Relationship? [online]." Los Angeles: University of California, University of California Transportation
Center (UCTC), [cited 16 September 2013]. Available from:
https://escholarship.org/uc/item/6bh2n9wx.pdf
Redfearn, Christian L. 2007. "The Topography of Metropolitan Employment: Identifying Centers of Employment in
a Polycentric Urban Area." Journal of Urban Economics 61 (3):519–541.
Redfearn, Christian L. 2009. "Persistence in Urban Form: The Long-Run Durability of Employment Centers in
Metropolitan Areas." Regional Science and Urban Economics 39 (2):224–232.
Richardson, Harry W. 1995. "Economies and Diseconomies of Agglomeration." In Urban Agglomeration and
Economic Growth, edited by Herbert Giersch, 123–155. Berlin-Heidelberg: Springer.
9
Small, Kenneth A, and Erik T Verhoef. 2007. The Economics of Urban Transportation. New York: Routledge.
Small, Kenneth A. 1997. "Economics and Urban Transportation Policy in the United States." Regional Science and
Urban Economics 27 (6):671–691.
Sweet, Matthias. 2013. "Traffic Congestion's Economic Impacts: Evidence from US Metropolitan Regions." Urban
Studies 51 (10): 2088–2110.
Taylor, Brian D. 2002. "Rethinking Traffic Congestion." ACCESS Magazine 1 (21): 8–16.

10
CHAPTER 2 AGGLOMERATION ECONOMIES, ACCESSIBILITY, AND
CONGESTION COSTS

—A review of some remaining questions

1 Introduction
Agglomeration economies and congestion costs are two fundamental forces in urban and regional
studies. In theory, the tradeoff between the two forces explains the formation of cities, the optimal size of
cities, the spatial distribution of economic activities within cities, and the growth or economic efficiency
of cities (Richardson 1995). In empirical studies, agglomeration economies and congestion costs have
been examined extensively in different fields of studies. For example, the urban growth studies explain
the variation of economic growth across different urban areas as a result of the differences in economy of
scale that firms would experience (see Melo, Graham, and Noland 2009 for a complete review), while
transportation studies usually evaluate the costs of traffic congestion by directly evaluating travel delays
in monetary terms, such as the total value of time and fuel costs lost in travel delays (e.g. Goodwin 2004;
Lomax, Schrank, and Eisele 2012).
Despite abundant evidence on the effects of agglomeration economies and congestion costs,
several questions remain regarding the nature and role of the two forces in urban/regional development
that deserve further theoretical and empirical exploration. First of all, questions remain regarding the
geographic extent at which the two opposing forces work and trade off each other. In other words, the
questions include: 1) how close is close enough to allow for spatial interactions between economic agents
(i.e., firms and workers) to generate agglomeration benefits; 2) at what spatial scale are the costs of
agglomeration generated due to competition between economic agents in close proximity, thus
counterbalancing the potential benefits of agglomeration? For example, while urban growth studies
usually consider metropolitan areas as club goods and evaluate and compare the agglomeration effects
11
across urban areas (Rosenthal and Strange 2003), we also see evidence that economic activities within
metropolitan areas are not evenly distributed but are concentrated in a few nodes (Agarwal, Giuliano, and
Redfearn 2012). Exploring the geographic extent of agglomeration economies and congestion costs could
improve our understanding of the nature of urban spatial structure and economic growth.
Questions also remain regarding the methods of measuring agglomeration economies and
congestion costs. For example, the positive effects of agglomeration economies have been tested in
different research fields and various measures have been developed. While early studies usually focused
on the agglomeration benefits associated with the scale of urban economy and used proxies such as urban
population size or employment density, some recent studies use measures such as accessibility indices to
allow for spatial spillover effects from neighboring locations. Thus, a set of questions would be raised
such as what the systematic reasons are behind the construction of each type of agglomeration measures,
whether different types and sources of agglomeration economies merit different measures, and more
importantly, whether different measures yield different results. Exploring these questions could assist in
the development of more consistent and meaningful measures of the two forces, which is one of the major
concerns of this study.
Compared with the extensive studies on the positive effects of agglomeration economies, the
negative effects of congestion costs and their relationship with agglomeration benefits are not often
examined in agglomeration researches, despite the theoretical emphasis (Hymel 2009). One possible
reason is that there is a lack of consensus on the most effective measure of various types and sources of
congestion costs. The other important reason is that congestion costs and agglomeration economies
coexist and overlap spatially (Graham 2007), so that congested cities or areas may also be economically
robust places with high growth potentials. For example, the three largest metropolitan areas in the U.S.,
Los Angeles, New York, and Chicago, are mostly traffic congested areas, but are also desirable places
because of the large variety of goods, services, and economic opportunities they provide to individuals
and firms (Taylor 2002). Thus, the final remaining question is how to evaluate the costs of agglomeration
12
separately from its potential benefits. This chapter reviews a few examples that identify the role of
congestion costs in economic growth and suggests implications for future studies.
The rest of this chapter is divided into three major sections. Section 2 briefly reviews the
theoretical basis of agglomeration economies and congestion costs: how and why the tradeoff between
them explains the structure and growth of cities/regions and at what geographic scales their effects work.
In section 3, I review empirical studies on firm location to explore how agglomeration economies of
different types and sources are measured and whether different measures yield different results. I also
illustrate whether a particular type of agglomeration economies exerts different impacts on firm location
decisions at different geographic scales. Section 4 reviews the methods and empirical evidence on the
economic impacts of traffic congestion to illustrate how congestion costs can be measured and how
congestion effects can be distinguished from agglomeration effects. This chapter concludes with a
summary of the major findings in the existing literature and provides implications for the later empirical
analyses in my dissertation.
2 Agglomeration economies and diseconomies—An overview of theoretical basis
and implications
2.1 Agglomeration economies: How Close is Close Enough
Theories of agglomeration economies explain cost savings and increasing returns external to
firms as a result of spatial concentration of economic agents, inputs, and various activities. Traditionally,
theories emphasize three sources of agglomeration economies tracing back to Marshall's (1920) seminal
work: reduced transportation costs associated with input-output linkages between firms in related
industries, a pooled market for specialized labor, and knowledge spillovers and information exchange
between firms. All three sources are facilitated by spatial proximity because of easier access to economic
inputs and easier interaction between economic agents.
Input-output linkages for firms located in spatial proximity to each other are enhanced because of
cost savings in market transactions such as buying, selling, and subcontracting (Folta, Cooper, and Baik
13
2006). Traditionally, the transaction cost approach of agglomeration economies suggests that vertically-
related firms are likely to be disintegrated and spatially concentrated when the internal transaction costs
are higher than the external transaction costs (Parr 2002; Scott 1988). When input-output relations are
small-scale, unstandardized, unstable, and not-easily manageable, transport costs per unit of flow would
be high relative to the output value, so that firms tend to agglomerate to achieve saving in transport costs
(Scott 1988, 44–58). On the contrary, when input-output relations are routinized and regulated, transport
costs would be low and firms tend to disperse from agglomerated regions (Scott 1988, 44–58). In other
words, spatial clustering is needed for the generation of these agglomeration economies only when the
costs of exchanging goods and services between firms are sensitive to distance (Montgomery 1988).
In addition to a reduction in physical transport costs, spatial proximity between firms also
facilitates the flow of information and knowledge exchange. This is because the exchange of information
associated with physical transactions, especially those involving non-coded information or tacit
knowledge, involves frequent and repeated face-to-face interactions that require geographic proximity
between economic agents (Torre and Rallet 2004). Face-to-face interactions include both formal and
informal interactions. Formal face-to-face meetings are especially important for firms' production
processes because they need to build up and maintain input/output linkages with their service
supplier/customers (Coffey, Drolet, and Polèse 1996; Ihlanfeldt and Raper 1990). Less formal face-to-
face contacts between employees of different firms, such as meetings and information exchanges held
over drinks or lunch, also contribute greatly to the business environment of agglomerations and help firms
learn from each other (Coffey, Drolet, and Polèse 1996; Ihlanfeldt and Raper 1990). In the modern
economy, cities continue to exist and growth despite the decline of transport costs because of their
efficiency in facilitating knowledge and information exchanges among firms and workers (Gordon and
McCann 2000; Parr 2004; Storper and Venables 2004).
In addition to the enhanced inter-firm linkages, theories of agglomeration economies suggest that
access to a larger labor pool with specialized skills contributes greatly to individual firms' productivity
14
because it increases individual firms' adaptability to shocks (Overman and Puga 2010): a positive shock
enables an establishment to respond by expanding production without raising wages, while a negative
shock enables an establishment to contract production without lowering wages. A larger labor pool also
leads to a finer division of labor (Rosenthal and Strange 2004), thereby reducing the costs associated with
the search for labor skills matching a firm's specific production requirements. A larger labor pool also
reduces training costs, especially for workers with more generalized skills rather than firm-specific human
capital (Montgomery 1988). Since workers are also potential consumers, easier access to workers also
implies access to a larger home market (Fujita and Mori 2005).
Despite the importance of spatial proximity and linkages between economic agents, theories of
agglomeration economies do not explicitly discuss the geographic extent of agglomeration economies or
the spatial decay pattern of agglomeration economies. For example, some of the theoretical studies on
industrial districts or industrial complexity do not specify the spatial dimension of agglomeration
(Drucker 2012). The answer to the question of "how close is close enough" to derive agglomeration
benefits, however, varies by different types and sources of agglomeration externalities. Empirical studies
suggest that the geographic extent of agglomeration economies depends on the nature of inter-firm
linkages or other spatial interactions: when interactions are time sensitive and mainly involve flows of
people or information, agglomeration economies operate at a much smaller distance; when less frequent
face-to-face interaction is required, agglomeration economies operate across a much larger distance
(Drucker 2012; Grek, Karlsson, and Klaesson 2009).
Thus, the spatial extent of Marshall's three externalities differs. For example, labor market
pooling effects are usually defined within a metropolitan area or because workers' acceptable commuting
distance/time is limited (Rosenthal and Strange 2001). The knowledge spillover effects work even at a
smaller scale, such as the zipcode level, and decay quickly with distance (Drucker 2012; Rosenthal and
Strange 2001). The spatial extent of other benefits associated with localized input-output linkages,
however, varies with the production technology, the types of goods and services transported, and the
15
associated transport costs (Drucker 2012; Scott 1988). It is also expected that firms in industries that are
sensitive to learning, innovation, and local availability of intermediate inputs or industries that rely on
face-to-face contacts for building up, maintaining, and adjusting relationships with suppliers and
customers are more likely to benefit from agglomerating at a much closer geographic distance (Drucker
2012).
Only a few empirical studies directly or indirectly test the geographic extent of agglomeration
economies (e.g. Drucker 2012; Rosenthal and Strange 2001). Most empirical studies do not look inside
the "black box" of agglomeration economies but rather measure their potential benefits through "catch-
all" agglomeration variables (such as the total employment/population of cities/regions) that do not
capture different sources of agglomeration economies (Rigby and Essletzbichle 2002). In terms of the
spatial extent of agglomeration benefits, some studies find that agglomeration economies operate at a
much smaller geographic scale than cities and attenuate quickly with distance (e.g. Rosenthal and Strange
2003; Van Soest, Gerking, and Van Oort 2006), while others find that agglomeration economies operate
at a much broader geographic scale, for example, the state or national level (e.g. Ciccone and Hall 1996;
Dekle and Eaton 1999). However, since these studies on the spatial extent of agglomeration benefits
apply different measures of agglomeration economies and different dependent variables to capture the
effects, we should not conclude here that the results contradict each other. Moreover, the construction of
agglomeration variables also determines the resultant geographic extent of agglomeration effects (Graham
and Melo 2011). The pros and cons of different measures of agglomeration economies, the resultant
effects, and their geographic extents are discussed in detail in section 3.
2.2 Theoretical basis of congestion costs (diseconomies)
One by-product of agglomeration is congestion, which arise as economic agents compete for
limited land, limited output markets, and limited fixed public infrastructure such as highways and
communication facilities. Other negative consequences of agglomeration such as environmental pollution,
noise, and high crime rates may also occur.
16
In general, congestion externalities originate from the difference between private and social costs.
The literature on optimal city size suggests that households and firms choose to locate in a city by
evaluating the prevailing average social costs (AC)—such as local taxes or commuting/transport costs in
the city, but ignoring the increased social costs their arrivals generate for the whole urban population
(Alonso 1971; Richardson 1972; Mills and De Ferranti 1971). Therefore, the average social cost would
usually be lower than the marginal social cost (MC) born by the city as a whole (Alonso 1971;
Richardson 1972; Mills and De Ferranti 1971). Firms and residents would continue to move into the city
as long as marginal benefits, as represented by marginal value of outputs (MP) in the city, exceed the
marginal social cost plus the opportunity cost of moving (MP > MC + opportunity cost for moving)
(Alonso 1971; Richardson 1972). The result is that cities would grow to a size that is more congested than
the optimal size for maximizing urban production, that size being the point where the difference between
the average costs and value of output (AP − AC) are maximized (Alonso 1971; Richardson 1972).
There are many sources of congestion externality. One of the most studied negative externalities
is traffic congestion, which also results from the difference between the marginal cost and social cost
associated with using the transportation infrastructure (Sullivan 1983). For example, a car commuter
entering a road network faces the average social cost of commuting (AC), not the marginal social cost of
commuting (MC), which takes into account the external costs his decision would have on other users
(Sullivan 1983). The basic idea is that roads and highways are specific types of public capital stock that
are not pure public goods; the service degrades as the number of users increases (Boarnet 1997; Fernald
1999; Montolio and Solé-Ollé 2009). It is also suggested that the travel costs per trip rise with the number
of users and the point at which the marginal cost exceeds the average cost (MC > AC) is reached even
when the use of the transport system exceeds the existing capacity by a small amount (Downs 1992; Mills
and Ferranti 1971; Walters 1961).
In addition to the congestion effects associated with heavy usage of roads and other public
infrastructure, there are other types of diseconomies in the form of registered market prices such as higher
17
rent and higher wages, i.e., the so-called "pecuniary diseconomies" (Richardson 1972). For example, in
the new economic geography (NEG) theory, the basic dispersion force that limits the size of an
agglomeration (or a city) is the competition for scarce land in the location, which generates high land
values that increase land input costs and diminish profit benefits for firms in spatial proximity (Venables
1996). Another important source of pecuniary diseconomies—high wages—can have different
implications for businesses and residents: for firms, higher wages mean higher costs for labor factor
inputs, while for individual workers higher wages imply higher compensations for negative externalities
(Richardson 1995). On the other hand, high land rents and wages, like productivity and employment,
reflect a high level of agglomeration benefits (Rosenthal and Strange 2004, 2147). For example, firms
benefitting from concentrating in larger cities would be more willing to pay for agglomeration benefits,
which would be reflected in higher office or industrial rents (Drennan and Kelly 2011). There is also
empirical evidence that spatial variation in wages reflects variation in labor productivity and that wage
levels increase with the scale or density of cities or regions (Fingleton 2000, 2006) or with market and
supplier access across regions (e.g. Mion 2004; Rice, Venables, and Patacchini 2006). Thus, like the costs
of traffic congestion, the negative effects of wages and land rents might not be evident until reaching a
threshold beyond which further growth of cities or regions would be inhibited.
The distinction between pecuniary diseconomies and other congestion externalities, however,
may not be that straightforward. There is some evidence that non-market congestion costs or disamenities
are reflected in factor prices. For example, some empirical studies examine the extent to which spatial
variation of local traffic density, air pollution, and other disamenities within a metropolitan area can be
reflected by internal variation in residential property values within a metropolitan area (e.g. Chay and
Greenstone 1998; Harrison Jr and Rubinfeld 1978; Hughes and Sirmans 1992).
3
For the city as a whole,
total land rents measures net benefits of agglomeration, not total congestion costs, because land values are
capitalized values of total rents—the difference between the total costs and total value of output (Edel

3
The theoretical underpinning of how costs of traffic congestion affect spatial variation of rents and housing prices
within a metropolitan area is discussed in Chapter 4.
18
1971). Edel (1971) also suggests that land rents do not perfectly measure net benefits of agglomeration
because not all gains and costs are capitalized into land markets. As suggested by Richardson (1995),
wage differentials are a better proxy than rent differentials for reflecting differences in disamenities at the
inter-metropolitan scale. This is because workers are mobile but land is not, which invalidates the
existence of equilibrium distribution of land rents across cities (Richardson 1995). However, Richardson
(1995) also suggest that neither wages nor land rents are reliable measures of agglomeration economies
(or diseconomies) because they also reflect labor or land market conditions in addition to the presence of
net agglomeration benefits or diseconomies (Richardson 1995). Thus, we should be cautious about
interpreting wages or land rents as the capitalized effects of agglomeration economies or diseconomies.
2.3 Implications for metropolitan spatial structure
Like discussions on agglomeration economies, most discussions on the tradeoff between
agglomeration economies and congestion costs focus on the metropolitan scale. For example, the
literature on optimal city size evaluates agglomeration function and congestion cost function for an entire
city, while some empirical studies on urban productivity or growth also identify size-related or density-
related negative externalities for metropolitan areas (e.g. Moomaw 1985; Sveikauskas, Gowdy, and Funk
1988). However, there is also evidence that congestion costs are not evenly distributed throughout a
metropolitan area. For example, the level of traffic congestion varies across different routes and road
segments in the transportation network, as well as across different destinations connected by the transport
system given their popularity (Mondschein, Taylor, and Brumbaugh 2011). As suggested by Richardson
(1995), the tradeoff between agglomeration economies and congestion costs could also work at an intra-
metropolitan scale. Specifically, spatial variation of agglomeration economies and congestion costs
determines the general urban form to be either polycentric or dispersed that
"If agglomeration economies have a narrow spatial ambit and cluster at specific (subcenter)
locations and offset density-related congestion costs at these locations, then the equilibrium
distribution of economic activity and population is a polycentric pattern characterized by several
subcenter peaks. If, on the other hand, congestion costs offset agglomeration economies
uniformly over geographical space, we end up with a situation where each location is, more or
19
less, equally efficient. The results should be a very dispersed pattern in terms of both economic
activity and population, the 'dispersed metropolis' rather than the polycentric metropolitan
structure. (Richardson 1995, 134)"
In other words, the geographic extent at which agglomeration economies work and trade off with
congestion costs determines the spatial clustering or dispersion pattern of economic activities within an
urban context. Moreover, employment centers within a metropolitan area are "analogous to the system of
cities in a larger regional economy" (Anas, Arnott, and Small 1998, 1427) and can have different sizes
and industrial compositions. This means that the strength of agglomeration economies differ across
employment centers with different industrial specialization characteristics and solve with congestion costs
at different levels within a metropolitan area (e.g. Anderson and Bogart 2001; Giuliano and Small 1991;
Giuliano et al. 2007).
On the other hand, the micro-economic foundation for the decentralization and formation of
subcenters within a metropolitan area can be explored through the location behavior of firms and
residents. For example, some theoretical models explain the transformation from monocentricity to
polycentricity as a result of aggregate locational decisions of individual firms and households to
maximize their own profit or utility levels in response to the increasing congestion costs as the city size
gets larger (e.g. Mun and Yoshikawa 1993; Wheaton 2004).
4

Although we have plenty of evidence that employment activities are clustered in a few centers
rather than distributed evenly throughout a metropolitan area, the tradeoff between congestion costs and
agglomeration benefits are not often examined directly, due to the difficulty in operationalizing the two
forces (Richardson 1995). As implied by Richardson (1995), in order to better identify agglomeration
economies and diseconomies and their roles in urban form, an attractive way is to observe the micro
behavior of economic agents and incorporate individual heterogeneity into the empirical analysis. After
all, firms and households are the actual economic units that respond to congestion costs by choosing

4
The theoretical underpinning of how congestion affects individual firms and households is discussed in detail in
Chapter 3 and Chapter 4, respectively.
20
where to locate and where to travel. Thus, the latter two empirical chapters would try to fill in the gap in
the current empirical research on agglomeration and urban form by following this logic.
3 Do economies of proximity matter for firm location decisions?
Firm location choice is of interest in many different fields of study. It has important implications
for regional economic development because the attraction of new businesses or new investments means
new job opportunities as well as new tax bases for a region. At a much finer geographic scale, predicting
firm location is also important for land use and transportation planning, because new jobs may generate
more travel demands and flows, and the placement of new jobs would influence the spatial distribution of
economic opportunities and alter the intra-urban travel pattern (Franklin and Waddell 2003). Despite a
wide difference in study purpose, spatial units and data chosen, and econometric techniques applied,
agglomeration effects are usually considered an important determinant of firm location and are the most
studied factor in empirical studies of firm location choice (Arauzo-Carod, Liviano-Solis, and Manjón-
Antolín 2010). Thus, studies of firm location choices provide a good example for examining some
important remaining issues in agglomeration studies, e.g. the geographic extent of agglomeration
economies and the spatial decay mechanism of the effects. Different measurements of agglomeration
economies applied by researchers of different fields also allow us to examine whether the results of
agglomeration effects are sensitive to the construction of the variable.
This section reviews empirical studies across various fields that focus on the determinants of firm
location choice and the role that agglomeration economies play. In doing so, I draw together related
literature on urban economics and regional science, urban/economic geography, and transportation. Two
fundamental questions are addressed in this study. First, despite the difference in the measures of
agglomeration economies, the use of spatial units, types of data and estimation methods across studies, is
there a consensus regarding the positive role agglomeration economies play in firm location choices? Or
is there any evidence that different measures of agglomeration yield different results, when factors such as
geographic scales and industrial sectors are controlled? Second, could we find a consistent geographic
21
range of agglomeration effects on firm location decisions? Or, do different sources of agglomeration
economies have different geographic extent? I also review and discuss firm characteristics and other
locational variables controlled in different studies.
In the following, I first review the theoretical expectation of agglomeration effect on firm location
choices, different measures of agglomeration widely used in empirical studies, and the rationale behind
the construction of each measure. Then I summarize the various measures of agglomeration economies
and their estimated effects on firm location, grouped by the geographic scale of location alternatives
specified for firms' decisions. Other information contained in each study, including study area, study
period, choice of dataset, and estimation method, are also exhibited. This section concludes with the main
findings from the existing literature and discusses possible improvements for developing more cohesive
measures of agglomeration economies.
3.1 Measures of agglomeration economies
As discussed, individual firms would tend to agglomeration because of the benefits made possible
by spatial proximity between economic agents. The existence of agglomeration economies in a location
would attract more firms to locate there, which would in turn increase the agglomeration benefits. Similar
to the urban growth studies, most studies in the firm location choice literature also distinguish between
localization and urbanization economies, i.e., whether it is spatial proximity to other firms within a
particular industry that attracts more firms or the proximity to a large concentration of diverse economic
activities that matters. Moreover, instead of evaluating the two "catch-all" agglomeration effects, some
studies also distinguish between different sources of agglomeration effects such as access to suppliers,
laborers, or markets (e.g. De Bok and Sanders 2005; Shukla and Waddell 1991).
Another source of differences across studies is the measurement of agglomeration economies. As
suggested by Graham and Melo (2011), the construction of agglomeration variables is closely related to
the understanding of the spatial extent of agglomeration benefits. In firm location studies, the definition of
22
geographic range for agglomeration effects is also related to the spatial dimension of firm location
decisions each study focuses on, i.e., whether the location choices are at the intra-metropolitan scale or
the inter-metropolitan scale. Measures of agglomeration economies also differ in many aspects, such as
whether spillovers of agglomeration economies from neighboring locations are allowed for and whether
the dimension of transportation is included. The following summarizes different measures of
agglomeration economies mostly used in, but not limited to, firm location studies.
3.1.1 Distance to Central Business District (CBD)/central city
The construction of this measure derives from the monocentric model in standard urban economic
modeling: assuming that the central location of a city or region is the only export node and that workers
are distributed evenly throughout the city or region, the central location would have a natural advantage in
terms of maximum accessibility to workers and customers (Alonso 1964). As suggested by NEG theory,
the locational advantage of a central location would generate agglomeration economies in a cumulative
causation process; otherwise, the higher rents and associated costs at the location would dampen the
locational advantage and drive economic activities to decentralize (Fujita and Mori 2005). Historically, a
CBD in a city (or a central city in a region) preserves the advantages in terms of the quantity and quality
of service and infrastructure provided, as well as the rich business and market linkages already formed,
which are closely associated with general urbanization effects. Today, the need for face-to-face contact is
one of the reasons that central locations are still valued by firms of some specific functions in some
sectors (such as those high-order office firms or headquarters) (Coffey, Drolet, and Polèse 1996). Thus,
the measure of proximity to central locations, using either straight line distance or travel times, reflects
not only the decentralization pattern of firms, but also to what extent firms value the benefits associated
with central agglomerations (Coffey, Drolet, and Polèse 1996). A positive coefficient indicates the
diminished attraction of urbanization economies associated with a central location, as well as the
decentralization trends of firms.
23
One of the main advantages of this measure is that the decay gradient of agglomeration
economies (Graham and Melo 2011) would be estimated directly. In other words, the coefficients for this
measure directly indicate the extent to which the influence of CBD or central city declines with distance
or travel time. However, by defining central locations as the only common reference points (Vickerman
1974), this measure assumes that only agglomeration benefits within the central location are valued by
firms, while potential benefits of agglomeration at other locations are ignored. Similar to the
agglomeration benefits of CBD (or central city), the existence of polycentric structures at both the urban
and regional scale have been proved in theoretical and empirical studies (e.g. Giuliano and Small 1991;
Kloosterman and Musterd 2001). To take into account the influence of subcenters or secondary cities,
proximity to these locations may also be included in the estimation of agglomeration effects. This,
however, may fail to account for the difference in the scale of agglomeration economies at each reference
point. Moreover, the inclusion of multiple proximity variables referring to different attractors might
generate multicollinearity problems in empirical studies (e.g. Heikkila et al. 1989; McDonald and Prather
1994).
3.1.2 Scale/density measures
The total scale of concentration, such as urban or industrial size, is a direct measure of
agglomeration economies in empirical studies. This measure is consistent with the theory that the
potential benefits of agglomeration economies are derived from the spatial concentration of economic
activities. As suggested by Ciccone and Hall (1996), the density of economic activity is a better measure
of spatial proximity between firms than the size measure because the scale of the concentration is
normalized with the physical area. In empirical studies, size and density are the most commonly used
measures of all types and sources of agglomeration effects. For example, the aggregate employment size
or density is usually used as a proxy for urbanization economies (e.g. Ciccone and Hall 1996; Fogarty and
Garofalo 1988; Malmberg, Malmberg, and Lundequist 2000; Moomaw 1985), while the industrial
employment size or density is used as a proxy for localization economies (e.g. Carlino 1979; Henderson
24
1986; Henderson, Kuncoro, and Turner 1995; Henderson and Mitra 1996). Population size or density is
usually used as a proxy for access to the labor market or final market demands (e.g. Arauzo-Carod and
Viladecans-Marsal 2009; Erickson and Wasylenko 1980; Melo, Graham, and Noland 2010; Wennberg
and Lindqvist 2010).
One obvious drawback with this measure, however, is the arbitrary definition of the geographic
range of agglomeration economies. Consistent with the various spatial dimensions focused on by different
firm location studies, the defined spatial scale of agglomeration economies also varies widely, ranging
from census tracts and counties to metropolitan areas or even states. This definition assumes that only
proximity to opportunities within the predefined unit matters and those opportunities outside the boundary
play no role. However, this assumption could be problematic because spatial interaction is not confined
by statistical or administrative units. Moreover, this measure does not account for the possible spatial
decay effects of agglomeration within the predefined units, which would be problematic when the spatial
units are defined at a large geographic scale, such as metropolitan areas, where the internal relative
proximity between economic agents is also important for urban economic efficiency (Prud'homme and
Lee 1999).
3.1.3 Cumulative opportunities
The cumulative opportunities measure of agglomeration economies usually sums up the size of
economic activities within a pre-defined distance or travel time threshold. The measure basically assumes
that the agglomeration effect is only realized over a certain distance or time; beyond that threshold spatial
interaction is far less efficient and the potential benefits of agglomeration cease to exist. As implied by
Handy and Niemeier (1997), the cumulative measure of agglomeration economies emphasizes the scale of
potential opportunities available to firms or economic agents without discounting distance between them.
In empirical studies, different thresholds may be applied to measure different types or sources of
agglomeration economies, based on the assumption that their geographic ranges also differs. To account
25
for the spatial decay effects of agglomeration, some studies uses a set of distance/time bands to measure
whether those opportunities at more distant bands are less relevant for firms at the focal location. Some
studies argue that travel time, instead of physical distance, should be used to set up the threshold for
agglomeration measures because the spatial decline of interactions is more economically relevant with
travel time than with physical distance (Rice, Venables, and Patacchini 2006). This issue also occurs in
the construction of measures of agglomeration economies based on accessibility indices, as discussed in
the following.
3.1.4 Gravity-based accessibility measures
The gravity-based accessibility measure, first developed by Hansen (1959), is the classical way to
operationalize spatial proximity variables (Karlsson and Manduchi 2001). In transportation studies, this
measure has its roots in theories of travel behavior and is constructed based on the gravity model for trip
generation (Handy and Niemeier 1997). In general, the gravity-based accessibility measure can be
described as the total number of opportunities reachable from a focal location discounted by the travel
impedance between the location and all other reference locations (Bruinsma and Rietveld 1998). In
agglomeration studies, however, this measure is constructed to allow for spillovers of agglomeration
economies from neighboring locations and to explore the spatial decay pattern of agglomeration effects
(Melo, Graham, and Noland 2009). The general form of this agglomeration index is specified as follows
(e.g. Graham, Gibbons, and Martin. 2009; Melo, Graham, and Noland 2009):

∑ · ·
(2-1)
where represents the market size or economic mass in location j and f (
•
) is the impedance function
that puts weight on the influence of economic mass from j to location i, with those closer locations
assumed to have a higher weight and those farther locations applied with a lower weight. In the
impedance function,
ij
d is the measure of travel impedance between location i and location j and is the
impedance parameter that specified the decay rate of agglomeration economies.
26
The most common form of accessibility indices in agglomeration studies is the market potential
measure, which extends from scale measures of agglomeration. This type of measure usually uses the
simple inverse distance function with the rate of decay set as 1 without any theoretical basis (e.g. Graham
and Kim 2008; Graham and Van Dender 2011; Melo, Graham and Noland 2010). The negative
exponential function is also used in many studies, which implies a faster attenuation of agglomeration
effects. Various impedance parameters are also applied and are not subject to theoretical explanations in
the accessibility literature (see a detailed discussion in the following section).
As is the case when measuring cumulative opportunities, the measure of travel impedance
between locations is another important factor in the construction of an accessibility-type agglomeration
indice. As suggested by Graham and Melo (2011), the classic forms of market potential-type measures
differ from traditional accessibility measures in transportation studies, as the former usually uses straight-
line distance instead of road network travel time to measure the travel impedance between locations.
Straight-line distance, however, is a poor proxy for the relative closeness between agents not only becaus
e transportation infrastructure is unevenly distributed (Duschl et al. 2011; Graham and Melo 2011;
Moodysson and Jonsson 2007), but also because the actual travel impedance between locations is not
static but subject to changes depending on the utilization and congestion levels of transportation
infrastructure (Gilly and Wallet 2001). Graham's (2007) study also suggests that the use of straight-line
distance in the impedance function without any transportation dimension may produce downward biased
estimates of the agglomeration effects and "makes locations that suffer from congestion seem more
accessible than they really are" (108).
3.2 Empirical evidence: the geographic extent of agglomeration benefits
To demonstrate how agglomeration effects work differently at different spatial scales of firm
location choices, I find eight representative papers on intra-metropolitan firm location choices from
different fields of studies, i.e., urban/regional economics, urban/economic geography, and transportation.
These studies are chosen because they are the only few studies I could find that econometrically
27
investigate the driving forces behind the location decisions of firms within an urban context. Besides,
these studies also apply different measures of agglomeration economies. I also choose from a number of
empirical studies on inter-metropolitan (or regional) firm location eight other papers, which are also from
different fields of study and differ in terms of agglomeration measures and econometric methods. Table 2-
1a (in the appendix) illustrates the basic information of each chosen study, i.e., the data and methods used,
the definition of spatial units, the study area, the study period, and the focused economic sectors. Table 2-
2a (in the appendix) summarizes the estimated effects of agglomeration economies on firm location
decisions in different studies. The following discussion focuses on how different types of agglomeration
economies are measured in different studies and whether results vary based on which measure of
agglomeration is used.
3.2.1 Urbanization effects
In this review, the term "urbanization economies" refers to the agglomeration benefits derived
from the general scale and scope of economic activities. Various measures of urbanization economies
have been applied in firm location studies, such as distance to CBD/central location, size or density of
employment, and accessibility-type measures. Population density is also considered a measure of
urbanization economies because it includes "all kinds of regional influence, such as availability of
qualified labor, house prices, local demand, and the level of knowledge spillovers" (Audretsch and Fritsch
2002, 120, repeated in Arauzo-Carod and Viladecans-Marsal 2009, 554). The results show that
urbanization effects play a different role in intra-urban and inter-regional firm location decisions.
— Intra-metropolitan firm location
There is some evidence that the effects of urbanization are not always that important in intra-
metropolitan firm location decisions and their impacts vary across firms in different industrial sectors or
types. For example, using the distance to CBD measure, Erickson and Wasylenko (1980) find that only
firms in construction and wholesale trade sectors value urbanization benefits associated with Milwaukee's
28
CBD in their location choices. Shukla and Waddell (1991) indicate that only the service sector is
consistently influenced by the distance to both Dallas' and Fort Worth's CBDs. For other sectors, while
the influence of distance to Fort Worth's CBD is always significant, distance to Dallas' CBD shows
opposite effects, implying firms in most sectors tend to decentralize from the Dallas CBD. Arauzo-Carod
and Viladecans-Marsal (2009) identify that the influence of distance to CBD on the location of firms in
the manufacturing sectors within each Spanish municipality also differs across high-tech and low-tech
firms: while agglomeration benefits associated with the CBD consistently influence the location of high-
tech firms, their impacts on low-tech firms are not significant. For office firms, Ihlanfeldt and Raper
(1990) show that distance to CBD plays no role in the location of independent office firms and shows
opposite effects on the location of branch office firms in the Atlanta region. This result contradicts the
theoretical expectation that proximity to the CBD is most important for the location of office activities.
Using size or density measures of urbanization effects, empirical results on urbanization effects
also differ greatly across different types of firms. For example, Charney (1983) indicates that while high
employment density at the zipcode area level drives away manufacturing firms in general, it attracts
manufacturing firms in detailed sectors—durable and non-durable goods manufacturing sectors. Waddell
and Shukla (1993) show that the total size and employment density of employment centers reduce their
attractiveness to manufacturing firms. On the other hand, the effects of population density vary by
different study areas at different geographic scales. For example, Arauzo-Carod and Viladecans-Marsal
(2009) suggest a positive relationship between municipalities’ population density and their attractiveness
to manufacturing firms of all technology levels within each Spanish metropolitan area. There is also
evidence that at a much finer geographic scale (such as the zip code level or the census tract level), high
population density drives away manufacturing firms (e.g. Charney 1983; Maoh and Kanaroglou 2009).
One possible explanation for the negative role of local employment or population density at the finer
geographic scale is that higher density implies more intense competition between firms and higher land
29
rents, which would reduce the attractiveness of the locations to firms within a metropolitan area,
especially for manufacturing firms.
Some studies also use the accessibility-type urbanization indices, which capture the influence of
access to general economic activities without imposing a restriction on their spatial distribution. For
example, besides the distance to CBD measure, Ihlanfeldt and Raper (1990) also use the gravity-based
accessibility measure of urbanization economies, which is constructed as the sum of aggregate
employment reachable for each census tract discounted by the inverse of distance squared between census
tracts. However, their results indicate that office firms in the Atlanta region not only show aversion to the
central location, but also to those locations with advantages of accessing economic activities at the
regional scale. These results again contradict with the theoretical predictions.
— Inter-regional firm location
Similar measures of urbanization economies are applied in inter-regional firm location studies.
Population size or density is the most often used proxy (e.g. Devereux, Griffith, and Simpson 2007; Melo,
Graham, and Noland 2010), while distance to the central city (or capital city) is also used in some studies
(e.g. Guimaraes, Figueiredo, and Woodward 2000; Hansen 1987). Contrary to intra-metropolitan firm
location studies, most of the studies mentioned here show that urbanization economies have positive
impacts on the location decisions of firms in various sectors at the regional level. Some studies also show
that proximity to large or central cities exerts a larger influence on firm location decisions than local
employment density does, all else equal (Guimaraes, Figueiredo, and Woodward 2000).
3.2.2 Localization effects
Localization economies here are defined as the externalities arising from proximity to other firms
of the same sector. Similarly, various measures of localization economies have been applied and the most
frequently used measures include employment share or employment size of a particular sector within an
area. At the intra-metropolitan scale, more precise measures such as cumulative opportunity measures and
30
gravity-based accessibility indices are applied. The following discussion focuses on the geographic scale,
industrial sectors, and localization measures used in different studies.
— Intra-metropolitan firm location
Empirical studies on intra-metropolitan firm location show that unlike urbanization effects,
localization economies usually increase a location's attractiveness to firms in the same sector. For
example, Erickson and Wasylenko (1980) use each municipality's fraction of regional employment in
each particular 1-digit sector classified on a Standard Industrial Classification (SIC) basis as a simple
measurement of localization economies. Their results show that municipalities taking larger shares of
employment in each sector within the Milwaukee region are more attractive for relocated firms in the
same sector. Waddell and Shukla (1993) instead proxy localization economies in manufacturing by the
percentage of employment in the sector for each employment center and find positive evidence that
localization economies increase the probability of employment centers being chosen by new
manufacturing firms. Arauzo-Carod and Viladecans-Marsal (2009) use the total manufacturing
employment in each municipality as a simple measure of localization economies and find it significantly
increases municipalities' attractiveness for new manufacturing firms of all technology levels within each
metropolitan area.
Empirical studies applying more complicated measures also indicate positive effects of
localization economies. For example, cumulative opportunities measures are used by De Bok and Sanders
(2005) and Maoh and Kanaroglou (2009). The two studies focus on the intra-metropolitan location of
firms in 1-digit SIC sectors. In constructing the cumulative opportunities measures, the two studies use
different distance thresholds. Results of the two studies show that localization effects are significant for
most sectors, except for some larger firms (with more than 20 employees) in wholesale trade and retail
trade sectors.
31
The gravity-type accessibility indices are also applied by Shukla and Waddell (1991) in their
examination of the location choices of firms in various 1-digit SIC sectors within the Dallas-Fort Worth
region. Specifically, their study captures the localization economies emanating from all zones in the area
by the following function:
A
∑
(2-2)
where is the total employment for firms in three broad economic sectors —
manufacturing/construction/wholesale trade sectors, mining/transport/finance, insurance, and real estate
(FIRE) sectors, and retail trade/services sectors—at location j,
represents the straight-line distance
between location j and focal location i, and α is the decay parameter (= 0.75). Interestingly, their study
also applied this specification to measure market access but with a much flatter decay parameter (= 0.25),
implying that localization effects decay over a narrow spatial extent and at faster rates. Their empirical
results, however, indicate positive effects of localization economies on firms in all sectors except the
FIRE sector. This counter-intuitive result may be attributable to their broad categorization of industrial
groups that combines the FIRE sector with other un-related sectors such as mining, transportation, and
communication, which might dampen the true localization effects generated from the FIRE sector.
— Inter-regional firm location
Similarly, a simple measure of the employment for firms in a particular sector at each location is
mostly used in empirical studies and its positive effects are usually identified. For example, using this
simple measurement for each metropolitan area, Hansen (1987) finds significant and positive effects of
localization economies on the location choices of manufacturing firms in Brazil. Devereux et al. (2007)
find that the total employment in the manufacturing sectors at the county level increases each county’s
attractiveness to manufacturing firms in Great Britain. Head, Ries, and Swenson (1995) also find that the
32
total manufacturing employment for each state in the U.S. increases each state’s probably of being chosen
by new manufacturing firms with Japanese investors.
Other two measures, including the employment share or the location quotients in a particular
sector for a location and each location's fraction of regional employment in a sector, are also used in many
empirical studies (e.g. Gabe and Bell 2004; Guimaraes, Figueiredo, and Woodward 2000; Melo, Graham,
and Noland 2010). Although these two groups of measures both the relative size of sectoral employment,
the former emphasizes how specialized a location is in a particular sector, while the latter emphasizes
how concentrated a particular sector is in a location. Despite these differences in localization measures,
most empirical studies still show positive effects of localization economies. There is also some evidence
that spillovers of localization economies exist at a broader geographic scale. For example, Head, Ries and
Swenson (1995) construct a group of variables as the sum of employment in related manufacturing
activities in neighboring states that share borders with the focal states. Their results imply that localization
benefits may not be confined by the state border and that states neighboring other states with a large
concentration of firms in the manufacturing sectors are more likely to be chosen by new manufacturing
investments.
3.2.3 Inter-industrial linkages
As suggested by agglomeration theories, inter-industrial linkage is an important source of
agglomeration economies that is closely associated with cost savings in input-output flows and/or
information exchanges between firms located in spatial proximity. In empirical studies, the benefits from
proximity to firms in other sectors are sometimes evaluated separately from the overall urbanization
effects. However, most studies include measures of proximity to other sectors without exploring the
actual vertical linkages among the sectors. Thus, empirical results are mixed as to the effects of inter-
industrial linkages.

33
— Intra-metropolitan firm location
Similar to localization effects, the effects of inter-industrial linkages are measured in various
ways in different studies. The simplest measure is the total local employment in other sectors or the total
employment share in other sectors in a location. For example, Waddell and Shukla (1993) use the percent
of employment in the wholesale trade and the FIRE sectors within each employment center to examine
their influence on the location choices of manufacturing firms among employment centers within the
Dallas-Fort Worth region. Their results show that those centers concentrated with employment in the
wholesale trade sectors are more attractive to manufacturing firms, while those concentrated with FIRE
employment activities tend to drive away manufacturing firms.
More complicated measures are also applied. Specifically, the gravity-based accessibility indices
are applied in many empirical studies, with different specifications of impedance functions applied and
different results generated. For example, Ihlanfeldt and Raper (1990) specify the impedance function as
the inverse of distance squared in their measure of each census tract's access to suppliers of support
services and business amenities. Moreover, employment in financial, legal, business, and miscellaneous
services sectors are used to proxy for the availability of service suppliers, while employment in food and
drink industries are used to proxy for business amenities (Ihlanfeldt and Raper 1990). Their results show
that while independent offices value access to service suppliers and business amenities, branch offices do
not consider the location of service suppliers to be an important location determinant. Shukla and Waddell
(1991) use the same specification of impedance function for constructing inter-industrial linkage variables
as equation (1). However, their results indicate that access to other groups of sectors contributes
negatively to a location's probability of being chosen by firms in most sectors. Again, a possible
explanation is that their study divides industrial sectors into three broad categories without considering
the economic linkages between the industries. Thus, firms of different broad categories may be weakly
related, so that spatial proximity to each other does not generate positive externalities. Another example
of applying the gravity-based accessibility measure of inter-industrial linkages is the study by De Bok and
34
Sanders (2005), which computes business accessibility as the sum of employment of suppliers and
customers weighted by travel time. Their results find that access to upstream and downstream industries
significantly impacts the location choices of firms in business services but not in other sectors.
— Inter-regional firm location
Contrary to intra-metropolitan firm location studies that usually indicate the importance of local
linkages in the location of office and/or services firms only, there is some evidence that vertical linkages
at larger geographic scales are important for the location of manufacturing firms. For example, using the
number of establishments affiliated within the same vertically linked groups, Head, Ries, and Swenson
(1995) find that positive effects of vertical linkage exist within the states, but there are also spillovers
across the state borders.
Guimaraes, Figueiredo, and Woodward (2000) use simple measures of local employment shares
in the business and financial services sectors and find that they play positive roles in influencing the
location of new manufacturing firms. Holl (2004) applies cumulative opportunities measures and uses 30
minute as the travel time threshold. He distinguishes between the effects of access to intermediate
demands and access to suppliers. Using finer categories of manufacturing industries, the study indicates
that while access to intermediate demands reduces municipalities’ attractiveness to most categories of
manufacturing firms, access to suppliers increases municipalities’ attractiveness to all manufacturing
firms.
3.2.4 Labor force access/market access
Labor market access is another important source of agglomeration economies according to the
Marshallian tradition. On the other hand, market potential linkages are emphasized in the NEG theory,
which predicts that the input factor price (wages) are higher in locations which are spatially close to other
locations with larger market size and stronger purchasing power (Fujita and Mori 2005).
35
In empirical studies, accessibility-type indices are usually applied to measure access to workers,
which also reflects the home market effect because workers are also potential customers. The accessibility
indices may also reflect other urbanization benefits such as infrastructural and amenity advantages
(Shukla and Waddell 1991). However, I specifically discuss these effects separately in this review to
distinguish the role of labor force access and/or market access in firm location decisions. To examine the
spatial scale at which the effects of access to workers/markets are more relevant, the following discussion
is also categorized by the spatial scale of choice sets in firm location studies.
— Intra-metropolitan dimension
Using cumulative opportunities measures or gravity-based accessibility measures, the positive
effect of labor force access or market access on intra-metropolitan firm location is usually identified in
empirical studies. The specification of economic mass/market size and the specification of impedance
function, however, vary across different studies. Usually, the total population or the total number of
inhabitants/households is used as a proxy for both the market size and the scale of labor forces. For
example, Shukla and Waddell (1991) use the total population of each zipcode area as a proxy for market
size and specify the impedance function as equation (1). Their results identify positive and significant
effects of market access on the location choices of firm in all sectors within the Dallas-Fort Worth region.
However, the decay parameter specified for proximity to residential concentration is much smaller than
that for localization effects, implying a flatter decay pattern of this effect. De Bok and Sanders (2005) use
the inverse of travel time instead of straight-line distance to weigh the influence of distant inhabitants.
However, their study find that locations with better market access are only attractive to large retail firms
or large office firms with a high car dependency.
Corresponding to the standard market potential function in the NEG theory, some studies directly
use the household income level as a proxy for the purchasing power of each location, while other studies
use the relative size of lower-income households or racial groups as a proxy for negative market strength.
36
For example, besides proximity to residential concentration, Shukla and Waddell (1991) construct two
other variables — sum of median household income and the percentage of the black population weighted
by distance with a steeper decay coefficient. Their results imply that while proximity to high income
households increases locations' attractiveness to all firms except construction firms, proximity to the
black population acts as a disamenity only for firms in the FIRE sectors and services sectors.
To identify the importance of access to workers of particular skills, some studies use the
concentration of residents employed in relevant sectors or occupations as a proxy for the economic mass
measure in the construction of accessibility indices. For example, Erickson and Wasylenko's 1980 study
measures each municipality's advantage of access to skilled workers in each sector as the cumulative
number of residents employed in each sector reachable for each municipality within a median commuting
distance. Similarly, Ihlanfeldt and Raper (1990) measure accessibility to labor skills for each census tract
by the sum of workers employed in managerial and professional occupations as well as those in
administrative support occupations, discounted by the inverse of distance squared. Their results show that
while locations with better access to high skilled managerial workers are more attractive to office firms,
those with better access to low skilled clerical workers tend to detract office firms.
— Inter-regional dimension
The locational advantage of access to workers or markets is also stressed in inter-regional firm
location studies. However, empirical results are mixed regarding the effects of the accessibility advantage.
For example, McConnell and Schwab's 1990 study finds that each state’s market access measured by the
distance discounted sum of personal income in other states does not significantly increase its
attractiveness to new motor vehicle firms. Holl's 2004 study finds that the cumulative population of other
municipalities within the 30-minute travel time threshold of the focal municipality significantly reduces
the municipality’s attractiveness for most manufacturing firms. The positive influence of market potential
37
is also identified in the study by Melo, Graham, and Noland (2009), in which the accessibility variable is
simply specified as the distance weighted population in large cities.
3.3 Summary
This section reviews the effects of agglomeration economies on firm location choices based on
the empirical results of previous studies. Agglomeration economies are categorized into four types:
urbanization economies, localization, inter-industrial linkages, and labor force/market access. For each
type, empirical evidence on intra-metropolitan firm location studies and that on inter-metropolitan firm
location studies are reviewed and compared. The construction of different measures of agglomeration
economies is also reviewed and the resultant effects on firm location behavior are examined.
So how close is close enough? Based on the results of the review, the answer is not that
straightforward. Basically, different types of agglomeration economies seem to vary in their geographic
range of effectiveness. The following summarize the empirical evidence:
1) Localization effects seem to work well at either the intra-metropolitan or the inter-regional
scale. There is also evidence that the effects of localization economies can spill over across states.
2) Density-related urbanization measures show positive effects on firm location choices at the
inter-regional scale, but seem to drive away businesses at a much finer geographic scale (such as the
zipcode level), which might indicate the negative effects of dense local land use intensity and land
competition.
3) Inter-industrial linkages work at different geographic scales for different industries. At the
intra-metropolitan scale, the effect is mainly effective for the location of firms in business services or
other high-order office activities, while at the inter-metropolitan scale, this effect is usually effective for
firms in manufacturing sectors. This provides evidence that different types of industries require different
types of inter-firm linkages. For example, office activities are more likely to depend on face-to-face
38
interactions with customers and suppliers that are more sensitive to travel time and distance, while
manufacturing activities are more likely to keep linkages with upstream and downstream sectors through
physical goods flows that are less sensitive to transport costs at longer distance.
4) In terms of labor force/market access, there is some evidence that access to labor force and/or
markets work at either the intra-metropolitan or the inter-regional scale. However, some inter-
metropolitan studies show negative results.
For the second question: do different measures of agglomeration economies yield different results?
Based on the review, I do not find much evidence for that. However, the effects of urbanization
economies could be a special case. For example, if we consider market access as a measure of
urbanization economies, conflicting evidence may be found: while density-based measures usually have
negative impacts on intra-metropolitan firm location, accessibility indices usually show positive effects
on firm location at the intra-metropolitan scale. This could also imply that urbanization effects work at the
urban scale, rather than the finer spatial scale. Moreover, accessibility-type measures of agglomeration
economies still show some ability to capture the geographic extent at which agglomeration effects matter
and whether there are spatial spillover effects from neighboring locations' employment/population
concentration in general or in particular sectors.
One element that is important in agglomeration theories, but not so often included in the
construction of agglomeration measures, is the time dimension in measuring spatial proximity between
economic agents. As briefly discussed in the theoretical review, agglomeration economies today depend
more on the exchange of information and knowledge, which in turn depends on human interactions that
are time-sensitive and time-consuming in nature (Karlsson and Manduchi 2001) and are more likely to be
effective at a geographic scale smaller than a metropolitan area.
To better capture the agglomeration benefits at a finer geographic scale, especially those
associated with the flows of people and information, we need to include this dimension in agglomeration
39
measures. For example, compared with physical distance to CBD/central city, travel time to the central
locations might better capture the benefits of central agglomeration spilled over to firms in other locations.
The use of accessibility-type agglomeration measures may also facilitate the inclusion of time dimension
to examine how quickly the impact of economic mass decline with travel time.
Moreover, the inclusion of time dimension in agglomeration measures would also facilitate the
introduction of congestion effects. The reason is that relative proximity between economic agents is not
static but is subject to congestion levels on the transportation network (Torreand Gilly 2000). As a result,
the efficiency of spatial interactions between economic agents and the productivity advantages derived
from interactions are dependent on the real network travel time (Graham and Melo 2011). Thus, estimates
of agglomeration benefits based on travel time accounting for congestion levels of road network, instead
of physical distance, would better capture the actual level of agglomeration economies experienced by
firms. This is especially important for measuring agglomeration effects at the sub-metropolitan level
where congestion costs vary over space and time.
4 Does traffic congestion reduce economic efficiency?
As discussed earlier, traffic congestion is presumed to be an important source of diseconomies in
the agglomeration theories. In empirical studies, however, the question of how congestion costs impose a
drag on the economic efficiency of cities/regions is not fully explored. One important reason is that
congestion costs are endogenous to the economy—higher congestion costs are more likely to occur in
more agglomerated regions which are usually attractive and economically vibrant (Graham 2007; Sweet
2011). Thus, to measure the effects of congestion costs on the urban/regional economy, we should find
appropriate methods to solve the endogeneity problem.
The other possible reason is that "(w)e have not developed effective methodology for measuring
each individual item within each individual category of agglomeration economy or congestion cost in
dollar terms" (Richardson 1995, 132). In other words, it would be difficult to operationalize the two
40
offsetting forces. Other factor price inputs, such as wages and rents, are better measures for net
agglomeration benefits than diseconomies. To measure congestion effect, we should also develop
effective methods of measuring traffic congestion in economic terms.
Despite these difficulties, however, there are still a few studies that examine the negative effects
of diseconomies, especially traffic congestion, on the economic competitiveness of cities and regions. In
this section, I review the empirical evidence on the relationship between traffic congestion and economic
outcome and the methods applied in these studies for addressing the measurement issue and the
endogeneity issue. The purpose of this review is to address the other remaining questions mentioned
earlier: how we might measure congestion and how we might separate the costs from the benefits of
agglomeration. I also examine the differential effects of congestion on various industries.
In the following, I firstly review the conceptual framework on how traffic congestion affects the
economy and discusses different measurements of congestion effects. Then I discuss the econometric
methods and methodologies used in different studies and summarize estimated effects of traffic
congestion on the urban economy. This section concludes with the lessons learned from the empirical
studies and discusses implications for future research.
4.1 Modeling the effects of traffic congestion on the economy
Traffic congestion raises costs for firms and households and hence affects economic efficiency.
There are a few empirical studies concerned with the wider economic impacts of traffic congestion on
output levels, productivity growth, and employment growth. However, the modeling framework and the
measures of congestion costs differ across various studies. The geographic scale, unit of analysis, time
periods, and instruments used in different studies are summarized in Table 2-a3 (in the appendix).
In general, there are two ways to measure the effects of traffic congestion on the economy. One
way is to view traffic congestion as the efficiency and level of service provided by the transportation
infrastructure and directly enter the congestion variable into the production function with the
41
transportation infrastructure (as an unpaid input) and other inputs (e.g. Boarnet 1995, 1997; Fernald 1999;
Hymel 2009; Montolio and Solé-Ollé 2009; Sweet 2013). In other words, the effect of traffic congestion
on the economy is traced through its negative impacts on the efficiency of the transportation sector of the
economy. The other way is to consider congestion costs as by-products of agglomeration benefits and
identify their effects indirectly through their constraints on the accessibility advantages and agglomeration
benefits of particular locations. Specially, to detect the effects of traffic congestion, the estimated
marginal benefits of agglomeration variables constructed under congested conditions are usually
compared with estimates based on agglomeration measures constructed without accounting for traffic
congestion (Graham 2007; Hartgen, Fields, and Moore 2009).
4.1.1 Direct examination of congestion costs
— Modeling framework
One group of studies uses the production function to examine the effects of traffic congestion on
economic output. The general structure of production function used by Boarnet (1997), Montolio and
Solé-Ollé (2009), and Fernald (1999) includes both the stock of public infrastructure and the measures of
congestion levels. It takes the following form:

,
,
,
(2-3)
where is output in time t,
t
A is the level of technology, and ,
,
,
denote labor, capital, stock
of road infrastructure and congestion measures of the road capital stocks, respectively.
In Boarnet's 1997 study, road capital stock and congestion levels are modeled as shifting factors
such that
,

,
(2-4)
42
where g (•) is the shift factor. Using the Cobb-Douglas production function and the second-order translog
expansion of the shift factor, the functional form for estimating labor productivity is constructed as
ln

ln
ln
ln ln
ln
ln
ln   ∑

(2-5)
where the
denotes the year dummy variables (Boarnet 1997). The quadratic forms for and
allows the effect of transportation stock and congestion level to vary across the sample with their
respective levels. Moreover, the interaction term for and implies that more congested locations
would experience greater benefits from road and highway stock expansion (Boarnet 1997).
On the other hand, Fernald (1999) and Montolio and Solé-Ollé (2009) construct the standard
production function such that

,
, ,
(2-6)
where the level of technology ( ) is added and the effect of road infrastructure ( ) is introduced as a
function of road infrastructure stock ( ) and a part of the private capital stock ( ). Their models applied
a simple way to include the congestion effects as the effective stock of road infrastructure such that
̀
(2-7)
where is a measure of the utilization of road stock and the parameter α measures the rate at which road
services received by firms fall as the total number of users increases, with a value of 0 representing pure
public goods and a value of 1 representing pure private goods. This congestion model is termed as the B-
D model by Montolio and Solé-Ollé (2009). Compared to Boarnet's (1997) flexible approach, this
functional form allows the effects of congestion to be estimated simply through a log-linear specification.
43
However, the drawback of this specification is that the output elasticity of congestion is restricted to be
constant and is allowed to vary with the level of congestion.
The effects of congestion on employment growth are also examined in some empirical studies.
For example, Hymel (2009) and Sweet (2013) examine how aggregate employment growth is affected by
the initial congestion levels and other lagged locational characteristics. The basic model can be expressed
in the following form:
ln

ln
ln
ln
ln
ln (2-8)
where

represents the employment growth between year t and t −k, and X is a vector of exogenous
explanatory variables, such as human capital levels, industrial composition, crime levels, urban spatial
structure, municipal governance and weather or amenities (Hymel 2009; Sweet 2013). Similar to the
functional specification for equation (2-5), the estimated coefficients measures the elasticity of
employment growth with respect to the initial level of congestion. The quadratic term for the congestion
variable ( ) is also included by Hymel (2009) to examine the non-linear effects of traffic congestion.
— Measures of congestion levels
A variety of measures of traffic congestion are developed and evaluated in transportation studies
(e.g. Bertini 2006; Lomax et al. 1997). Instead of doing a complete review on different measures here, I
describe the congestion measures used for examining wider economic impacts and discuss the differences
in meaning of these measures.
Since congestion is viewed as a measure of the inefficiency of the transportation infrastructure,
some studies simply model congestion as the level of aggregate usage of road infrastructure such as total
vehicle miles/kilometers traveled (VMT/VKT), which captures the demand side of the transportation
sector (e.g. Fernald 1999; Montolio and Solé-Ollé 2009) . Since the supply side is controlled by the road
44
infrastructure variable, the effects of congestion are captured in this modeling (Montolio and Solé-Ollé
2009). Other studies directly apply the level of services measures of traffic congestion constructed by the
Texas Transportation Institute (TTI) for the urbanized portion of each Metropolitan Statistical Area
(MSA), such as the travel time delays per capita that only account for congestion-induced travel time
delays (e.g. Hymel 2009; Sweet 2013), or the average daily traffic (ADT) per freeway lane across the
entire network of each metropolitan area that approximate the volume-to-capacity ratios of the network
for each area (Sweet 2013). Boarnet (1995, 1997) constructed a congestion index (named ) that
enables congestion levels to be measured at geographic scales ranging from individual highway segments
to specific routes to county to regional. The basic element in the measure is the capacity
adequacy ( ) variable—the ratio of highway capacity to the peak-hour travel flow, which measures the
inverse of congestion at a particular point on the highway network (Boarnet 1995, 1997). To measure
congestion levels of each highway segment, the variable is aggregated and the sum is weighted by the
ratio of ADT at each point (mile marker) to the total ADT on the highway segment (Boarnet 1995, 1997).
Moreover, the highway segment congestion measure ( is summed into a county-level congestion
measure as the weighted sum of , weighted by the percent of total ADT each highway segment
takes in the total ADT within a county (Boarnet 1995, 1997). Thus, this measure accounts for the level of
road usage while controlling for the stock of highway infrastructure.
What is the difference between travel time delay and road usage measures of congestion? As
suggested by Sweet (2013), these two measures can have different policy implications: if reduced
economic efficiency is associated with congestion-induced travel time delays, policies aimed at
alleviating peak-hour congestion would be more likely to reduce the costs of congestion to the economy;
if, on the other hand, the negative effects of traffic congestion on economic efficiency is mainly through
over-capacity flow of traffic, policies should be directed towards reducing aggregate road usage, not just
peak-hour congestion delays, throughout the entire day.
45
4.1.2 Indirect examination of congestion effects
Another way to identify the effects of traffic congestion is to test how agglomeration benefits are
inhibited when congestion costs exist. Although some empirical studies in the urban growth literature
demonstrate the existence of diminishing returns on agglomeration economies by using the quadratic
form of agglomeration variables (e.g. Moomaw 1985; Sveikauskas, Gowdy, and Funk 1988), the negative
impact of traffic congestion on urban growth is not specified.
The accessibility-type measures of agglomeration with a transport-dimension allows for the
effects of traffic congestion to be verified and distinguish from other sources of diseconomies (Graham
2007). In empirical studies, the accessibility indices can be included as input factors of production
function such that
,,, , (2-6)
where X represents all the other control variables. The quadratic specification for is usually applied
to capture diminishing returns.
5
Accessibility measures are usually constructed in two ways, one
describing agglomeration economies achievable under congested condition and the other describing
agglomeration levels under free-flow condition. To identify congestion effects, two separate regressions
are usually using the two types of measures and the estimates associated with the agglomeration variables
compared. The logic behind this method is that if we believe that measures of accessibility under
congested condition better captures the actual level of agglomeration experienced by firms, measures of
accessibility under free flow condition would make those locations more accessible than they really are
and estimation using the latter measure would be biased (Graham 2007). Hypothesizing that the
difference in the two measures tend to be larger as the level of agglomeration increases, estimates based
on the accessibility measure without congestion would be downward biased. Thus, the measure of

5
The construction of the Acc variable is introduced in section 3 of this chapter.
46
proximity between locations in the impedance function of the variable is essential to the identification of
congestion effects.
For example, Graham (2007) estimates firm-level production functions using two measures of
accessibility separately. One accessibility-type measure of agglomeration is constructed based on straight-
line distance and the other measure is based on real road network travel time that accounts for the travel
speeds and congestion information. Consistent with the hypothesis, he finds that the real travel time based
measure of agglomeration grow at a slower rate and yields lower extreme values than the distance based
agglomeration measure do, while the gap between the two agglomeration measures diverge as
agglomeration levels increase. Comparing the estimation results of the two measures from the two
separate regressions, he finds that the elasticity estimates of accessibility based on the real travel time is
usually higher than the physical distance based estimates. This result supports the hypothesis that traffic
congestion reduces the otherwise higher levels of productivity by constraining the agglomeration benefits.
Moreover, negative and significant estimated coefficients on
are also found for some industrial
sectors, implying that these industries experiences diminishing returns on agglomeration economies
(Graham 2007). However, the nature of diminishing returns varies based on the two agglomeration
measures: when agglomeration measures with the presence of traffic congestion are used, less industries
show diminishing returns on agglomeration economies (Graham 2007).
Similarly, Hartgen, Fields, and Moore (2009) use free-flow travel time and congested travel time
to measure the internal accessibility of each metropolitan area, including accessibility to CBD, suburban
centers, airports, and other activity centers within each metropolitan area. They also regress labor
productivity on the two groups of accessibility measures separately for eight U.S. metropolitan areas in
four base years (1999, 2000, 2001, and 2005).
6
Their results predict that labor productivity would increase
by as much as 5 percent if traffic congestion were eliminated from the measures of accessibility to CBD,

6
The 8 metropolitan areas are Atlanta, Charlotte, Dallas, Detroit, Salt Lake City, San Francisco, and Seattle.
47
while the elasticity of labor productivity with respect to accessibility to suburban centers would increase
by as high as 30 percent if traffic congestion were eliminated.
In summary, this group of empirical studies considers traffic congestion as an important source of
agglomeration diseconomies and identifies its negative impacts on economic efficiency by comparing the
productivity-agglomeration curves estimated using different agglomeration measures. The empirical
results show that traffic congestion constrains the benefits of agglomeration and reduces its marginal
productivity. Thus, the results support that the mitigation of traffic congestion would mitigate diminishing
returns on agglomeration economies by improving accessibility advantages of urban areas. However, we
should still be cautious when predicting of production benefits of agglomeration from congestion
mitigation, since the costs of eliminating or alleviating traffic congestion within urban areas could also be
high but are not accounted for in these studies.
4.2 The endogeneity of traffic congestion to the economy
As discussed, empirical studies estimating the effects of traffic congestion on the urban/regional
economy are challenging because congestion and agglomeration effects are endogenously determined.
Economically vibrant places are more likely to be congested because greater demands for spatial
interactions between economic agents are more likely to be generated there (Mondschein, Taylor, and
Brumbaugh 2011). To measure the causal impacts of traffic congestion on economic growth, most studies
apply two-stage least squares (TSLS) or generalized method of moments (GMM) estimation techniques
and use different instruments for congestion levels. The instrumental variables are expected to be
correlated with traffic congestion but not causally related to economic outcome. The choice of
instrumental variables, therefore, is key to separating congestion costs from agglomeration benefits.
In Boarnet's (1997) study, the congestion level of each county is estimated as a function of the
network and demographic characteristics that affect driving patterns and congestion only. Specifically, the
proportion of fatal traffic accidents and the ratio between highway miles and total road miles are specified
48
as the chosen road network characteristics, which reflect geographic and highway network design
characteristics that influence traffic flows and travel speeds, while vehicles per capita is the chosen
demographic characteristic that reflects residents' driving patterns and propensity to drive (Boarnet 1997).
The instrument traffic accident variable and the vehicles per capita variable are found not to be causes or
results of county output through the vector-autoregression techniques, while their coefficients are not
found to be significant when directly included in the production function with other factor inputs.
Based on the same assumption, Montolio and Solé-Ollé (2009) choose two instrumental variables
for traffic congestion—the number of cars and the increase in cars per capita. They demonstrate that the
two variables are valid instruments that are uncorrelated with economic growth. They also suggest that
the instrumental variables could be either positively or negatively correlated with congestion: a large
number of cars on the roads might imply high levels of congestion due to heavy road usage, but a lower
number of cars might also imply that the level of congestion is so high that individuals and firms are
unwilling to own or use their private vehicles.
Historical variables are also used as instruments for congestion levels. For example, Hymel (2009)
use two instruments: 1) a measure of 1947 planned radial road-miles per capita interacted with a linear
time trend and 2) a measure of each metropolitan area's past congressional representatives assigned to the
House Transportation Committee, which is assumed to be positively correlated with prior transportation
infrastructure investments that would reduce subsequent congestion levels. Hymel (2009, 129) suggest
three reasons for justifying the use of the first instrumental variable: 1) radial highways are mainly
designed to provide inter-city access not intra-city access; 2) road-miles in center lines are less associated
with transportation infrastructure capacity than are lane-miles and are expected to be less associated with
future employment growth; and 3) the impacts of radial-road miles on employment growth would mainly
work through its impacts on congestion reducing firms' and households' travel time costs. Despite these
reasons, however, the validity of the distant highway plan instrument could still be questioned. For
example, those places historically vibrant would be expected to have more growth in the future and would
49
attract more road investments. The distinction in the functions between radial highway and beltways
would also be arbitrary. Even if such a distinction really matters in historical highway plans and radial
highways mainly improve intercity access, radial highways might still have a direct impact on urban
growth because, as discussed earlier, both intercity and intra-city access matter for the agglomeration
benefits experienced by firms within a region. Moreover, planned highways might serve as a proxy for the
stock of the current road infrastructure, which could be considered a direct input in the aggregate
production function. In other words, the impacts of planned highways on economic growth might not
need to rely upon their effects on congestion.
Similarly, Sweet (2013) applies instruments very similar to that used by Hymel (2009), including
the number of radial highways and beltways planned according to the U.S. interstate plan of 1955 and the
number of radial interstate highways planned according to the "Toll Roads and Free Roads" report of
1939. The chosen instrumental variables are sufficiently lagged to ensure their exogeneity to the
economic growth of metropolitan areas. However, the validity of these instruments has not been tested
and could be doubted for the same reasons discussed above for Hymel's 2009 study.
In summary, the choice of instrumental variables is essential to solve the endogeneity of
congestion. Although a variety of instruments are used in empirical studies, the validity of these
instruments could be subject to doubt. Given that transportation investments have been approved to
influence future urban growth, researchers should perhaps be more cautious regarding the use of historical
transportation plans as a proxy for congestion levels. Other socio-economic variables that reflect the
demand side of the transportation sector, such as vehicle ownership and stocks, might be better
instruments than variables that reflect transportation supply and capacity.
4.3 Empirical results
The estimation results of most studies support the prediction that traffic congestion places a drag
on otherwise high levels of economic efficiency in cities and regions, despite the differences in modeling
50
frameworks and methodologies used across various studies. Moreover, most studies also identify that the
effects of congestion is non-linear, which means that congestion costs may not impose a drag on the
economy until reaching a threshold beyond which further growth would be inhibited.
As discussed, the flexible model allows the effects of traffic congestion to be non-linear. The
regression results by Boarnet (1997) suggest that both the effects of (inverse of congestion) and
the effects of highway stock are quadratic when constant returns to scale are imposed in the production
function, while the interaction term between and highway capital is negatively significant,
implying that more congested areas will benefit less from highway investment. However when a ten-year
difference specification is used, neither the interaction term nor the quadratic term of highway stock is
significant. Boarnet (1997) interpret this result to mean that congestion reduction policies are not
"substitute policies" for highway stock increases.
Montolio and Solé-Olléb (2009) also use the flexible congestion model that includes the quadratic
terms for congestion levels and the stock of road infrastructures and the interaction term for congestion
and road stock. The estimated coefficients for congestion/utilization levels are found to be negatively
significant in their results. However, they also show that the flexible congestion model provides no more
information than the B-D model does. That is, the quadratic term of the congestion variable is not
statistically significant, while the interaction term is only significant when OLS regression is applied.
Hymel (2009), however, finds significant negative coefficients for the congestion variables and
the quadratic term of the congestion variables. Thus, instead of identifying a congestion threshold beyond
which further growth would be inhibited, his study suggests that traffic congestion will have stronger
negative impacts on the employment growth of more congested cities.
Sweet's 2013 study finds that higher levels of restrained travel capacity, measured by the ADT
variable, reduce employment growth rates and productivity growth per worker. Quadratic effects of
congestion are also detected in this study. Specifically, higher average daily traffic per lanes (ADT) is
51
associated with faster employment growth until reaching a threshold of 4.5 minutes of delay per one-way
commute (Sweet 2013). On the other hand, higher levels of peak-hour travel time delays measured by the
travel delay per capita appear to be associated only with decreasing employment growth, not productivity
growth (Sweet 2013). This result has a strong policy implication: since the aggregate road usage is a
stronger predictor of employment growth and productivity growth, it might be more important to alleviate
road usage in general instead of placing too much emphasis on peak-hour travel time delays (Sweet 2013).
However, this conclusion may require further examination given the possible problems in the research
design. For example, the inclusion of urban spatial structure variables in this study, such as the CBD
distance gradient and indices for job-housing balance, might have taken over the influence of peak-hour
travel time delays, because travel time delays would also affect the locational patterns of firms and
households by increase their travel costs. Without a more complete examination on the causal relationship
between congestion costs, urban form, and urban growth, policy implications based on this conclusion
may be doubted.
The effect of congestion is also found to be non-linear in empirical studies that evaluate the
effects of traffic congestion indirectly. For example, Graham (2007) finds that not only would the firm-
level productivity-urbanization curve shift outwards when congestion is removed, but the gap between the
two measures of market potential increases as the level of agglomeration benefits grows. In other words,
the constraints of traffic congestion on agglomeration benefits would become larger as the city size or the
level urbanization economies grow.
5 Conclusions and extensions
In this chapter, I focus mainly on three groups of remaining questions in agglomeration research:
1) at what geographic scale agglomeration economies/diseconomies matter; 2) how prior studies construct
measures for potential benefits of agglomeration and whether the resultant effects of agglomeration
economies differ due to the different measures used; 3) how prior studies measure congestion costs and
unravel the costs of congestion from the potential benefits of agglomeration. To address these questions, I
52
review the theoretical basis of agglomeration economies and congestion costs and summarize the results
of some empirical studies. Using studies on firm location choices as examples, I find that most empirical
studies indicate that agglomeration economies, as expected, positively affect the location of firms.
However, there is some evidence that different types of agglomeration economies work at different
geographic scales. For example, while localization economies usually play a positive and significant role
in affecting firms' location decisions at all geographic scales, density-related urbanization economies
usually play a positive role at a broader geographic scale, such as the urban scale. There is also some
evidence that different measures of urbanization economies can yield different results and that
accessibility-type measures better capture urbanization effects by allowing for spatial spillover effects
from neighboring locations.
Due to the possible limited availability of transportation network data, the time dimension is not
so often included in agglomeration measures in current studies. However, this dimension is important to
measure some types of agglomeration economies, such as those derived from information exchange and
human interactions that are more sensitive to actual travel time than physical distance. The use of
accessibility-type agglomeration measures with a time dimension would better estimate agglomeration
effects, especially those operating at smaller geographic scales.
Focusing on studies of the effects of traffic congestion on economic efficiency, I review the
modeling framework and methods used in empirical studies that disentangle congestion effects from
possible agglomeration effects. There are generally two ways to measure the negative effects of
congestion: first, the congestion levels on the transportation network may be entered directly into the
production function with other inputs to reflect their reduction on the efficiency of transportation input;
second, the costs of traffic congestion may be measured indirectly by examining their diminishing effects
on the accessibility advantage of locations. The second way to measure congestion effects depends on the
construction of accessibility-type agglomeration variables with a time dimension. Most studies use
instrumental variables in TSLS or GMM models to account for congestion's potential endogeneity in the
53
economy. Despite the differences in modeling framework, assumptions, methodology, as well as the
geographic scales used in various studies, most studies reveal that traffic congestion imposes a drag on
the urban/regional economy. There is also some evidence that the effect of traffic congestion is nonlinear
and that higher congestion levels are associated with faster economic growth until a particular congestion
threshold is met.
Reviewing the empirical literature on agglomeration economies and congestion costs, I find that
most of the existing literature focuses on the agglomeration benefits and/or congestion effects at the inter-
metropolitan or inter-regional level. There is little discussion regarding the trade-off between the two
effects at the sub-metropolitan level, even though congestion and agglomeration is obviously not uniform
within them. Thus, a major purpose of my study is to provide some direct empirical evidence on how
congestion costs affect the intra-metropolitan structure by offsetting agglomeration benefits and other
associated locational advantages. To examine this issue, I apply a behaviorist approach and look closely
at firms' and households' locational responses to congestion costs in chapter 3 and 4, respectively.
Chapter 2 References
Agarwal, Ajay, Genevieve Giuliano, and Christian L Redfearn. 2012. "Strangers in Our Midst: the Usefulness of
Exploring Polycentricity." The Annals of Regional Science 48 (2): 433–450.
Alonso, William. 1964. Location and Land Use: Towards a General Theory of Land Rent. Cambridge: Harvard
University Press.
Alonso, William. 1971. "The Economics of Urban Size." Papers in Regional Science 26 (1): 66–83.
Anas, Alex, Richard Arnott, and Kenneth A. Small. 1998. "Urban Spatial Structure." Journal of Economic
Literature 36 (3): 1426–1464.
Anas, Alex, and Ikki Kim. 1996. "General Equilibrium Models of Polycentric Urban Land Use with Endogenous
Congestion and Job Agglomeration." Journal of Urban Economics 40 (2): 232–256.
Anderson, Nathan B., and William 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.
Bertini, Robert L. 2006. "You are the traffic jam: an examination of congestion measures." Paper presented at the
85th annual meeting of the Transportation Research Board, Washington, D.C., January 22–26.
Boarnet, Marlon G. 1995. "The Economic Effects of Highway Congestion [online]." Berkeley: University of
California, University of California Transportation Center (UCTC), [cited 16 September 2013]. Available
from: https://escholarship.org/uc/item/6tk2m53k.pdf
54
Boarnet, Marlon G. 1997. "Infrastructure Services and the Productivity of Public Capital: the Case of Streets and
Highways." National Tax Journal 50 (1): 39–58.
Bruinsma, Frank, and Piet Rietveld. 1998. "The Accessibility of European Cities: Theoretical Framework and
Comparison of Approaches." Environment and Planning A 30 (3): 499–521.
Carlino, Gerald A. 1979. "Increasing Returns to Scale in Metropolitan Manufacturing." Journal of Regional Science
19 (3): 363–372.
Chay, Kenneth Y, and Michael Greenstone. Does Air quality matter? Evidence from the Housing Market.
Cambridge: National Bureau of Economic Research, 1998.
Ciccone, Antonio, and Robert E. Hall. 1996. "Productivity and the Density of Economic Activity." The American
Economic Review 86 (1):54–70.
Coffey, William J., Réjean Drolet, and Mario Polèse. 1996. "The Intra-metropolitan Location of High Order
Services: Patterns, Factors and Mobility in Montreal." Papers in Regional Science 75 (3): 293–323.
De Bok, Michiel, and Frank Sanders. 2005. "Firm Relocation and Accessibility of Locations Empirical Results from
the Netherlands." Paper presented at the 84th annual meeting of the Transportation Research Board,
Washington, D.C., January 9–13
Dekle, Robert, and Jonathan Eaton. 1999. "Agglomeration and Land Rents: Evidence from the Prefectures. Journal
of Urban Economics 46 (2): 200–214.
Downs, Anthony. 1992. Stuck in Traffic: Coping with Peak-Hour Traffic Congestion. Washington, D.C.: Brookings
Institution Press.
Drennan, Matthew P, and Hugh F Kelly. 2011. "Measuring Urban Agglomeration Economies with Office Rents."
Journal of Economic Geography 11 (3): 481–507.
Drucker, Joshua. 2012. "The Spatial Extent of Agglomeration Economies: Evidence from Three US Manufacturing
Industries." U.S. Census Bureau Center for Economic Studies, [cited 16 September 2013]. Available from:
http://dx.doi.org/10.2139/ssrn.1995507
Edel, Matthew. 1971. "Land Values and the Costs of Urban Congestion: Measurement and Distribution." Social
Science Information 10 (6): 7–36.
Fernald, John G. 1999. "Roads to Prosperity? Assessing the Link between Public Capital and Productivity."
American Economic Review 89 (3): 619–638.
Fingleton, Bernard. 2000. "Spatial Econometrics, Economic Geography, Dynamics and Equilibrium: a 'Third
Way'?" Environment and Planning A 32 (8): 1481–1498.
Fingleton, Bernard. 2006. The New Economic Geography versus Urban Economics: an Evaluation Using Local
Wage Rates in Great Britain." Oxford Economic Papers 58 (3): 501–530.
Folta, Timothy B, Arnold C Cooper, and Yoon-suk Baik. 2006. "Geographic Cluster Size and Firm Performance."
Journal of Business Venturing 21 (2): 217–242.
Franklin, Joel P, and Paul Waddell. 2003."A Hedonic Regression of Home Prices in King County, Washington,
Using Activity-Specific Accessibility Measures." Paper presented at presented at the 82nd annual meeting
of the Transportation Research Board 82nd Annual Meeting, Washington, D.C., January 12–16
Fujita, Masahisa, and Tomoya Mori. 2005. "Transport Development and the Evolution of Economic Geography."
Portuguese Economic Journal 4 (2): 129–156.
55
Gilly, Jean-Pierre, and Frédéric Wallet. 2001. "Forms of Proximity, Local Governance and the Dynamics of Local
Economic Spaces: The Case of Industrial Conversion Processes." International Journal of Urban and
Regional Research 25 (3): 553–570.
Giuliano, Genevieve, Christian Redfearn, Ajay Agarwal, Chen Li, and Duan Zhuang. 2007. "Employment
Concentrations in Los Angeles, 1980 – 2000." Environment and Planning A 39 (12): 2935–2957.
Giuliano, Genevieve, and Kenneth A. Small. 1991. "Subcenters in the Los Angeles Region." Regional Science and
Urban Economics 21 (2): 163–182.
Goodwin, Phil. "The Economic Costs of Road Traffic Congestion [online]." London: University College London,
Rail Freight Group, [cited 16 September 2013]. Available from: http://eprints.ucl.ac.uk/1259
Gordon, Ian R, and Philip McCann. 2000. "Industrial Clusters: Complexes, Agglomeration and/or Social
Networks?" Urban studies 37 (3): 513–532.
Graham, Daniel J. 2007. "Variable Returns to Agglomeration and the Effect of Road Traffic Congestion." Journal of
Urban Economics 62 (1): 103–120.
Graham, Daniel J, and Patricia C Melo. 2011. "Assessment of Wider Economic Impacts of High-Speed Rail for
Great Britain." Transportation Research Record: Journal of the Transportation Research Board 2261 (1):
15–24.
Grek, Jenny, Charlie Karlsson, and Johan Klaesson. 2009. "Market Potential and New Firm Formation [online]."
Stockholm: The Royal Institute of Technology, Center of Excellence for Science and Innovation Studies
(CESIS), [cited 16 September 2013]. Available from: https://static.sys.kth.se/itm/wp/cesis/cesiswp 202.pdf
Handy, Susan L, and Debbie A Niemeier. 1997. "Measuring Accessibility: an Exploration of Issues and
Alternatives." Environment and Planning A 29 (7): 1175–1194.
Hansen, Walter G. 1959. "How Accessibility Shapes Land Use." Journal of the American Institute of Planners 25
(2): 73–76.
Harrison Jr, David, and Daniel L Rubinfeld. 1978. "Hedonic Housing Prices and the Demand for Clean Air."
Journal of Environmental Economics and Management 5 (1): 81–102.
Hartgen, David T, M Gregory Fields, and Adrian T Moore. 2009. "Gridlock and Growth: the Effect of Traffic
Congestion on Regional Economic Performance (Policy Study 371) [online]." Los Angeles: Reason
Foundation, [cited 16 September 2013]. Available from:
http://reason.org/files/ps371_growth_gridlock_cities_full_study.pdf
Heikkila, E., P. Gordon, J. I. Kim, R. B. Peiser, H. W. Richardson, and D. Dale-Johnson. 1989. "What Happened to
the CBD-Distance Gradient?: Land Values in a Policentric City. " Environment and Planning A 21
(2):221–232.
Henderson, J. Vernon. 1986. "Efficiency of Resource Usage and City Size." Journal of Urban Economics 19 (1):
47–70.
Henderson, Vernon, Ari Kuncoro, and Matt Turner. 1995. "Industrial Development in Cities." The Journal of
Political Economy 103 (5): 1067–1090.
Henderson, Vernon, and Arindam Mitra. 1996. "The New Urban Landscape: Developers and Edge Cities." Regional
Science and Urban Economics 26 (6): 613–643.
Hughes, William T., and C. F. Sirmans. 1992. "Traffic Externalities and Single-Family House Prices." Journal of
Regional Science 32 (4): 487–500.
56
Hymel, Kent. 2009. "Does Traffic Congestion Reduce Employment Growth?" Journal of Urban Economics 65 (2):
127–135.
Ihlanfeldt, Keith R, and Michael D Raper. 1990. "The Intra-metropolitan Location of New Office Firms." Land
Economics 66 (2): 182–198.
Karlsson, Charlie, and Agostino Manduchi. 2001. Knowledge Spillovers in a Spatial Context: A Critical Review and
Assessment. In Knowledge, Complexity and Innovation Systems, edited by Manfred M. Fischer, Josef
Fröhlich, 101–123. Berlin-Heidelberg: Springer.
Kloosterman, Robert C., and Sako Musterd. 2001. "The Polycentric Urban Region: Towards a Research Agenda."
Urban Studies 38 (4): 623–633.
Lomax, Tim, David Schrank, and Bill Eisele. 2012. 2012 Annual Urban Mobility Report. College Station: Texas
A&M Transportation Institute.
Lomax, Tim, Shawn Turner, Gordon Shunk, Herbert S. Levinson, Richard H. Pratt, Paul N. Bay, and G. Bruce
Douglas. 1997. Quantifying Congestion. Volume 1: Final Report (National Cooperative Highway Research
Program Report 398). Washington, D.C.: National Academy Press.
Marshall, Alfred (1920), Principles of Economics. London: MacMillan.
McDonald, John F, and Paul J Prather. 1994. "Suburban Employment Centres: the Case of Chicago." Urban Studies
31 (2): 201–218.
Melo, Patricia C, Daniel J Graham, and Robert B Noland. 2009. "A Meta-Analysis of Estimates of Urban
Agglomeration Economies." Regional Science and Urban Economics 39 (3): 332–342.
Mills, Edwin S., and David M. de Ferranti. 1971. "Market Choices and Optimum City Size." The American
Economic Review 61 (2): 340–345.
Mion, Giordano. 2004. "Spatial Externalities and Empirical Analysis: the Case of Italy." Journal of Urban
Economics 56 (1): 97–118.
Mondschein, A, Brian D. Taylor, and Stephen Brumbaugh. 2011. "Congestion and Accessibility: What's the
Relationship? [online]." Los Angeles: University of California, University of California Transportation
Center (UCTC), [cited 16 September 2013]. Available from:
https://escholarship.org/uc/item/6bh2n9wx.pdf
Montgomery, Mark R. 1988. "How Large Is Too Large? Implications of the City Size Literature for Population
Policy and Research." Economic Development and Cultural Change 36 (4): 691–720.
Montolio, Daniel, and Albert Solé-Ollé. 2009. "Road Investment and Regional Productivity Growth: the Effects of
Vehicle Intensity and Congestion." Papers in Regional Science 88 (1): 99–118.
Moomaw, Ronald L. 1985. "Firm Location and City Size: Reduced Productivity Advantages as a Factor in the
Decline of Manufacturing in Urban Areas." Journal of Urban Economics 17 (1): 73–89.
Mun, Se-il, and Kazuhiro Yoshikawa. 1993. "Communication among Firms, Traffic Congestion and Office
Agglomeration." The Annals of Regional Science 27 (1): 61–77.
Overman, Henry G, and Diego Puga. 2010. "Labor Pooling as a Source of Agglomeration: An Empirical
Investigation." In Agglomeration Economics, edited by Edward L. Glaeser, 133–150. Chicago: University
of Chicago Press.
Parr, John B. 2002. "Missing Elements in the Analysis of Agglomeration Economies." International Regional
Science Review 25 (2): 151–168.
57
Parr, John B. 2004. "Economies of Scope and Economies of Agglomeration: The Goldstein-Gronberg Contribution
Revisited." The Annals of Regional Science 38 (1): 1–11.
Prud'homme, Remy, and Chang-Woon Lee. 1999. "Size, Sprawl, Speed and the Efficiency of Cities." Urban Studies
36 (11): 1849–1858.
Torre, Andre, and Alain Rallet. 2005. "Proximity and localization." Regional studies 39 (1): 47–59.
Rice, Patricia, Anthony J. Venables, and Eleonora Patacchini. 2006. "Spatial Determinants of Productivity: Analysis
for the Regions of Great Britain." Regional Science and Urban Economics 36 (6): 727–752.
Richardson, Harry W. 1972. "Optimality in City Size, Systems of Cities and Urban Policy: a Sceptic's View." Urban
Studies 9 (1): 29–48.
Richardson, Harry W. 1995."Economies and Diseconomies of Agglomeration." In Urban Agglomeration and
Economic Growth, edited by Herbert Giersch, 123–155. Berlin-Heidelberg: Springer.
Rigby, David L., and Jürgen Essletzbichler. 2002. Agglomeration economies and productivity differences in US
cities. Journal of Economic Geography 2 (4):407–432.
Rosenthal, Stuart S., and William C. Strange. 2001. "The Determinants of Agglomeration." Journal of Urban
Economics 50 (2): 191–229.
Rosenthal, Stuart S, and William C Strange. 2003. "Geography, Industrial Organization, and Agglomeration."
Review of Economics and Statistics 85 (2): 377–393.
Rosenthal, Stuart S, and William C Strange. 2004. "Evidence on the Nature and Sources of Agglomeration
Economies." In Handbook of Regional and Urban Economics, Volume 4, edited by J. Vernon Henderson
and Jacques-François Thisse, 2119–2171. Amsterdam: ELSEVIER.
Scott, Allen John. 1988. Metropolis: from the Division of Labor to Urban Form. Berkeley and Los Angeles:
University of California Press.
Shukla, Vibhooti, and Paul Waddell. 1991. "Firm Location and Land Use in Discrete Urban Space: a Study of the
Spatial Structure of Dallas-Fort Worth." Regional Science and Urban Economics 21 (2): 225–253.
Storper, Michael, and Anthony J Venables. 2004. "Buzz: Face-to-Face Contact and the Urban Economy." Journal of
Economic Geography 4 (4): 351–370.
Sullivan, Arthur M. 1983. "The General Equilibrium Effects of Congestion Externalities." Journal of Urban
Economics 14 (1): 80–104.
Sveikauskas, Leo, John Gowdy, and Michael Funk. 1988. "Urban Productivity: City Size or Industry Size." Journal
of Regional Science 28 (2): 185–202.
Sweet, Matthias. 2011. "Does Traffic Congestion Slow the Economy?" Journal of Planning Literature 26 (4): 391–
404.
Sweet, Matthias. 2013. "Traffic Congestion's Economic Impacts: Evidence from US Metropolitan Regions." Urban
Studies 51 (10): 2088–2110.
Taylor, Brian D. 2002. "Rethinking Traffic Congestion." ACCESS Magazine 1 (21): 8–16.
Van Soest, Daan P., Shelby Gerking, and Frank G. Van Oort. 2006. "Spatial Impacts of Agglomeration
Externalities." Journal of Regional Science 46 (5): 881–899.
Venables, Anthony J. 1996. "Equilibrium Locations of Vertically Linked Industries." International Economic
Review 37 (2): 341–359.
58
Vickerman, Roger W. 1974. "Accessibility, Attraction, and Potential: a Review of Some Concepts and Their Use in
Determining Mobility." Environment and Planning A 6 (6): 675–691.
Waddell, Paul, and Vibhooti Shukld. 1993. "Manufacturing Location in a Polycentric Urban Area: a Study in the
Composition and Attractiveness of Employment Subcenters. Urban Geography 14 (3): 277–296.
Walters, Alan A. 1961. "The Theory and Measurement of Private and Social Cost of Highway Congestion."
Econometrica 29 (4): 676–699.
Wheaton, William C. 2004. "Commuting, Congestion, and Employment Dispersal in Cities with Mixed Land Use."
Journal of Urban Economics 55(3): 417–438.

59
Appendix for Chapter 2
Table 2-a1 Basic information of selected empirical studies on firm location choice
Author Journal Study Area
Choice
sets/Alternatives
locations
Study
period
Data source Model using
Intra-metropolitan firm location studies
Erickson and
Wasylenko
(1980)
Journal of Urban
Economics
Milwaukee
region
Municipalities
1964–
1974
Unemployment compensation
records, from Wisconsin
Department of Industry, Labor and
Human Relations
Logistic
Regression
Charney
(1983)
Journal of Urban
Economics
3-County
metropolitan
area of Detroit
Zipcode zones
1970–
1975
Manufacturing dictionary of
Michigan
OLS
Ihlanfeldt
and Raper
(1990)
Land Economics Atlanta region Census tracts 1980
Atlanta Regional Commission: 1980
employment survey
Tobit Model
Shukla and
Waddell
(1991)
Regional Science and
Urban Economics
Dallas-Fort
Worth region
Zipcode zones
1985–
1987
County Business Patterns by
Zipcode
Conditional Logit
Model
Waddell and
Shukla
(1993)
Urban geography
Dallas-Fort
Worth region
Employment centers
1985–
1987
County Business Patterns by
Zipcode
Multinomial
Logit/Conditional
Logit Model
Coffey,
Drolet, and
Polèse
(1996)
Papers in regional
science
Montreal
Census
Metropolitan
Area
CBD/ rest of the
metro area;
CBD/rest of
city/inner
suburb/outer suburb
1992–
1993
A detailed survey of 324 high order
service establishments
Logistic
Regression
60
De Bok and
Sander
(2005)
Journal of the
Transportation
Research Board
3 LISA
registration
areas in the
Netherlands
Real estate objects
1988–
1997
WMD -dataset
Conditional Logit
Model
Arauzo-
Carod and
Viladecans-
Marsal
(2009)
Regional Science Spain
Municipalities inside
13 biggest
metropolitan areas in
Spain
1992–
1996
REI (Spanish Industrial
Establishments Register)
Poisson/Negative
Binomial Model
Maoh and
Kanaroglou
(2009)
Journal of the
Transportation
Research Board
Hamilton/
Ontario
200 m 200 m grids
1996–
1997/
2001–
2002
Statistic Canada BR
Conditional Logit
Model
Inter-metropolitan firm location studies
Hansen
(1987)
Journal of Urban
Economics
Sao Paulo State,
Brazil
Metropolitan areas
1977–
1979
350 sample branches and transfer
plants in the study area
Nested Logit
Model
McConnell
and Schwab
(1990)
Land Economics U.S. Counties
1973–
1982
Dun and Brad
Conditional Logit
Model
Head, Ries,
and Swenson
(1995)
Journal of
International
Economics
U.S. States
1979–
1987
Japanese Economic Institute
Conditional Logit
Model
Guimaraes et
al. (2000)
Journal of Urban
Economics
Portugal "Concelhos"
1982–
1992
Department of Statistics
Conditional Logit
Model
Holl (2004)
Journal of Urban
Economics
Spain Municipalities
1980–
1984
Register of Industrial
Establishments by Spanish Ministry
of Industry
Poisson Model
Gabe and
Bell (2004)
Journal of Regional
Science
Maine, U.S. Municipalities
1993–
1995
Covered Employment and Wage
(ES-202) data
Poisson/Negative
Binomial Model
61
Devereux,
Griffith, and
Simpson
(2007)
Journal of Public
Economics
Great Britain Counties
1986–
1992
Office for National Statistics Annual
Respondents Database for Great
Britain
Conditional Logit
Model
Melo,
Graham, and
Noland
(2010)
Journal of the
Transportation
Research Board
Portugal Municipalities
1995–
2003
"Quadros de Pessoal", from
Portuguese Ministry of Labor and
Social Solidarity
Poisson Model

62
Table 2-a2 Summary of empirical studies on firm location choices
Author
Economic
sector
Proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market
access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Intra-metropolitan firm location studies
Erickson
and
Wasylenko
(1980)
Construction
A.
population
density;
B. distance
to the
Milwaukee
City CBD
Share of
employees in
each particular
sector in each
municipality to
all non-central
city
Cumulative
number of worker
residents
employed in each
particular sector
within medium
commuting
distance of each
suburban
municipality
B. (+) (+)

(+)
Manufactur
-ing
B. NA (+) (+)
Transportat
-ion
B. NA (+) (+)
Wholesale
trade
B. (+) (+) (+)
Retail
Trade
A. NA; B. NA (+) (+)
FIRE A. NA; B. NA (+) (+)
Services A. NA; B. NA (+) (+)
Charney
(1983)
All
manufactur
-ing (all
size type/
small firms/
medium
firms;
large firms)
A.
employment
density;
B. net
population
density

Gravity-based
accessibility
measure of low-
income
households
(prediction of
negative
neighborhoods
characteristics)
A. (-) for all size
type, but not for
medium firms;
B. (+) for all size
types firms, but
not for small and
medium firms

(+)
Durable
goods
A. (+);
B. (-)
(+)
Non-
durable
goods
A. (+);
B. NA
(+)

63
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
Proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market
access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Intra-metropolitan firm location studies
Ihlanfeldt
and Raper
(1990)
Independent
office
A. distance to
CBD;
B. gravity-
based
accessibility
measure of
total
employment

Gravity-based
accessibility
measure of
employment in
A. financial,
legal, business
Services;
B. eating and
drinking
1. Gravity-based
accessibility
measure of
workers
employed in
A. managerial
and professional
occupations;
B. administrative
support;
2. Number of
poor households
A. NA;
B. (-)
A. (+);
B. (+)
1A. (+);
1B. (-);
2. (+)
Branch
office
A. (+);
B. (-)
A. NA;
B. (+)
1A. (+);
1B. (-);
2. (+)
Shukla and
Waddell
(1991)
Manufactur-
ing
A. distance to
Dallas CBD;
B. squares of
A;
C. distance to
Forth-worth
CBD;
D. squares of
C
Gravity-based
accessibility
measure of
A.
construction,
manufacturing
and wholesale
trade;
B. mining,
transport and
FIRE;
C. retail
trade/services
Gravity-based
accessibility
measures of
employment in
other two
groups of
sectors
Gravity-based
accessibility
measure of
A. total
population
B. median
household income
C. percent of
black population
A. NA; B. (-);
C. (+); D. (+)
(+) B. (-); C. (-)
A. (+);
B. (-);
C. NA
Construction
A. (+); B. (-);
C. (+); D. (+)
(+) B. (-); C. (-)
A. (+);
B. NA;
C. NA
Wholesale
trade
A. NA; B. (-);
C. (+); D. (+)
(+) B. (-); C. (-)
A. (+);
B. (-); C. NA
Retail Trade
A. (+); B. (-);
C. (+); D. (+)
(+) A.(+); B. (-)
A. (+);
B. (-); C. NA
FIRE
A. (+); B. (-);
C. (+); D. (+)
(-) C. (-); B. (+)
A. (+);
B. (+); C. (-)
Services (+) (+) A. NA; B.(-)
A. (+);
B. (+);
C. (-)

64
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
Proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market
access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Intra-metropolitan firm location studies
Waddell and
Shukla
(1993)
3-digit
manufactur-
ing industry
A. total
employment in
center;
B. net
employment
density in
center
Percentage of
center
employment in
manufacturing
Percentage of
center
employment in
A. wholesale
trade;
B. FIRE
Gravity-type
accessibility
measure of
population,
weighted by
distance
A.(-);
B.(-)
(+)
A.(+);
B. (-)
(+)
Coffey,
Drolet, and
Polèse
(1996)
All high-
order Service

Percentage of
revenues from
specific
geographic
areas (
A. CBD,
B. suburb,
C. elsewhere)

A. (+);
B. (+);
C. (+)

producer
Service
A.(+);
C. (+)
FIRE
A. (+);
B. (+)
Arauzo-
Carod and
Viladecans-
Marsal
(2009)
High-tech
manufactur-
ing sectors;
A. population
density;
B. distance to
the Central
City
Previous entries
in the same
sector

A: (+);
B.(+)
(+)

Intermediate-
tech
manufactur-
ing sectors
B.(+) (+)
low-tech
manufactur-
ing sectors
A.(+);
B.NA
(+)

65
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market
access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Intra-metropolitan firm location studies
Maoh and
KaNAroglou
(2009)
Manufactur
-ing
A. distance to
CBD;
B. population
Density
Cumulative
number of
establishments
for each 1-digit
sector within
500 m from the
center of each
grid

A. density of
households;
B. density of
average
household income
A. NA;
B. (-)
(+)

Construct-
ion
A. NA (+)

A. (+)
Transportat
-ion
A. NA (+)

Wholesale
Trade
A. NA (+)

Retail
Trade
(+)

B. (+)
Service A. NA (+) B. (+)
De Bok and
Sander
(2005)
Industrial
land
market:
1a. Space
extensive
industry;
1b.
Agriculture
and
Mineral
extracting

Cumulative
number of jobs
in the three
broad industry
types within 800
meters
Gravity-based
accessibility
measures to
suppliers and
customers,
weighted by
distance
Gravity-based
accessibility
measures to
inhabitants,
weighted by
distance

1a. NA;
1b. NA;

66
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
Proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market
access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Intra-metropolitan firm location studies
De Bok and
Sander
(2005)
Office
market:
2a. Offices
trade and
industry;
2b. Offices
with high
car-
dependency;
2c. Offices
business
Services;
2d. Offices
government;
2e. Social
Services

Cumulative
number of jobs
in the three
broad industry
types within 800
meters
Gravity-based
accessibility
measures to
suppliers and
customers,
weighted by
distance
Gravity-based
accessibility
measures to
inhabitants,
weighted by
distance

2a. NA;
2b. (+) for all
except
firms>20
employment;
2c. (+);
2d. (+);
2e. (+) for all
firms
2c. (+);

Retail land
market:
3a. Public
Services;
3b. Medical
and sports
facilities
3a. NA
3b. NA
4a. NA

67
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
Proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Inter-metropolitan firm location studies
Hansen
(1987)
Manufactur
ing
A. road travel
time to the city
of Sao Paulo;
B. number of
manufacturing
employee
Local
employment in
manufacturing
sector

A. (+);
B. (+)
(+)

McConne
ll and
Schwab
(1990)
Motor
vehicle
industry
(SIC 371)
County
production
worker hours in
manufacturing
industries
State production
worker hours in
SIC 37

Gravity-based
accessibility measure
of personal income in
other states, weighted
by distance
(+) (+)

NA
Head,
Ries, and
Swenson
(1995)
Manufactur
-ing
A. number of
U.S.
establishments
in the same 4-
digit industry
within a state;
B. number of
Japanese-based
activity in the
same 4-digit
industry within a
state;
C. sum of A
and B
A. number of
establishments
in the same
vertical linked
group affiliation
within a state;
B. number of
establishments
in the same
group in the
bordering states
within a state

A. (+);
B. (+);
C. (+)
A. (+) ;
B. (+)

68
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Inter-metropolitan firm location studies
Guimarae
s,
Figueired
o, and
Woodwar
d (2000)
Manufactur
-ing
A. (log of)
population
density;
B. (log of)
travel time to
Porto and
Lisbon(capital
cities)
A. share of
manufacturing
employment in
the same 3-digit
sector;
B. share of
employment in
foreign plants
Share of
employment
in business
and financial
services
sector

A. (+);
B. (+)
A. (+).
B. NA
(+)

Devereux,
Griffith,
and
Simpson
(2007)
Manufactur
-ing
Total working
age population
A. number of
plants in each 4-
digit industry in
each county-
year;
B. number of
foreign-owned
plants in each
industry in each
county-year

(+) (+)

Melo,
Graham,
and
Noland
(2010)
Primary
industries
Population
density of each
municipality
Location
quotient of each
municipality in
each industry

Gravity-based measure
of populations of
Lisbon, Porto and the
capital of the Distrito to
which the municipality
belongs, weighted by
distance
(+) NA

(+)
Manufacturi
ng
(+) (+) (+)
Electricity/
gas/ water
(+) NA

NA
Construction (+) NA (+)
Wholesale
and Retail
Trade
(+) (+)

(+)
69
Table 2-a2 Summary of agglomeration effects on firm location choices (continue)
Author
Economic
sector
proximity-related agglomeration variable Significance of agglomeration variable
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor) market access
Urbanization
effects
Localization
effects
Inter-industry
linkage
(labor)
market
access
Inter-metropolitan firm location studies
Gabe and
Bell
(2004)
All sectors
Municipality
population size
Location
quotient of 1-
digit SIC
category

(+) (+)

Holl
(2004)
Manufactur
ing:
a. pooled
sample;
b. primary
metal;
c. minerals;
d. chemical
products;
e. metal
products;
f. transport
equipment;
g. food and
beverage;
h. textile
and
clothing;
i. paper and
printing;
j. wood and
furniture;
k. plastics
and others
population
size
A. location
quotient for each
sector;
B. share in
total national
industry
employment
A.
intermediate
demand:
Cumulative
number of
road flows of
merchandise
trade from
other
municipalities
that are
within half an
hour travel
time;
B. supplier
access:
cumulative
number of
value added
in each sector
in other
municipalities
that are
within half an
hour
A. Inter-regional:
Cumulative number of
population in other
municipalities that are
within half an hour
travel time
B. Intra-regional:
shortest travel time from
each municipality to its
provincial capital city
(+) for
pooled
sample and
all sectors
except d, f
sectors
A.(+) for
b,d,i,j sectors,
(-) for e, g, h
sectors;
B. (-) for
pooled
samples and
c,d,e,g,h,j
sectors
A.(-) for all
samples and
for c, e, g, h,
i, j sectors;
B. (+) for
c,e,f,i,j,k
A. (-) for all
samples and
c,e,g,h,i,j
sectors;
B. (-) for c, e,
g, i, j sectors

70
Table 2-a3 Summary of empirical studies on the negative impacts of congestion on economic efficiency
Methods to
examine
congestion
costs
Study Journal Unit of analysis Study period Measures of congestion
Measure of economic
efficiency
Instruments
used for
congestion
measures?
congestion
reduces
transport
infrastructure
efficiency
Boarnet
(1997)
National Tax
Journal
Counties of
California
1977–1988
(Inverse of) the
index
County output
(productivity) per worker
Yes
Fernald
(1999)
American
Economic
Review
National (U.S.) 1953–1989
Total miles driven by
trucks and auto
Growth rate of
productivity for each
industrial sector
Yes
Hymel
(2009)
Journal of
Urban
Economics
U.S. metropolitan
areas
1982–2003
TTI's index of travel
delay per capita
Growth rate of
employment
Yes
Montolio
and Solé-
Ollé (2009)
Papers in
Regional
Science
Spanish Provinces
(NUTS 3)
1984–1994
Total kilometers driven
by trucks and auto
Growth rate of total factor
productivity
7

Yes
Broersma
and Dijk
(2008)
Journal of
Economic
Geography
Netherland
provinces (NUTS
2)
1995–2002 Cars per kilometer
Growth rate of total factor
productivity
No
Sweet
(2013)
Urban studies
U.S. metropolitan
areas
1993–2008 (for
employment
growth);
2001–2007 (for
productivity
growth per
worker)
1) TTI's index of travel
delay per capita
2) Average daily traffic
per lane (ADT)
1) Growth rate of
employment
2) Growth rate of
productivity per worker
Yes
congestion
alters
accessibility
Graham
(2007)
Journal of
Urban
Economics
UK Wards 1995-2002
"Effective density":
straight-line distance
based measure vs. travel
time based measure
Firm-level outputs No

7
Total factor productivity growth is the "output growth not produced by private inputs of production" (Montolio and Solé-Ollé 2009, 104).
71
Hartgen,
Fields, and
Moore
(2009)
Report by the
Reason
Foundation
U.S. metropolitan
areas
1999, 2000, 2001,
and 2005
Absolute / relative
accessibility to key
destinations within each
region: free-flow travel
time based measure vs.
congested travel time
based measure
Productivity per worker No

72
CHAPTER 3 TRAFFIC CONGESTION, POLYCENTRICITY, AND INTRA-URBAN
FIRM LOCATION CHOICES

A study of the Los Angeles metropolitan area

1 Introduction
Traffic congestion has been intensively studied in transportation research. Many empirical studies
focus on the causes, social consequences, and policy solutions of traffic congestion, while the economic
costs of traffic congestion are evaluated in the most direct way. For example, the estimation of the
economic value of time is the key issue in measuring the extra costs for transport system users caused by
congestion delays (e.g. Goodwin 2004; Lomax, Schrank, and Eisele 2012). Except for the immediate
costs for transport users, Sweet (2011) suggests that traffic congestion also causes second-order impacts
to the economy that extend beyond the transportation system. However, not many empirical studies
examine the wider impacts of congestion because it is difficult to identify how businesses, individuals,
and land-use patterns respond or adapt to traffic congestion (Sweet 2011).
Traffic congestion, like land rents and pollution, is an important source of diseconomy in
agglomeration research. The view that the tradeoff between the costs of congestion and agglomeration
benefits explains the structure and growth of cities and regions has been repeated in many agglomeration
theories (e.g. Fujita and Mori 2005; Richardson 1995). While the positive benefits of agglomeration
economies have been intensively examined in empirical studies, the dampening effect of traffic
congestion has only been examined in a few studies (e.g. Boarnet 1997; Graham 2007; Hymel 2009). One
major reason is the endogeneity of traffic congestion to the economy, in that congested locations may also
be economically robust places with high growth potentials (Hymel 2009). How to unravel the costs of
congestion from the potential benefits of agglomeration remains an important issue in empirical studies.
73
This study aims to estimate the economic impact of traffic congestion on agglomeration through
the lens of firms' location decisions. Specifically, I focus on the location choice of new establishments
among employment centers within the Los Angeles metropolitan area and test whether congestion costs
are incorporated in firm location choice. Drawing on the theories of industrial location and agglomeration,
the central hypothesis is that traffic congestion reduces the accessibility advantages and agglomeration
benefits of employment centers within a metropolitan area, and reduces the probability of these centers
being chosen by firms, all else being equal. Moreover, firms from different industrial sectors are expected
to vary in their spatial responses to traffic congestion, and those firms placing a higher value on the
economies of proximity are more likely to endure congestion costs. By focusing on the location choices
of new businesses, all locational attributes can be considered as exogenous, so that the costs of congestion
and the benefits of agglomeration can be examined and evaluated separately.
The structure of the chapter is as follows. Section 2 reviews the theoretical basis of the possible
impact of traffic congestion on firm location, spatial location patterns, and the general literature on
determinants of firm location choice. Section 3 describes the conceptual model and the estimation
approach. Section 4 describes the various data sources used, while Section 5 presents descriptive results
and estimation results of the discrete choice model. The chapter concludes with a summary of the findings
and discusses the remaining questions for future work.
2 Literature review: Transport, traffic congestion, and firm location
In this section, I review relevant theories of industrial location, agglomeration economies, and
urban structure that focus on individual firms' location choice. The purpose is to identify the general role
that transport has, and the possible impact that traffic congestion has on firm location, especially within
an urban context. Related empirical studies in transportation analysis are also discussed to derive the
implications of congestion effects on individual firms' location behavior.

74
2.1 Theoretical basis
Traditionally, transport costs play a key role in industrial location theory. In Weber's (1929)
classic triangle problem, the optimal location of an individual firm would be the one that minimizes the
total transportation costs to input suppliers and to customers. When input factors are allowed to substitute
each other, the optimal location becomes the profit-maximization one and this would not overlap with the
transport-minimization location (Alonso 1967; Moses 1958). Further theoretical developments of firm
location choices in a transportation-network space suggest that when transport rates decrease with
distance, the optimal location would be market sites, nodes on the network with a high degree of
centrality, or junction points for mode transfers, depending on their relative locations on the network, and
on the transport cost functions (Louveaux, Thisse, and Beguin 1982).
Another tradition of location theory, namely the "Von Thunen" tradition, focuses on the
competition of firms for locations with market-access advantages. Bid-rent curves are used to describe
firms' willingness to pay for locations to achieve maximum profit at a given level of output (Alonso 1967).
Following this tradition, the standard monocentric urban model—and the later development of the
polycentric model—solves the bid-rent function of firms as a result of the tradeoff between the proximity
to markets and the proximity to an unevenly distributed labor force (Solow 1972; White 1976). In terms
of individual firms' location, the pull of land rents is determined exogenously by the equilibrium rent
structure—the outer envelope of bid-rents—and can be treated similarly to the pulls of other factor inputs
(Alonso 1967).
Tracing back to Marshall's (1890) economic analysis, theories of agglomeration economies
suggest that individual firms benefit from proximity to other firms because of intermediate input sharing,
labor-market pooling, and knowledge spillover effects. It can be directly derived from agglomeration
theories that transportation can enhance the agglomeration effects by substituting for spatial proximity
and increasing the connectivity between firms and workers in the cluster (Graham and Melo 2011).
75
The recent development of the new economic geography (NEG) theory takes into account the
interaction between transportation costs and agglomeration economies and implies the condition under
which individual firms would choose the locations where other firms agglomerate (Krugman 1991). In a
review of previous NEG modeling, Fujita and Mori (2005) summarize how firms will tend to agglomerate
only when the transport costs of the final differentiated products are moderate: when transport costs are
high, firms will tend to disperse over space to serve dispersed local demands from the workers in the
traditional sector. When transport costs are low, high land rents in the cluster and the resultant high labor
costs will be the dominant driving forces for firms to disperse (Fujita and Mori 2005). Their market
potential function approach suggests that the propensity of a single firm to locate in an agglomeration (or
a city) is determined by: 1) the labor cost advantage of dispersed locations, 2) the market size of the
cluster, 3) accessibility to the market in the cluster, and 4) the intensity of competition for differentiated
products in the city (Fujita and Mori 2005). Thus, firms will tend to locate away from the agglomeration
due to the lower competition intensity and labor costs in lower density locations.
Transport costs and rates are assumed to be exogenous in industrial location theories and NEG
modeling. However, in reality, transport costs vary over space and time because of traffic congestion,
which occurs when the travel demands exceed the transportation network capacity. Since the demand for
travel is derived from the need for interactions between economic agents, the aggregate results of firms'
and residents' location decisions will also affect transport cost outcomes (e.g. Anas and Kim 1996; Mun
and Yoshikawa 1993; Tauchen and Witte 1983; Wheaton 2004). In other words, the transport cost may be
endogenous to firm location decisions because of the limited transportation supply. Moreover, firms not
only benefit from spatial interactions with each other through exchanges of information and inputs, but
also suffer from the costs of traffic congestion associated with these interactions (e.g. Mun and
Yoshikawa 1993; Tauchen and Witte 1983).
The endogeneity of transport costs and traffic congestion has been introduced in some theoretical
models of intra-urban firm locations (e.g. Anas and Buyukeren 2013; Anas and Kim 1996; Anas and Xu
76
1999; Mun and Yoshikawa 1993; Tauchen and Witte 1983; Wheaton 2004). These models assume that
the location choices of firms/households and land rents are simultaneously determined with the travel-
time matrix and congestion distributions on the road network (e.g. Anas and Kim 1996; Anas and Xu
1999; Mun and Yoshikawa 1993; Tauchen and Witte 1983; Wheaton 2004). Resorting to the numeric
solution, some models suggest that compared with the scenario without endogenous congestion, the
accessibility advantage is diminished, and firms are more likely to decentralize and disperse over space
(e.g. Anas and Kim 1996; Mun and Yoshikawa 1993; Tauchen and Witte 1983; Wheaton 2004). Some
models also suggest that the higher initial density of a location, the greater the density loss in the presence
of endogenous traffic congestion, such that the natural accessibility advantage of a central location will be
diminished the most, and may even be cancelled out (e.g. Mun and Yoshikawa 1993; Tauchen and Witte
1983).
In sum, some general implications on the location behavior of a single firm may be derived from
these models.
1) The location choice of a firm results from the tradeoff among proximity to markets, labor force,
other intermediate inputs, and firms of the same kind. When fixed transport costs dominate, individual
firms tend to choose "corner locations," input/output place, or the nodes and junctions on the transport
network.
2) Changes in the transport rate will change the relative attractiveness of different input factors
and final demands as well as the balance between agglomeration and dispersion forces.
3) Traffic congestion increases an individual firm's travel costs and shifts its bid-rent curve,
making those locations connected by congested routes less profitable for the firm.

77
2.2 Possible impacts of traffic congestion on firms
Possible impacts of traffic congestion on firms' operation and production have been examined in
some transportation analyses and can be summarized in terms of two aspects (Sweet 2011; Weisbrod et al.
2001; Weisbrod et al. 2003).
The first aspect concerns the direct impact of travel costs. Analogous to transportation
improvements, traffic congestion directly increases costs for existing travel by causing travel-time delays
and unreliability (Sweet 2011). According to the report by the Economic Development Research Group
(2005), traffic congestion not only increases workers' commuting costs, but also increases firms' time
costs for business-related trips that are subject to schedule requirements. On the one hand, firms in those
production-related (or trade-related) industries such as manufacturing, transportation, and the wholesale
and retail trade may have to 1) adjust their schedules to avoid the delays caused by peak afternoon
congestion for inbound shipments, 2) increase their inventories in terms of both volumes and variety to
reduce the uncertainties and delays in scheduled deliveries, and 3) deploy extra vehicles and crews due to
longer travel times (Economic Development Research Group 2005; Weisbrod and Fitzroy 2008). On the
other hand, firms engaged in office-related activities or local services industries may also be adversely
affected by congestion because of possible delays in business meetings and conferences as well as
increased access costs for their inbound or outbound services (Economic Development Research Group
2005).
There is also an indirect impact. By reducing the efficiency in the flow of goods, people, and
information between locations, congestion effects may extend beyond the existing travel within the
transportation system, and further reduce the agglomeration and accessibility advantage of a location. For
example, firms' relative ease of access to specialized workers and suppliers within a given travel time may
be decreased due to traffic congestion, and firms may have to substitute those inputs with others that do
not match their production needs (Weisbrod et al. 2001; Weisbrod et al. 2003). The size of the output
delivery market served by firms would also shrink (Weisbrod et al. 2001; Weisbrod et al. 2003).
78
In sum, by directly increasing firms' operation costs and indirectly reducing their access to
markets, traffic congestion adversely impacts business' productivity and locations' competitiveness for
retaining and attracting economic activities. Thus, firms may avoid choosing congested areas or those
areas connected by congested routes when opening new businesses.
3 Research approach
Based on the literature review, two basic hypotheses are developed here:
H1: Firms are less likely to be attracted to traffic-congested locations, all else being equal.
H2: Firms in industrial sectors that benefit more from agglomeration economies will be less responsive to
congestion costs.
To test these basic hypotheses, this study develops and estimates discrete choice models of the
establishment of location choice within the Los Angeles metropolitan area. Following the pioneering
work of McFadden (1974), the location behavior of an individual new business is modeled as the
probability of choosing from among a set of alternative locations based on the attributes of each
alternative and the firm's own attributes. This approach is consistent with the "Weber" tradition of
industrial location theory that assumes individual firms rationally choose the optimal location to
maximize their expected profits (Arauzo-Carod, Liviano-Solis, and Manjón-Antolín 2010; Guimaraes,
Figueiredo, and Woodward 2004). Moreover, the location decisions of individual firms are assumed to be
separated from other considerations (such as the decision to run a business or the choice of business size)
and each firm is considered to be a price taker of all markets and its location decision making is motivated
to maximize its potential profit (utility). By focusing on new businesses, the inertial behavior of firm
migration due to high moving costs is avoided, so that different locational characteristics may be
separated and compared. This is especially important for evaluating the effects of congestion costs, which
could be simultaneously determined with existing land-use and employment patterns, but can be
considered as exogenous in the eyes of new businesses. As I will discuss in a later section, the location
79
pattern of new firms and existing firms do not seem to differ significantly from each other, so that my
conclusions regarding the congestion effects on new firm locations might also be extended to firm
location decisions in general.
3.1 The basic model
Following the profit-maximization assumption, each new business i attaches an expected profit to
each alternative location j consisting of an observed component and a random unobserved component:

, 1, …;   1, … (3-1)
where X
is the vector of the observed and exogenous location-specific variables and ε

is an
independently and identically distributed (i.i.d.) error term capturing other unobserved factors. Each firm
chooses the alternative location that yields the highest expected profit. As suggested by McFadden (1973),
the probability of firm i choosing location j takes the following form if the random component follows the
i.i.d. distribution:


exp
/ ∑ exp
(3-2)
The coefficients of (3-2), β, would be estimated with the maximum likelihood method.
3.2 Definition of choice sets
Previous studies on intra-metropolitan firm location usually use constant-boundary administrative
units such as zip code areas as available locations (e.g. Charney 1983; Shukla and Waddell 1991). This
implies a large number of spatial alternatives and possible computational difficulty. As suggested by
McFadden (1978), this problem can be addressed by replacing the full choice set with the chosen
alternative and a randomly drawn sample of other alternatives. Another potential problem with the use of
these narrowly defined spatial units is that the unobserved portion of utility could be spatially correlated
across neighboring locations that would violate the independence of irrelevant alternatives (IIA)
assumption in the discrete choice model. Thus, some studies introduced accessibility variables to control
80
for spatial association effects (e.g. Shukla and Waddell 1991; De Bok, Michiel, and Sanders. 2005).
Waddle and Shukla (1993), instead, use employment centers as the basic spatial unit to explore the role of
intra-urban agglomerations and the polycentric nature of metropolitan areas.
This study follows Waddell and Shukla’s (1993) method and uses employment centers as the
alternative sets for the following reasons. First, as explained by NEG theory, the existence of employment
centers is a result of the tradeoff between agglomeration economies and diseconomies within
metropolitan areas (Agarwal, Giuliano, and Redfearn 2012; Giuliano et al. 2011). Empirical studies also
show that employment centers are the "right unit" to analyze agglomeration effects because they act
persistently as magnets of intra-urban growth (Giuliano and Redfearn 2009; Redfearn 2009, Redfearn
2014). Secondly, based on the definition of employment centers, the component spatial units within each
center are relatively homogeneous in terms of, for example, minimum density, or employment/population
ratios. Previous empirical studies also show that the employment centers differ from each other in terms
of size, industrial composition, and other spatial characteristics (Giuliano and Redfearn 2009; Giuliano
and Small 1991). This provides sufficient heterogeneity among spatial alternatives as required by discrete
choice analysis. Moreover, employment centers are predicted in NEG theory to be spatially distant from
each other to avoid being in the market shadow of each other (Fujita and Mori 2005), which might reduce
the spatial correlation problem at the micro-geographic level. Thus, using employment centers as a
potential set of alternatives allows us to examine the nature of intra-metropolitan agglomeration effects
and polycentricity from the perspective of firm location behavior.
3.3 Definition of explanatory variables
The observable component in equation (3-1) can be specified as a function of locational
characteristics in a linear additive form, including the non-purchased locational characteristics and
spatially variable input factor price. Within an urban context, the most important input price for firms
would be industrial and commercial land rents (Shukla and Waddell 1991). However, these price factors
81
are not generally available. Since land rents are also functions of locational characteristics, I follow earlier
studies and summarize the influence of land costs by other non-purchased locational attributes.
The set of locational attributes in the expected profit function consists of the following: 1) traffic
congestion measures, which is the most important location factor for the purpose of this study; 2)
agglomeration measures, which arise from the proximity to labor forces, the proximity to other related
businesses, and the local concentration of aggregate employment and employment in related industrial
sectors; 3) other accessibility-related attributes, including distance to freeways and arterials, and distance
to the airports that proxies for inter-regional connection; 4) land use characteristics, which are represented
by the fraction of local land in related use types; and 5) local labor quality measures. The construction of
the explanatory variables is discussed below.
3.3.1 Traffic congestion
The quantification of traffic congestion has been widely discussed in transportation studies and
various measures have been constructed. As suggested by Mondschein, Taylor, and Brumbaugh (2011),
the concept and measures of traffic congestion differ from those of accessibility measures in two ways: 1)
congestion measures focus on the transport network as the unit of analysis, while accessibility measures
usually focus on individuals and places; and 2) congestion measures focus on the performance of the
transport system as the ends instead of the means of economic interactions. Among the various congestion
measures, however, travel-time measures reflect the effectiveness of the transportation system as well as
business' and residents' concerns about time costs (Lomax et al. 1997).
This study applies travel time-related measures because the impacts of traffic congestion on firm
behavior are mainly through travel-time delays, which directly increase transport costs, and indirectly
reduce locations' accessibility advantages. Moreover, because employment centers are defined as the
spatial choice set, two types of congestion are defined here: one associated with travel-time delays in
accessing the labor force and other businesses within the metropolitan area, and the other associated with
82
delays in accessing other businesses within each center. In other words, the first describes to what extent
congestion affects agglomeration economies accruing to firms at the regional scale, while the second
describes how firms' benefits of locating within an employment center are diminished by local congestion.
—Regional congestion
The regional congestion measure derives from the "delay rate" measure of congestion that was
originally used to measure the difference in the travel rate between congested and uncongested conditions
for specific road segments or trips (Lomax et al. 1997). To convert the trip-based measure into a location-
based measure, I use the average delay rate for all possible trips from each location to all other locations
that are within a certain distance boundary as a measure of the average congestion level for a location i:


∑

  /
,

(3-3)
where n is the total number of tracts that are accessible to location i within distance boundaries, and

under different scenarios is the network travel time between focal location i and any other location j
within the travel-distance threshold. The morning peak is used as the congested condition because it is
"nearly uniform, consistent and predictable" (Lomax et al. 1997), while the night period is used as the
free-flow or benchmark condition because it is usually easily identifiable by travelers as a desirable or
acceptable condition. For each pair of locations (i and j),
for AM peak/Night period is the shortest-
path travel time on the network for each period; within each census tract i,
is estimated as the average
travel time for all road segments (excluding limited-access highways) within each census tract for each
period. For each employment center, the congestion level of its peak-density tract will be used to
represent the regional congestion level experienced by the employment center.
The choice of the distance boundary depends on what type of trip purpose associated with firms is
studied. As discussed, various types of trips, such as commuting, business-related trips, and deliveries are
83
related to firms' operation and transport costs directly or indirectly. This study mainly uses the average
commuting trip length as the distance boundary, which is approximately 12 miles, according to the 2001
National Household Travel Survey (NHTS). This distance threshold is chosen because journey-to-work
trips take up a much larger proportion of the total number of personal trips than work-related trips do.
8

Thus, the effects of metro-wide congestion costs imposed on firms are more likely to be through workers'
commuting time delays that yield higher wages and labor costs for firms. Limiting the distance threshold
to a smaller range close to the local labor-market area also allows the spatial variation in congestion levels
to be better captured.
—Within-center congestion
Similar to the regional congestion measure, within-center congestion is also based on the "delay
rate" measure that captures to what extent within-center trips are delayed:
_

1

, , / , ,
, ,
(3-4)
where a and b represent component tracts of employment center i, , is the network travel time between
a and b for different periods, and m is the total number of routes within centers.
9
This measure proxies for
the efficiency of within-center trips most likely to be associated with face-to-face contacts between firms,
such as short-distance trips for business meetings, conferences, or consulting services. Analogous to the
general belief that larger cities are more congested, I expect that the within-center congestion level will
increase with the size of centers, because a larger concentration of firms generates greater travel demands,
and more intensive use of the roads and highways within each center.

8
The NHTS defines work-related trips as those trips for "attending business meetings" and "other work-related
trips." According to the 2001 NHTS' "Day Trip File for Public Use," the total number of journey-to-work trips
(Code = 11) is 47,475, while the total number of trips for "attending Business meetings" (Code = 13) and "other
work-related trips" (Code = 14) is 1,316 and 11,692, respectively.
9
Assuming employment center i has k census tracts, the total number of trips is calculated as
,
which is the total number of pairs of tracts within i plus the total number of component tracts of the center.
84
The two measures of congestion may not be highly correlated with each other. For example, some
centers experiencing high levels of regional congestion may still have efficient travel at the local scale
because of the benefits associated with proximate locations within a center, which also facilitates
interactions between firms.
3.3.2 Accessibility
Accessibility to the labor force, customers, and suppliers is an important source of metro-wide
agglomeration economies. Accessibility variables are also important background variables for firm
location choice. The most commonly used form of accessibility measure is the gravity-based measured,
which is a weighted sum of opportunities at the destination discounted by distance to the focal location.
However, since distance-decay effects in this study are captured by congestion measures, I use the
cumulative opportunity form to measure access to potential workers and other related businesses:
∑ ,

(3-5)
, ∑
, ,

(3-6)
where the labor force ( ) is proxied by the whole working-age (18–65) population and the total
employment in the same industry m ( , ) is used as objective opportunities for business access. This
form of accessibility measure considers all opportunities available within certain thresholds as equally
important (Handy and Niemeier 1997). Consistent with congestion measures, the 12-mile distance
threshold is used for both of the two measures. Similarly, the accessibility levels of the peak-density tract
for each employment center are used to proxy for the center's accessibility.
One possible problem with the cumulative measure is that the two accessibility variables are
likely to be highly correlated with each other. Moreover, this form of measure does not consider the
competition for opportunities between firms. Following Giuliano et al. (2011), I construct another
measure of "relative labor-force accessibility" ( ) that takes the following form:
85

∑

∑
, ,
,
} (3-7)
This accessibility measure assumes the probability of the labor force in location j choosing the
job position in center i as center i's employment share of total employment that is accessible within the
commuting distance boundary (12 miles) of location j. Thus, the total number of individuals in the labor
force in location j available for center i would be the product of the total labor force in j and the
probability of those individuals in the labor force choosing jobs in center i. This relative accessibility
measure not only accounts for the competition for labor between centers, but also helps to reduce the
interaction between the accessibility variables.
3.3.3 Urbanization/localization economies
Besides the accessibility advantages, agglomeration benefits within centers are measured by the
local concentration of employment activities. I construct two measures for urbanization effects: the total
employment size ( ) and the total employment density of each center ( ). The
employment density variable would also proxy for per unit land costs and land competition between firms
(Giuliano et al. 2011). Thus, the expected sign for this variable could be either positive or negative.
On the other hand, the localization effects are measured as the percentage of each center's
employment in the same industrial sector ( ). Employment centers concentrated with economic
activities in a specific sector are expected to attract new firms in the same sector. Thus, this variable is
expected to exert a positive influence on firm location.
3.3.4 Other control variables
Other control variables include land-use characteristics, proximity to transport facilities, and local
labor quality. The percentage of a center's land that has been developed into related land-use types is used
to represent the supply side of land use (_ ). This variable could either imply the amount of
available rental space for particular use or a constraint on future development for the same use type
86
(Erickson and Wasylenko 1980). Thus, the coefficients for land-use variables are expected to have either
positive or negative signs.
The variables for transport-access measures include a straight line to the nearest highway ramp
( ) and the straight-line distance to the nearest major airports ( ).
10
As suggested by Holl
(2004), proximity to the transport infrastructure should be considered as a separate locational factor from
market access because road corridors may act as extended markets. However, as implied by other
empirical studies, highway access may play a minor role in firm location choices at the intra-metropolitan
scale because this access is almost ubiquitous across different locations within metropolitan areas
(Giuliano et al. 2011; Giuliano and Small 1999). Distance to the airport is also controlled to represent the
proximity to inter-regional transport facilities, which have been proven to play significant roles in
previous empirical studies (Button et al. 1995; Cie ślik 2005a, 2005b; Egeln, Gottschalk, and Rammer
2004).
Finally, the quality of the local labor force is represented by the percentage of the population over
25 with at least a bachelor's degree for each center (_ ). This variable reflects the availability of
local human capital and wage levels. As summarized by Arauzo-Carod, Liviano-Solis, and Manjón-
Antolín (2010), firms might tend to choose those locations with workers having higher education levels
and with potentially higher human capital, or they may avoid these locations because of higher wage
levels. Moreover, this variable may confound the influence of residential land prices that implies general
land costs and may reduce the attractiveness of locations to new firms. Thus, a positive sign is expected
when firms value the local quality of labor skills and potential capital, while a negative sign is expected
when firms seek locations with lower labor costs or land costs.

10
The nine commercial airports in the LA region include Bob Hope Airport, John Wayne - OR County Airport,
Long Beach Municipal Airport, Los Angeles International Airport, Palm Springs Regional Airport, Ontario
International Airport, Palmdale Air Terminal, Southern California Logistics, and March Global Port.
87
4 Data
This study uses the Los Angeles region as the study area, which has been considered as the most
congested metropolitan area in the U.S.A. for the past 30 years. The definition of the Combined Statistical
Area (CSA) of the LA region is used, which consists of the five counties of Los Angeles, Orange,
Riverside, San Bernardino, and Ventura.
11
The sources and descriptions of the data on employment, the
population, the transportation network, and land use for the region are discussed below.
4.1 Employment data
The National Establishment Time Series (NETS) is used to measure firms' location choices
within the LA region. The NETS database provides establishment-level information such as address,
geographic coordinates, employment, sales, and establishment categories, and follows the lifecycle of
establishment over time by assigning a unique D-U-N-S number to each establishment,
12
which is not
allowed to change (Walls and Associates 2008). An establishment in the dataset is defined as a standalone
firm, a headquarters, or a branch or division (Walls and Associates 2008). The birth of a new business is
detected through multiple sources, including public records, government registrations, and news and
media reports, and only enters into the database when it is confirmed as having started doing business
(Walls and Associates 2008). As suggested by Kolko and Neumark (2008), a key strength of the NETS
data is that it distinguishes true birth and death from establishment relocations, which fits well with my
study purpose.
The NETS datasets for the Los Angeles region are extracted from the California datasets for the
years 2000 to 2005, with the year 2000 set as the base year for measuring the locational variables.
Focusing on the intra-metropolitan scale, new businesses are defined as all types of newborn

11
I have also experimented with the urbanized area of the Los Angeles region. However, the key statistics, including
the location quotient of center locations, and accessibility and congestion measures do not change that much because
the non-urbanized areas do not accommodate much employment activity. The results of the regression analysis do
not change that much either and are not reported here.
12
The D-U-N-S number means the “Data Universal Numbering System” established by Dun and Bradstreet (D&B)
in 1963. Source: http://www.dnb.com/content/dam/english/dnb-data-insight/duns_number_overview_2011.pdf
88
establishments. All the self-employed standalone firms (employment size = 1), and all the "cottage firms"
with less than three employees (and based on residence) are excluded because they would have little
demand for space, and their locational behavior would be more likely to be footloose. The agriculture
(SIC1 = 1 or NAICS2 = 11) and mining (SIC1 = 2 or NAICS2 = 21) sectors are also excluded in advance
because these firms are more likely to be oriented toward material inputs and may not fit well with my
model of urban firm location. Based on the definition, there were 399,527 new establishments born in the
LA region in the 2001–2005 period, associated with 2,102,718 new jobs.
Employment centers identified
The classic method of center identification by Giuliano and Small (1991) is applied in this study.
The establishment-level NETS data are matched to the 2000 census-tract boundaries, and the 10 jobs per
acre density/10,000 total jobs cutoff (referred to as a "10/10" center) is used to identify the employment
centers in the base year. Using the 2000 employment data of NETS, 47 10/10 centers are identified in the
2001–2005 period, accommodating 132,285 new establishments. The geography of the employment
centers is illustrated in Figure 3-1, with the centers ranked by employment size.
89

Figure 3-1 Spatial distribution of employment centers identified in 2000 (10/10 criterion)

Choice of industrial sectors
In the 2001–2005 study period, the percentage of new establishments that chose to locate in the
employment centers of the region was 33.1 percent, while the share of employment centers in the total
employment of new establishments was slightly higher (about 36.8 percent). Not all firms value the
agglomeration benefits associated with centers and their sensitivities to congestion costs may also differ.
For example, agglomeration theories suggest that those firms valuing the importance of face-to-face
interactions are more likely to choose to locate where other firms concentrate. Since this study aims to
examine the nature and role of employment centers, I choose those sectors that are spatially concentrated
within centers. To examine this, the location quotient (LQ) for each industrial sector in all center locations
is calculated as each sector's share of center locations for new establishment counts divided by the sector's
90
new establishment share of the regional total. If the center location's LQ for a sector is larger than 1.1, the
employment centers are considered to have a higher concentration in the sector than in the whole region.
Using this LQ threshold, I explicitly choose those sectors that already favor center locations, which is
consistent with the conditional choice model focusing on firms' location choices among employment
centers, based on the fact that the center locations have already been chosen.
Table 3-a1 (in the appendix) indicates the number of new establishments, associated employment
counts, and the LQs for all the NAICS3 sectors (or NAICS4 sectors for the professional services sectors)
in center and non-center locations. In the study period, the total number of new establishments born in
those sectors meeting the LQ threshold is 45,563 and the total employment of those establishments is
267,787. To examine if the location pattern of new firms across center and non-center locations is
significantly different from the location pattern of existing firms, the same information for existing firms
in the base year (2000) is indicated in Table 3-a2, and t-tests on the equality of the means for LQs are run
(see Tables 3-a3 and 3-a4 in the appendix for details). The difference in the means of LQs in center
locations between new firms and existing firms is less than 0.03, while the t-statistic for the difference is
only 0.71 (df = 190), which is not significant at any level of significance. Similarly, the difference in the
means of the LQs in non-center locations is −0.01 and is not significant either (t
190
= −0.67). This result
implies that the location decisions of new firms can be representative of the location decisions of firms in
general.
To run the conditional logit model, I choose those sectors that not only show a pattern of
concentration in centers but also have at least 1000 new establishments locating in centers (see the
highlighted rows in Table 3-a1).
13
This threshold for the sample size is chosen because, as stated by
McFadden (1984), "sample sizes which yield less than thirty responses per alternative produce estimators

13
Some sectors also fit these standards but are not chosen because by comparing the NETS dataset with the County
Business Pattern (CBP) dataset, I found that employment shares in these sectors at the NAICS2 level for the entire
Los Angeles region are undercounted (e.g. NAICS2 = 55) or over-counted (e.g. NAICS2 = 71) in the NETS dataset
by more than 20%.
91
which cannot be analyzed reliably by asymptotic methods" (1442). However, this study lessens the
standard slightly so that some finer industrial sectors in some broad economic sectors could be included
(e.g. the printing sector in the manufacturing sectors). Thus, the 1000 sample-size threshold is set up to
ensure that the dependent variable is not zero and that there are at least 20–30 respondents per spatial
alternative, so that parameters for the expected profit function can be reliably estimated.
Figure 3-2 illustrates the LQs of center and non-center locations for new establishments in those
selected sectors. The table and the figure indicate that sectors concentrating in centers include a variety of
economic activities. However, most of the chosen sectors are office-related activities, such as the
information, financial and insurance, and professional service sectors.

Figure 3-2 Spatial concentration pattern of chosen sectors across center and non-center locations
To examine how firms in different sectors value the benefits of locating within centers differently,
and to predict their sensitivities to congestion costs and delays, I categorize the employment centers by 3-
quantiles of the centers' employment densities, and calculate the LQs for each category. As suggested by
Ciccone and Hall (1996), the density of economic activities is a better proxy for agglomeration economies
than the size of economic activities. Thus, those sectors valuing agglomeration benefits of proximity to
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
Center
Non‐Center
92
other firms within centers should be more concentrated in denser centers. Figure 3-3 shows the results
with all sectors ranked by the LQs in centers with the highest employment densities. Moreover, sectors
are also categorized into 3-quantiles based on their LQs in the densest centers. The "high-concentration,"
"median concentration," and "low concentration" groups of industries are defined to match the three
categories (see Figure 3-3).

Figure 3-3 Spatial concentration pattern of chosen sectors across centers of different density
categories

The three high-concentration sectors are all office-related activities, in which the motion picture
and legal services sectors take relatively large shares of the total employment in the region for the base
year, at about 1.55 percent and 1.56 percent, respectively. Previous studies imply that the "production"
process of office activities may rely heavily on frequent contacts with outside customers and suppliers,
which often takes the form of face-to-face meetings (Ihlanfeldt and Raper 1990), or less formal face-to-
face contacts between employees of different firms (Coffey, Drolet, and Polèse 1996). Employment
centers, including the traditional central business district (CBD) and subcenters, may provide inviting
business environments for these new office firms because spatial proximity between firms saves travel
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
High
density
centers
Medium
density
centers
Low density
centers
High
concentration
Medium
concentration
Low
concentration
93
costs for personnel for face-to-face meetings, and there are more chances for learning and information
sharing (Coffey, Drolet, and Polèse 1996). Figure 3-3 shows that those sectors showing a higher level of
concentration in denser centers are usually the so-called front-office activities that have a higher need for
face-to-face interactions (e.g. Coffey and Bailly 1991; Shearmur and Coffey 2002). Thus, these sectors
are also expected to be more likely to endure congestion delays and slower travel speeds as long as
interaction costs within centers are still reduced by spatial proximity between firms (Mills and De Ferranti
1971). On the other hand, congestion delays in terms of access to workers and other related businesses
surrounding employment centers at the regional scale may also negatively affect centers' attractiveness for
these firms.
Some production-related industrial sectors also show patterns of concentration within centers.
Different from those office activities, the agglomeration forces that drive firms in these sectors to
concentrate in centers are more likely to be the savings in transport costs and input sharing. For example,
as a specific wholesale trade sector, the merchant wholesalers–non-durable goods sector may act as an
intermediate between the manufacturing and retail sector so that the transport costs of goods may take a
large share of the firms' production costs. Thus, by co-locating with each other, wholesalers can benefit
from reductions in transport costs, repair and maintenance costs, and other operation costs by sharing
transport and storage capacity (Heuvel et al. 2013). Figure 3-3 indicates that new firms in this sector are
more concentrated in the least dense centers where land rents and within-center congestion levels are
expected to be lower. Similarly, by co-locating within centers, printing firms may benefit from
information and intermediate input sharing. However, new firms in this sector are mostly concentrated in
the least dense centers and are least concentrated in the densest centers. This implies that these
production-related sectors are more sensitive to congestion costs and are less likely to be attracted to
highly congested centers. Since shipments and scheduled deliveries are important for production-related
industries, firms in these sectors are also expected to avoid centers having long delays to other
destinations in the region.
94
Moreover, new firms in one local service sector—the clothing and clothing accessories stores—
are also found to favor centers. This sector takes a relatively large share of employment for the base year
(about 1.23 percent). The agglomeration mechanism for this sector could also be attributable to the
shopping externalities besides the production-related agglomeration benefits (Anas and Kim 1996). This
sector is also expected to be more sensitive to congestion delays at the regional scale or within centers.
4.2 Transportation network data
The transportation network data for the Los Angeles region are obtained from the Southern
California Associations of Governments (SCAG) base-year network files for the Regional Transportation
Plan (RTP) of 2003. The road network files contain detailed information for each link and node, including
length (miles), lane miles, and capacity characteristics. The Regional Transportation Model also estimated
the travel time and speed of each link for four time periods in a day (AM peak, mid-day, PM peak, and
the night period) after assigning to the road network the vehicle volumes for all trip purposes, which are
estimated from the contemporary land-use pattern, and are validated by comparing them with the actual
average daily traffic counts across a set of screen lines.
14
As discussed, the AM peak is defined as the
congested condition, while the night period (NT) is defined as the free-flow condition. To calculate the
travel time between each pair of census tracts under the two scenarios, the centroid of each census tract is
assigned to the closest network nodes falling within them, and the shortest-path algorithm is applied.
4.3 Land use data
The land-use information is obtained from SCAG's region-wide land-use (LU) dataset, which
includes land-use codes and areas for each parcel in different survey years.
15
To quantify land-use
characteristics within each census tract and center, the 2001 parcel-level data are spatially joined to the
2000 census tract. The areas of each land-use type are then summed for each employment center. To
assign the land-use types to corresponding economic activities, I categorize land use into three broad

14
Source: SCAG 2003 Model Validation Summary Final Report
15
The survey years include 1993, 1995, 2001, 2005, and 2008.
95
types—industrial and transportation use, office and commercial use, and facilities and education use—
which correspond to the three broad economic sectors of production-related industries, commercial and
private services, and education and public services (see Table 3- a5 in the appendix).
16

4.4 Population data
The data on the socio-demographic profiles of the population and households at the census-tract
level come from the 2000 U.S. Census of the Population. Relevant information such as age structure, the
education level of the local population, household income levels, and housing values/rents are obtained.
The summary statistics of all the explanatory variables are included in Table 3-a6 (in the
appendix). The distribution pattern of the spatial variables is discussed further in the next section.
5 Results
This section starts with a brief descriptive analysis to explore the relationship among firm
location, agglomeration effects, and congestion costs. I also explore the spatial pattern of accessibility and
congestion and their relation to other agglomeration measures. The purpose of the descriptive analysis is
to detect a possible multicollinearity problem. If different attributes of the centers were not significantly
correlated, firms would be able to choose among centers with different bundles of locational attributes so
that the individual effects could be separately assessed. This is especially important for the study purpose,
which is to unravel the effects of congestion costs from agglomeration benefits. In the second part, the
estimation results of the discrete choice model are presented and discussed.
5.1 Descriptive results
5.1.1 Urbanization effects
As suggested by the theory, employment centers with a higher level of economic concentration
should take a larger share of new establishments in the same sector. Figure 3-4 indicates the distribution
of new establishments across centers of different size categories. Consistent with our expectations, large-

16
The categorization of land-use types and the corresponding economic sectors derive from e Silva et al. (2014).
96
sized centers attract a larger share of new establishments. A few sectors, such as motion pictures, legal
services, and accounting service industries have a slightly larger portion of firms locating in the 20k–50k
size group than in the larger 50k–100k size group. The distribution of all the new establishments in the 34
sectors that meet the LQ threshold (LQ of center locations>1.1) also indicates a similar trend where the
largest centers attract more than 60 percent of the new businesses.

Figure 3-4 Distribution of new establishments across centers of different size groups

Figure 3-5 indicates the distribution of new establishments across centers of different density
levels, which is divided by 3-quantiles of the centers' employment density. Similarly, a consistent trend is
found in that centers of higher employment density usually attract a larger share of new establishments in
the aggregate 34 sectors or in any specific industrial sector.
All 34
sectors
Legal
Services
Motion
pict.
Advertis ing Publishing
Clothing
stores
Securities
Merchant
wholesalers
Account ing Managem.
Telecommu
nic.
Computer
Syst. design
Printing
>100K Jobs 68.8 69.8 79.1 68.8 66.3 71.7 66.0 70.6 60.9 63.7 59.5 67.7 61.5
50K‐100K Jobs 13.5 12.0 8.3 13.0 17.8 10.6 14.4 13.5 15.5 16.2 23.2 13.9 18.9
20K‐50K Jobs 12.6 14.3 10.1 12.6 10.4 11.4 14.1 11.1 18.2 13.6 12.2 12.6 14.0
10K‐20K Jobs 5.0 3.9 2.5 5.6 5.6 6.3 5.5 4.7 5.5 6.5 5.1 5.7 5.6
0
10
20
30
40
50
60
70
80
90
Percentage of new establishments within centers of different size category
97

Figure 3-5 Distribution of new establishments within centers of different density (jobs per acre,
by 3-quintiles)

5.1.2 Localization effects
To explore whether employment centers that are more concentrated in a particular sector are more
attractive to new firms in the same sector, centers are categorized into 3-quantiles for each broad NAICS2
sector based on the value of the centers' employment share for each NAICS2 sector (the
variable). The summary statistics of each variable within the different 3-quantiles
are listed in Table 3-a7 (in the appendix). The distribution of new establishments in each chosen sector is
then examined across the three types of centers.
Figure 3-6 indicates a general trend that centers having higher concentrations of employment in a
NAICS2 sector are also more attractive to new firms in the detailed sectors within the same 2-digit sector.
This trend is true for the aggregate result of the aggregate 34 sectors specialized within centers, and for all
the office-related sectors, including the information, finance and insurance, and professional service
sectors. For example, the 15 centers belonging to the top 3-quantiles in terms of concentration in the
information sector (NAICS2 = 51) attract more than 50 percent of new publishing firms (NAICS3 = 511),
80 percent of new motion picture firms (NAICS3 = 512), and about 60 percent of new telecommunication
firms (NAICS3 = 517) choosing center locations.
All 34
sectors
Legal
Services
Motion
pict.
Advertis ing Publishing
Clothing
stores
Securities
Merchant
wholesalers
Account ing Managem.
Telecommu
nic.
Computer
Syst. design
Printing
High 73.1 80.7 82.0 73.1 72.0 75.4 73.6 69.5 67.2 69.5 69.1 66.1 60.2
Medium 17.6 14.9 15.6 18.2 19.6 14.5 17.6 16.6 24.0 19.5 19.2 24.2 23.1
Low 9.3 4.3 2.4 8.7 8.4 10.1 8.8 14.0 8.8 11.0 11.7 9.8 16.6
0
10
20
30
40
50
60
70
80
90
Percentage of new establishments within centers of different density category
98
Other sectors show different patterns of distribution. New firms in the wholesalers of non-durable
goods sector (NAICS3 = 424), the clothing and clothing accessory stores sector (NAICS3 = 448), and the
printing sector (NAICS3 = 323) are more attracted to centers that are moderately concentrated in their
corresponding NAICS2 sectors. This might imply that firms in these aggregate NAICS2 sectors differ
greatly from each other so that locating close to each other does not necessarily generate localization
benefits, while the competition for land and other resources within clusters may generate diseconomies.
For example, printing firms may not benefit from locating close to other heavy manufacturing firms that
have no direct input–output linkages with them, while industrial land scarcity and pollution generated by
the latter might drive away new printing firms.

Figure 3-6 Distribution of new establishments across centers of different levels of local
specialization in the same 2-digit sectors

5.1.3 Congestion costs vs. local concentration
The spatial pattern of metro-wide traffic congestion measured by the delay rate is shown in
Figure 3-a1 (in the appendix), with the darker color representing higher quintiles of congestion level. The
boundaries of the employment centers are also plotted as a reference. The map indicates that the severest
congestion delays mainly concentrate in a few areas. The first area starts in the downtown area, extends
All 34
sectors
Legal
Services
Motion
pict.
Advertising Publishing
Clothing
stores
Securities
Merchant
wholesalers
Accounting Managem.
Telecommu
nic.
Computer
Syst. design
Printing
High 54.8 74.3 84.5 73.2 54.9 13.4 78.1 24.7 62.6 68.0 60.6 67.3 22.9
Medium 37.8 22.6 13.8 21.3 36.4 75.1 13.5 68.3 30.5 24.8 30.3 24.3 68.4
Low 7.4 3.0 1.7 5.5 8.8 11.5 8.4 6.9 6.9 7.2 9.1 8.5 8.7
0
10
20
30
40
50
60
70
80
90
Percentage of new establishments within centers of different concentration levels for each
NAICS2 sector
99
southeast along the I-5 (Santa Ana Freeway) until meeting the S-91 Freeway, and extends south along the
I-710 (Long Beach Freeway). Another highly congested area spreads along the I-405 corridor, including
the intersection of I-405 and I-101 (Ventura Freeway), the intersection of I-10 and I-405 (San Diego
Freeway) near Santa Monica City, and the area around LAX.
Figure 3-a1 also indicates that census tracts whose congestion levels are highest are usually a
component unit of employment centers. A rough visual examination shows that about 24 centers have
their boundaries overlapping with areas of the highest quintile of congestion. To examine whether center
locations are more congested than non-center locations, Table 3-1 presents the summary statistics of
regional congestion measures using the 12-mile distance boundary for the component tracts of centers and
other non-center tracts. As expected, delay rates are always higher for center locations at all percentiles.
Table 3-1 Summary statistics of congestion levels at different locations (12-mile boundary)
Location meansd min p10p25p50p75 p90 max
Center tracts (n = 378) 0.56 0.14 0.16 0.37 0.45 0.59 0.67 0.73 1.04
Non-center tracts (n = 2972) 0.37 0.18 0 0.12 0.23 0.38 0.5 0.61 0.89

Furthermore, to examine the relationship between congestion levels and the scale of centers, both
the regional congestion measure and the within-congestion measure are plotted with the employment
size/density of centers (see Figures 3-7 and 3-8).
Figure 3-7 shows that centers' regional congestion levels are weakly correlated with their
employment size or density. This might be explained by the fact that a large portion of the traffic that
flows on the road network in the Los Angeles region consists of "through traffic," so that the congestion
pattern at the regional scale is weakly determined by local travel demand and local economic
concentrations. Figure 3-8 shows that congestion delays within centers are more relevant to the scale
rather than the intensity of economic activities within centers. However, the correlation between within-
center congestion and center size is only moderate, and this is not entirely consistent with general
100
predictions in urban economic theory that larger agglomerations are associated with high congestion costs.
This might again suggest the weak role of local travel demands in determining the intensity of road use.

Figure 3-7 Center size/density vs. regional congestion level (12-mile boundary applied)

Figure 3-8 Center size/density vs. within-center congestion level

5.1.4 Accessibility pattern
The accessibility variables are used to measure the proximity to the labor force and to firms in
related sectors, which are defined as the same NAICS2 sector. Figures 3-a2 to 3-a8 (in the appendix)
show the spatial pattern of accessibility variables measured using the 12-mile boundary. Similarly, all
census tracts are categorized by quintiles based on their accessibility levels, with darker colors
representing higher quintiles.
Cor(Delay Rate, EmpSize)=0.22
.2 .3 .4 .5 .6 .7
， Delay Rate (min per mile) 12-mile boundary
0 500000 1000000 1500000
Employment size of centers
Cor(Delay Rate, EmpDen)=0.20
.2 .3 .4 .5 .6 .7
Delay Rate (min per mile), 12-mile boundary
10 20 30 40 50 60
Employment density of centers (jobs per acre)
Cor(Delay Rate, EmpSize)=0.47
0 .2 .4 .6 .8
Delay Rate (min per mile)
0 500000 1000000 1500000
Employment size of centers
Cor (Delay Rate, EmpDen)=-0.15
0 .2 .4 .6 .8
Delay Rate (min per mile)
10 20 30 40 50 60
empden
101
Figure 3-a2 (in the appendix) indicates that the spatial pattern of labor-force accessibility
resembles concentric rings, and the locations with the highest level of labor-force accessibility
concentrate in a core area of the S-134 Freeway in the north, the I-405 Freeway in the west, the S-91
Freeway in the south, and the I-605 Freeway in the east. The spatial patterns of access to employment in
manufacturing, and in the wholesale trade and retail trade sectors are very similar to that of labor-force
access, with the level of accessibility declining as the distance to the core area increases. This result
implies that employment centers locating in the core area will be attractive to those firms valuing metro-
wide access to the labor force or to firms of the same kind in the three sectors.
The spatial patterns of access to employment in information, the finance and insurance, and
professional service sectors are very different from that of labor-force accessibility. Instead of showing
patterns of concentric rings, Figures 3-a3 to 3-a8 (in the appendix) indicate potential divisions of market
areas served by two groups of centers: those locating along the LA downtown–Hollywood–Santa Monica
corridor and those large centers in Orange County. The I-605 Freeway seems to serve as a natural
boundary between the two market areas. This provides evidence on the polycentric nature of the Los
Angeles region, where, except for centers in the downtown area, large centers in Orange County are also
important in providing business services. The result also implies that firms valuing metro-wide access to
business services face multiple choices in terms of employment centers.
The maps also show that not all the centers locate in the most accessible areas, while not all tracts
belonging to the highest quintile accessibility levels are component units of employment centers. To
explore this point further, summary statistics of accessibility variables for center and non-center locations
are calculated (see Table 3-a8 in the appendix). As expected, the average values of all accessibility
variables for component tracts of centers are higher than for non-center locations. This implies that
centers are on average more accessible than those non-center locations.
102
Comparing the maps of accessibility patterns and the regional congestion pattern, I also observe
that not all the highly accessible areas are associated with high levels of congestion. However, the
correlation coefficients between the regional congestion measure and all the accessibility measures are
moderately high, ranging from 0.53 to 0.77 (see Table 3-a9 in the appendix). The relative labor-force
accessibility variable is weakly correlated with congestion variables and other accessibility variables.
These results imply that regional accessibility measures and regional congestion measures are not
perfectly aligned, and the construction of the two variables enables the metro-wide agglomeration benefits
and congestion costs to be evaluated separately.
To further examine the relationship between accessibility, congestion, and other agglomeration
effects, the correlation matrix for the related variables is represented in Table 3-a9 (in the appendix). It is
easily identifiable from the table that the correlation coefficients between accessibility variables and
urbanization/localization variables are low. This implies that the two groups of variables measure
agglomeration benefits at different scales: the former measures metro-wide agglomeration benefits
accruing to firms, while the latter measures agglomeration benefits associated with economic
concentrations of centers.
5.2 Estimation results
As discussed in Section 3.3, the expected profit of firm i in sector m at center j is constructed as a
function of agglomeration, accessibility, and congestion effects, controlling for transport-network and
land-use characteristics. Considering the interactions between the spatial variables, the two following
specifications of the expected profit function are used:
,

_

_

(3-8)
103
,

_

_

(3-9)
Model 1 (equation 3-8) includes key variables of the centers' size and density, access to
employment in the same sector, and congestion measures. Model 2 (equation 3-9) includes the relative
accessibility of the labor force but drops the center size variable, because the former variable is
constructed to take the effect of the latter variable. The quadratic term of the within-center congestion
variable (_ ) is also entered into the function to identify whether an upper threshold
exists for individual firms to endure congestion costs within centers. The two models are run for each
chosen sector as well as the aggregate 34 sectors meeting the LQ threshold.
To estimate the discrete choice model, a randomly drawn choice-set approach is applied here to
reduce the computational burden of estimation with many alternatives. As suggested by McFadden (1978),
the use of random subsets of the full choice set has no effect on the consistency of parameter estimates in
a conditional logit framework given that the IIA assumption is held. Previous studies on discrete choice
models with many alternatives prove that 12.5 percent of the full choice set should be used as a minimum
in the conditional logit model to produce relatively stable coefficients (e.g. Feather 1994; Nerella and
Bhat 2004; Parsons and Kealy 1992). In this study, about 20 percent of the non-selected alternatives is
used to construct the individual choice set that for each new establishment, 9 alternative centers are
randomly drawn (without replacement) from the full choice set except the chosen center, and then added
the center actually chosen is added to that set.
17
Because the randomly drawn choice-set approach is used,
no alternative-specific constants are reported here and there is no base alternative (Waddell and Ulfarsson
2003).
Tables 3-2 and 3-3 report the estimation results for the two models. The coefficients for
congestion measures are of greatest interest in this study. Results suggest that congestion delays at the

17
I also construct individual choice sets with 12.5%, 50%, and 100% of the full choice set. However, the estimated coefficients
do not change much in terms of magnitude and significance, and are thus not reported here.
104
regional scale show negative effects on the centers' attraction for new firms in general as well as in most
of the sectors. This is consistent with existing empirical evidence that congestion delays increase firms'
costs indirectly by increasing workers' commuting costs and the costs for other business-related trips.
Among the three information sectors (motion pictures, publishing, and telecommunications), stronger
aversion to congestion is shown by the motion picture sector. Among all the professional service sectors,
the legal services sector is most sensitive to congestion delays. However, there is some evidence that
centers with high levels of regional congestion seem to attract firms specialized in wholesaler–non-
durable goods. This could correspond to the corridor of I-5 Freeway starting from downtown LA to the
intersection of I-5 and S-91 Freeway, which has a higher level of congestion, but is also the location for
most of the centers where wholesale trade firms concentrate. This result to some extent implies that a
higher concentration of wholesale trade firms at those locations contributes to the higher levels of
congestion there. Moreover, AM peak congestion delays show no effect on the location choices of
publishing firms, clothing stores, and printing firms. Since journey-to-work trips take a large share of the
total number of personal trips and might be most relevant to the metro-wide congestion effects on firms,
this result implies that regional labor-force access is less important for firms in the two sectors. To
examine if the coefficients for the congestion variables are contaminated by the correlation between the
congestion measures and accessibility variables, I also rerun all the regressions by dropping the
accessibility variables. The estimation results show that the coefficients for the congestion variables do
not change in terms of magnitude and significance, and they are thus not reported here.
105
Table 3-2 Estimation results of firm location choice models, 2001-2005 (1)

All 34
sectors
Legal
services
Motion
pict.
Advertis-
ing
Publish-
ing
Clothing
stores
Securities
Merchant
wholesaler
Account-
ing
Managem.
Telecomm-
unic.
Computer
Syst.
design
Printing
EmpSize
0.0031
***
0.0023
***
0.0029
***
0.00324
***
0.00401
***
0.00369
***
0.00294
***
0.00410
***
0.00236
***
0.00274
***
0.00346
***
0.00288
***
0.00384
***

(92.78) (17.10) (23.94) (15.54) (17.07) (31.58) (34.61) (34.07) (13.48) (17.13) (20.15) (13.88) (16.45)
EmpDen
0.00397
***
0.0340
***
0.00995 0.00376 -0.0297
**
-0.0177
***
-0.0273
***
0.00454 0.0023 -0.0146
*
-0.0398
***
-0.011 -0.0141
(3.36) (6.64) (1.49) (0.47) (-3.03) (-3.64) (-6.58) (1.10) (0.31) (-2.35) (-5.20) (-1.35) (-1.86)
Indu 0.0250
***
0.0529
***
0.0399
***
0.0339
***
0.0523
***
0.00147 0.0905
***
0.0924
***
0.0171 0.0363
***
0.0375
***
0.0355
***
-0.0031
Share (24.21) (8.60) (6.82) (3.44) (4.82) (0.31) (17.06) (16.66) (1.93) (4.99) (4.73) (3.72) (-0.62)
BizAcc_ 0.00353
***
0.0005 0.0156
***
0.0009 0.0006 0.00626
***
0.00642
***
-0.00126 -0.0004 0.00298
**
0.00236 0.00264
*
-0.00287
**

NAICS2 (15.97) (0.54) (11.92) (0.75) (0.32) (4.59) (6.05) (-1.22) (-0.32) (3.12) (1.70) (2.09) (-2.93)
Delay -1.717
***
-2.419
***
-3.968
***
-1.938
***
-0.686 -0.021 -1.438
***
1.659
***
-1.457** -1.775
***
-0.555 -2.252
***
-0.473
Rate (-18.15) (-6.19) (-8.87) (-3.59) (-1.08) (-0.05) (-5.62) (4.47) (-3.02) (-4.42) (-1.16) (-4.36) (-0.75)
Delay
Rate_inner
7.419
***
6.852
***
9.008
***
10.23
***
14.18
***
4.030
***
8.517
***
5.954
***
8.555*** 9.008
***
8.507
***
15.17
***
12.21
***

(35.97) (8.09) (10.52) (7.39) (9.66) (6.86) (15.60) (9.37) (6.88) (9.03) (7.58) (10.59) (9.16)
(Delay
Rate_inner)
2

-5.524
***
-2.269
*
-7.440
***
-8.129
***
-15.33
***
-4.223
***
-6.006
***
-7.567
***
-4.893** -7.490
***
-7.630
***
-13.60
***
-12.84
***

(-20.28) (-2.00) (-6.93) (-4.44) (-7.63) (-5.06) (-8.57) (-8.32) (-3.04) (-5.45) (-5.02) (-7.25) (-7.04)
DistHwy
-0.0511
*
-0.321
***
-0.0977 0.135 -0.0257 0.00298 0.0111 -0.106 -0.148 -0.246
**
-0.146 0.17 -0.0447
(-2.39) (-4.01) (-0.84) (1.18) (-0.20) (0.04) (0.21) (-1.52) (-1.38) (-2.77) (-1.30) (1.69) (-0.66)
DistAP
-0.0258
***
-0.0693
***
0.00963 -0.0536
***
-0.0350
**
0.0378
***
-0.0659
***
0.0137
**
-0.0341
***
-0.0400
***
-0.0366
***
-0.0232
*
-0.0454
***

(-15.64) (-10.16) (1.07) (-5.31) (-3.16) (7.17) (-13.86) (3.10) (-3.88) (-5.38) (-4.24) (-2.43) (-4.81)
Per_LU
0.0136
***
0.0755
***
-0.0172 0.0288
*
0.104
***
0.0447
***
0.0492
***
-0.00633
**
0.0594
***
0.0669
***
0.0756
***
0.0936
***
0.00792
(20.59) (9.67) (-1.57) (2.32) (6.69) (7.92) (8.34) (-3.29) (5.45) (7.32) (6.08) (7.83) (1.82)
Per_baplus
0.0105
***
-0.0031 0.0423
***
0.0145
**
0.0160
*
-0.00201 -0.000299 -0.0180
***
0.0202
***
0.00348 0.00336 0.00813 -0.0337
***

(13.85) (-0.85) (10.13) (2.68) (2.49) (-0.79) (-0.13) (-5.44) (4.04) (0.86) (0.70) (1.45) (-5.20)
N 455630 34760 47110 15280 10590 60850 69890 57350 15510 22170 16430 13230 11320
choosers 45563 3476 4711 1528 1059 6085 6989 5735 1551 2217 1643 1323 1132
pseudo R
2
0.461 0.486 0.716 0.472 0.429 0.518 0.435 0.501 0.375 0.37 0.419 0.369 0.316
ll -56597.9 -4110.4 -3085.8 -1859.4 -1392.9 -6759 -9090.3 -6588.6 -2232.3 -3217.5 -2198.2 -1922.1 -2606.5
ll_0 -104912.7 -8003.8 -10847.5 -3518.4 -2438.4 -14011.2 -16092.8 -13205.3 -3571.3 -5104.8 -3783.1 -3046.3 -1782.8
chi2 96629.6 7786.7 15523.4 3317.8 2091.2 14504.4 14004.9 13233.5 2678 3774.7 3170 2248.5 1647.4
t statistics in parentheses; * p<0.05, ** p<0.01, *** p<0.001

106
Table 3-3 Estimation results of firm location choice models, 2001-2005 (2)

All 34
sectors
Legal
services
Motion
pict.
Advertis-
ing
Publish-
ing
Clothing
stores
Securities
Merchant
wholesaler
Account-
ing
Managem.
Telecomm
-unic.
Computer
Syst.
design
Printing
RAccLF
0.00372
***
0.00276
***
0.00358
***
0.004
***
0.005
***
0.00474
***
0.00363
***
0.0051
***
0.00288
***
0.0034
***
0.00437
***
0.00353
***
0.00485
***

(89.80) (16.47) (23.78) (15.05) (16.41) (31.43) (33.59) (33.07) (13.05) (16.67) (19.78) (13.36) (15.86)
EmpDen
0.00607
***
0.0323
***
0.0071 0.0007 -0.033
***
-0.0208
***
-0.0308
***
0.00659 0.0007 -0.0182
**
-0.0435
***
-0.0139 -0.0113
(5.18) (6.23) (1.05) (0.09) (-3.32) (-4.25) (-7.30) (1.63) (0.10) (-2.87) (-5.56) (-1.69) (-1.51)
Indu
Share
0.0248
***
0.0552
***
0.0410
***
0.0355
***
0.0522
***
-0.0009 0.0898
***
0.091
***
0.0187
*
0.0379
***
0.0376
***
0.0378
***
-0.001
(23.97) (8.99) (6.97) (3.59) (4.67) (-0.18) (16.81) (16.33) (2.11) (5.21) (4.63) (3.95) (-0.27)
BizAcc_
NAICS2
0.00337
***
0.0007 0.0152
***
0.001 0.0007 0.00426
**
0.00748
***
-0.002 -0.0002 0.00318
***
0.0025 0.00281
*
-0.0035
***

(15.26) (0.80) (11.76) (0.99) (0.37) (3.08) (7.07) (-1.82) (-0.14) (3.34) (1.80) (2.24) (-3.53)
Delay
Rate
-1.939
***
-2.518
***
-3.928
***
-2.094
***
-0.8 0.378 -1.622
***
1.592
***
-1.556
**
-1.867
***
-0.588 -2.385
***
-0.468
(-20.57) (-6.40) (-8.72) (-3.86) (-1.24) (0.96) (-6.32) (4.32) (-3.21) (-4.63) (-1.22) (-4.60) (-0.74)
Delay
Rate_inner
7.887
***
7.156
***
8.741
***
10.65
***
14.83
***
4.273
***
9.094
***
6.555
***
8.862
***
9.420
***
8.980
***
15.52
***
13.43
***

(37.95) (8.32) (10.29) (7.62) (10.03) (7.26) (16.48) (10.14) (7.03) (9.32) (7.96) (10.72) (9.67)
(Delay
Rate_inner)
2

-6.091
***
-2.565
*
-7.229
***
-8.714
***
-16.21
***
-4.874
***
-6.771
***
-8.619
***
-5.270
**
-8.094
***
-8.495
***
-13.98
***
-14.57
***

(-22.12) (-2.22) (-6.79) (-4.68) (-7.96) (-5.77) (-9.49) (-9.21) (-3.22) (-5.77) (-5.52) (-7.33) (-7.58)
DistHwy
-0.0715
***
-0.342
***
-0.139 0.0846 -0.085 0.0132 -0.0365 -0.119 -0.172 -0.280
**
-0.188 0.126 -0.0424
(-3.35) (-4.29) (-1.19) (0.75) (-0.65) (0.19) (-0.69) (-1.72) (-1.62) (-3.19) (-1.69) (1.25) (-0.62)
DistAP
-0.0318
***
-0.074
***
0.0039 -0.059
***
-0.043
***
0.03
***
-0.0707
***
0.0108* -0.0374
***
-0.0446
***
-0.0417
***
-0.0286
**
-0.05
***

(-19.15) (-10.87) (0.43) (-5.86) (-3.91) (5.60) (-14.78) (2.42) (-4.27) (-6.04) (-4.82) (-3.02) (-5.49)
Per_LU
0.0142
***
0.084
***
-0.009 0.0414
***
0.121
***
0.0486
***
0.0612
***
-0.00683
***
0.0676
***
0.0777
***
0.0902
***
0.106
***
0.0065
(21.50) (10.60) (-0.87) (3.32) (7.68) (8.59) (10.28) (-3.56) (6.13) (8.40) (7.18) (8.73) (1.49)
Per_baplus
0.0115
***
-0.005 0.0421
***
0.0132* 0.0143* -0.003 -0.0015 -0.0176
***
0.019
***
0.0023 0.0025 0.0062 -0.034***
(15.20) (-1.36) (10.07) (2.41) (2.21) (-1.10) (-0.62) (-5.35) (3.79) (0.55) (0.51) (1.11) (-5.23)
N 455630 34760 47110 15280 10590 60850 69890 57350 15510 22170 16430 13230 11320
choosers 45563 3476 4711 1528 1059 6085 6989 5735 1551 2217 1643 1323 1132
pseudo R
2
0.458 0.485 0.716 0.47 0.425 0.518 0.433 0.5 0.374 0.369 0.419 0.367 0.313
ll -56835.5 -4119.2 -3085.5 -1865.3 -1401.7 -6748.7 -9119.3 -6607.2 -2237.4 -3222.4 -2198.7 -1928.2 -2606.5
ll_0 -104912.7 -8003.8 -10847.5 -3518.4 -2438.4 -14011.2 -16092.8 -13205.3 -3571.3 -5104.8 -3783.1 -3046.3 -1790.6
chi2 96154.4 7769.2 15524 3306.2 2073.5 14525.1 13947 13196.2 2667.9 3764.9 3168.9 2236.3 1631.8
t statistics in parentheses; * p<0.05, ** p<0.01, *** p<0.001

107
Results on the within-center congestion delays suggest that the quadratic form produced
significant estimates for all the chosen sectors and the aggregate 34 sectors. The negative and significant
coefficients for the quadratic term suggest that a congestion threshold exists for firms to endure the
congestion delays inside centers: before the threshold, economic interactions within centers may still be
efficient given the relative proximity to suppliers and customers within centers; beyond the threshold, a
further increase in congestion costs would diminish the agglomeration benefits of centers so that firms are
more likely to be driven away from those centers that have long delays for short-distance trips.
To examine this, I use the estimated coefficients to roughly calculate the "turning point" where
the _ variable reaches the maximum value. Figure 3-9 shows that for the regression
results of most sectors, the _ variable usually yields the maximum value at more than
the 75th percentile value (= 0.41 min per mile) of the variable. The only exception is the wholesaler–non-
durable goods sector. The publishing sector and the printing sector are also shown to have lower
congestion thresholds that are less than the 90th percentile value (= 0.56 min per mile) of the variable. On
the other hand, the legal services sector seems to be least sensitive to within-center delays, such that the
estimated threshold beyond which the centers' attractiveness for firms in the sector would be diminished
is very high, exceeding the maximum value of the variable. However, the two sectors with the 2nd and
3rd highest within-center congestion delay threshold are the accounting and securities sectors,
respectively, rather than the motion picture sector and the advertising sector. This might be because firms
in the two sectors are more diversified and composed of various office-related activities: although the two
sectors are generally less concentrated in denser centers, some firms in the two sectors might benefit more
from economies of proximity, and so would be more likely to endure congestion delays.
In general, Figure 3-9 does not indicate that industrial sectors ranking low in terms of
concentration within the densest centers always have lower congestion thresholds. This result does not
conflict with the results of the descriptive analysis: since congestion delays within centers are weakly
correlated with the employment density of centers, denser centers are not necessarily more traffic

108
congested, but could still be less attractive to firms in some sectors due to other diseconomies' effects,
such as higher land-use intensity, and higher land costs, while less dense centers could be traffic
congested but offer other locational advantages that attract firms in some sectors. However, when
comparing across broad economic sectors, the result still provides some evidence that firms relying on
face-to-face interaction and experiencing continuing benefits of agglomeration, such as those with office-
related activities, are less responsive to congestion delays within centers than are firms involved in
production-related activities (i.e. wholesalers and the printing sector) or the retail sector.

Figure 3-9 Estimated turning points for the Delay Rate_inner variable
In terms of urbanization effects, the results from all of the models indicate that the employment
size of the centers exerts a positive influence on the location choices of firms. This suggests that, all else
being equal, firms appreciate the benefits of the scale of economic concentration. However, the
magnitude and significance for the employment density variable suggests that only legal services firms
are strongly affiliated with centers of high employment density. Firms that belong to the motion picture,
advertising, and accounting sectors, and the two production-related sectors, do not show any preference to
locate in dense centers, while firms in other sectors tend to avoid locating in dense centers. The results
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
Congestion Threshold
model
1
model
2

109
imply that with other things being equal, the negative effects of employment density, such as those
associated with high land costs and intense competition between firms, dominate the location choices of
firms in most sectors.
The localization effect is assessed with the variable. Positive and significant
coefficients for this variable are found for the aggregated 34 sectors, and for most of the chosen finer
sectors. Exceptions are firms in the clothing store sector, the accounting sector, and the printing sector. It
might be possible that new firms in the accounting service sector are more customer-oriented and do not
show a preference for centers concentrated with other professional service activities when other locational
factors are controlled. Similarly, new clothing stores or new printing firms might have weak economic
ties with other retail firms or manufacturing firms in general, so that they do not prefer centers with high
levels of concentration in the broad NAICS2 sectors, all else being equal.
Accessibility variables consist of the relative labor-force accessibility variable and access to
employment in the same NAICS2 sectors, all of which are measured using the 12-mile distance threshold.
Positive and significant coefficients are observed for the variable for all sectors, implying that
"net labor access" that takes into account competition for workers from other centers or job locations is
important for centers to attract new firms in all sectors. This result is consistent with earlier findings by
Giuliano et al. (2011) that relative labor access is the most important access measure for employment
center growth and that centers with higher net labor access attract more growth.
In terms of the effects of access to other related businesses, the results are mixed. The results
suggest that firms in general favor centers that are more accessible to other firms of the same NAICS2
sector at the regional scale. However, results from the separate models for each detailed sector show that
centers with better regional access to businesses in related fields seem to attract firms in some sectors,
including the motion picture, clothing store, securities, computer system design, and management service
sectors. Except for the printing sector, firms in other sectors do not exhibit any preference to locate in

110
centers with better regional business access. The explanation for the results could vary among different
sectors. Some sectors showing no significant relationship between access to other businesses and new
firm births across employment centers, however, do have positive and significant estimated coefficients
for the variables. Examples include firms in the legal services, advertising, publishing,
telecommunication, and wholesale trade sectors. It might be possible that firms in these sectors only value
proximity to other related businesses at the localized scale (e.g. within the same center) rather than at the
regional scale. Results also show that printing firms tend to avoid centers with better regional access to
other manufacturing firms, which implies that these firms experience diminishing returns on metro-wide
localization economies.
Transport-access effects are measured by distance to a highway intersection and distance to the
nearest airport. The estimated coefficients for suggest that firms in most sectors exhibit no
preference for centers with better access to highway ramps. This result does not surprise us, since
highway network access is ubiquitous across the Los Angeles region, and most of the centers locate near
highways. On the other hand, negative and significant coefficients for the variable are found in
the regression results of the aggregate 34 sectors and most of the chosen sectors. This suggests that firms
in most sectors tend to locate in centers in close proximity to commercial airports.
In terms of land-use characteristics, results for the aggregate 34 sectors suggest that a larger
percentage of the land area being in relevant use generally increases the centers' attractiveness to new
firms in corresponding sectors. Specifically, results suggest that a larger percentage of land area being in
office and commercial use increases the centers' attractiveness for most office firms and clothing stores.
However, new firms in the non-durable goods–wholesale trade sector tend to avoid those centers having a
large percentage of land area developed into industrial uses, while firms in the printing sector show no
preference for centers with industrial space. The result implies that industrial land use in general may not
fit the needs of specific production-related activities and may even constrain further development of
industrial activities.

111
Finally, results related to the labor-quality variable (_ ) suggest that firms in general
favor centers with a larger percentage of highly educated worker residents. Results of separate regressions
for each sector indicate that new firms in the motion picture, advertising, publishing, and accounting
sectors tend to gravitate toward centers with a more educated labor force. However, this is not case for
firms in the two production-related industries, as indicated by the negatively significant estimated
coefficients for the variable. This is because the variable could also be used as a proxy for local wage
rates or even land prices. Thus, printing firms and merchant wholesalers tend to avoid centers where land
costs and labor costs would be more expensive.
6 Conclusions
This study examines whether or not congestion-induced travel-time delays would have an impact
on the intra-metropolitan location decisions of firms. It also investigates whether or not firms in different
sectors experiencing different levels of agglomeration economies vary in their response to congestion
delays at both the local scale and regional scale. A discrete choice model where spatial alternatives are
employment centers in the Los Angeles metropolitan area is applied to examine the nature of the
polycentric structure of the area from the firm location-behavior basis. Equations are estimated explaining
the location decisions of all firms concentrated in employment centers, as well as firms in finer industrial
sectors that are concentrated in employment centers in the 2001–2005 study period. The major findings
from the estimations are summarized as follows.
1) Congestion delays at the regional scale significantly affect the location decisions of firms in
general and firms in most of the finer industrial sectors. The negative location effects are the strongest,
both in terms of magnitude and significance, for firms specialized in office-related activities, such as the
motion picture, legal services, and telecommunication sectors, which might place a higher value on the
costs of access to workers. Firms in other broad economic sectors, such as manufacturing and retail, do
not show an aversion to locating where regional access is constrained by long travel-time delays. Firms

112
specialized in the wholesale trade sector are attracted to those congested locations which also concentrate
most of the existing wholesale trade activities.
2) Congestion delays within employment centers are significant location factors for all firms
concentrated in centers and for firms in all the finer sectors. The estimated coefficients for the within-
center congestion measure and its quadratic term suggest that a threshold exists beyond which traffic
congestion delays begin to outweigh the agglomeration benefits of centers and this constrains further
growth of the centers. Firms in office-related activities, especially those valuing agglomeration economies
derived from proximity to other related firms, usually have higher congestion thresholds, and are less
sensitive in response to congestion delays than firms are in the manufacturing, retail, and wholesale trade
sectors.
3) Employment size is important in terms of the probability of firms locating in a center.
However, the employment density of centers only increases their attractiveness to new legal services
firms. New firms specialized in other office-related activities, such as the publishing, securities, and
management sectors, and new clothing store firms tend to locate away from those high-density centers
due to possible high land costs and competition. The production-related sectors do not show any
sensitivity to center densities.
4) Localization economies at the local scale significantly increase the centers’ attractiveness to
firms in most sectors. However, metro-wide localization economies, as measured by the regional access
to businesses in related sectors, are significant location factors for firms in a few office-related activities
and retail trade sectors.
5) Transport access to highways and arterials does not significantly impact the intra-metropolitan
location decisions of firms in most sectors. On the contrary, spatial proximity to airports strongly predicts
location choices among centers within the metropolitan area for firms in most sectors.

113
6) New office firms and retail firms are attracted to employment centers that already have a large
share of land developed into office and commercial uses, while new wholesalers tend to locate away from
those centers with a large share of industrial land uses.
7) The share of highly educated worker residents within centers has a positive and significant
influence on the location choices of firms in only a few office-related sectors, including motion pictures,
advertising, and accounting. Firms in production-related activities are driven away from those centers of
high local labor-force quality, possibly due to the associated high labor and land costs.
In sum, this study provides some evidence that congestion delays would diminish the locational
advantages associated with agglomeration economies both at the regional scale and the local scale, and
would inhibit further growth of employment centers by reducing their attractiveness to new firms, when
controlling for agglomeration economies, land supply, and other locational attributes. It also proves that
firms in different industrial sectors vary in their responses to congestion costs at different geographic
scales. Given that the LQs in center and non-center locations do not significantly differ between new
firms and existing firms, my conclusions about the location decisions of new firms might also be
extended to firm locations in general. These results could be useful for urban policy makers to target
specific activities at existing agglomerations to enhance their advantages. For example, within
metropolitan areas, the high level of local congestion may drive away firms in manufacturing, wholesale
trade, and retail trade sectors, while long commuting delays that many employment centers have for
office workers may account for the migration of firms in high-order service activities. This study thus
supports public efforts to mitigate congestion delays at both the regional scale and local scale to influence
the future development pattern of different activities.
There are several more issues that need to be further examined. For example, some studies
suggest that the randomly drawn choice-set approach for discrete choice model estimation would be
statistically less efficient when spatial correlation across alternatives exists (e.g. Chen, Duann, and Hu

114
2005; Nerella and Bhat 2004). Although the definition of employment centers as the basic choice set
might diminish the spatial correlation problem to some extent, this assumption needs to be tested, and
additional work on this issue would be addressed in the next step of this research. Moreover, this study
focuses on the Los Angeles area, which is polycentric in nature, and allows for the existence of a variety
of employment centers with different characteristics that match firms specialized in different activities.
Although the major findings are interesting, a similar analysis for other metropolitan areas would identify
whether the location behavior of firms observed in the Los Angeles area is generalizable to other areas
with different urban spatial structures.

115
Chapter 3 References
Agarwal, Ajay, Genevieve Giuliano, and Christian L Redfearn. 2012. "Strangers in Our Midst: The Usefulness of
Exploring Polycentricity." The Annals of Regional Science 48 (2):433–450.
Alonso, William. 1967. "A Reformulation of Classical Location Theory and Its Relation to Rent Theory." Papers of
the Regional Science Association 19 (1):22–44.
Anas, Alex, Richard Arnott, and Kenneth A. Small. 1998. "Urban Spatial Structure." Journal of Economic
Literature 36 (3):1426–1464.
Anas, Alex and Akin C. Buyukeren. 2013. "A Simple Model of Road Congestion with Job Sprawl." Paper presented
at the 60th annual North American meeting of the North American Regional Science Council, Atlanta,
November 13–16.
Anas, Alex and Ikki Kim. 1996. "General Equilibrium Models of Polycentric Urban Land Use with Endogenous
Congestion and Job Agglomeration." Journal of Urban Economics 40 (2):232–256.
Anas, Alex and Rong Xu. 1999. "Congestion, Land Use, and Job Dispersion: A General Equilibrium Model."
Journal of Urban Economics 45 (3):451–473.
Arauzo-Carod, Josep-Maria, Daniel Liviano-Solis, and Miguel Manjón-Antolín. 2010. "Empirical Studies in
Industrial Location: An Assessment of Their Methods And Results." Journal of Regional Science 50
(3):685–711.
Arauzo-Carod, Josep-Maria, and Elisabet Viladecans-Marsal. 2009. "Industrial Location at the Intra-metropolitan
Level: The Role of Agglomeration Economies." Regional Studies 43 (4):545–558.
Arauzo-Carod, JosepMaria, and MiguelC Manjón Antolín. 2004. "Firm Size and Geographical Aggregation: An
Empirical Appraisal in Industrial Location." Small Business Economics 22 (3–4):299–312.
Ben-Akiva, Moshe E, and Steven R Lerman. 1985. Discrete Choice Analysis: Theory and Application to Travel
Demand, vol. 9. Cambridge: MIT press.
Button, Kenneth J., Scott Leitham, Ronald W. McQuaid, and John D. Nelson. 1995. "Transport and Industrial and
Commercial Location." The Annals of Regional Science 29 (2):189–206.
Charney, Alberta H. 1983. "Intraurban Manufacturing Location Decisions and Local Tax Differentials." Journal of
Urban Economics 14 (2):184–205.
Chen, Y., L. Duann, and W. Hu. 2005. "The Estimation of Discrete Choice Models with Large Choice-Set." Journal
of the Eastern Asia Society for Transportation Studies 6:1724–1739.
Cie ślik, Andrzej. 2005. "Location of Foreign Firms and National Border Effects: The Case of Poland." Tijdschrift
voor economische en sociale geografie 96 (3):287–297.
Cie ślik, Andrzej. 2005. "Regional Characteristics and the Location of Foreign Firms within Poland." Applied
Economics 37 (8):863–874.
Coffey, William J. and Antoine S. Bailly. 1991. "Producer Services and Flexible Production: An Exploratory
Analysis." Growth and Change 22 (4):95–117.
Coffey, William J., Réjean Drolet, and Mario Polèse. 1996. "The Intrametropolitan Location of High Order Services:
Patterns, Factors and Mobility in Montreal." Papers in Regional Science 75 (3):293–323.
Downs, Anthony. 1992. Stuck in Traffic: Coping with Peak-hour Traffic Congestion. Washington, D.C.: Brookings
Institution Press.

116
e Silva, Filipe Batista, Eric Koomen, Vasco Diogo, and Carlo Lavalle. 2014. "Estimating Demand for Industrial and
Commercial Land Use Given Economic Forecasts." PloS one 9 (3):e91991.
Economic Development Research Group. 2005. "The Cost of Congetion to the Economy of the Portland Region
[online]." Boston: Economic Development Research Group, Inc., [cited 16 September 2013]. Available
from: http://www.portofportland.com/PDFPOP/Trade_Trans_Studies_CoCReport1128Final.pdf
Egeln, Jürgen, Sandra Gottschalk, and Christian Rammer. 2004. "Location Decisions of Spin-Offs from Public
Research Institutions." Industry and Innovation 11 (3):207–223.
Eisenberg, Joseph H. and Frank G. Mittelbach. 1991. "The Intrametropolitan Location of Corporate Offices:
Implications for Economic Development Analysis." Economic Development Quarterly 5 (3):229–240.
Erickson, Rodney A. and Michael Wasylenko. 1980. "Firm Relocation and Site Selection in Suburban
Municipalities." Journal of Urban Economics 8 (1):69–85.
Feather, Peter M. 1994. "Sampling and Aggregation Issues in Random Utility Model Estimation." American Journal
of Agricultural Economics 76 (4):772–780.
Fujita, Masahisa and Tomoya Mori. 2005. "Transport Development and the Evolution Of Economic Geography."
Portuguese Economic Journal 4 (2):129–156.
Giuliano, Genevieve and Christian L. Redfearn. 2009. "Employment Center Hierarchy & Function: The
Manifestation of Agglomeration in the Los Angeles Metropolitan Area." Los Angeles: University of
Southern California. Unpublished work.
Giuliano, Genevieve, Christian Redfearn, Ajay Agarwal, and Sylvia He. 2011. "Network Accessibility and
Employment Centers." Urban Studies 49 (1):77–95.
Giuliano, Genevieve and Kenneth A. Small. 1991. "Subcenters in the Los Angeles Region." Regional Science and
Urban Economics 21 (2):163–182.
Giuliano, Genevieve and Kenneth A. Small. 1999. "The Determinants of Growth of Employment Subcenters."
Journal of Transport Geography 7 (3):189–201.
Goodwin, Phil. "The Economic Costs of Road Traffic Congestion [online]." London: University College London,
Rail Freight Group, [cited 16 September 2013]. Available from: http://eprints.ucl.ac.uk/1259.
Graham, Daniel J. and Patricia C Melo. 2011. "Assessment of Wider Economic Impacts of High-Speed Rail for
Great Britain." Transportation Research Record: Journal of the Transportation Research Board 2261
(1):15–24.
Guimaraes, Paulo, Octávio Figueiredo, and Douglas Woodward. 2004. "Industrial Location Modeling: Extending
the Random Utility Framework." Journal of Regional Science 44 (1):1–20.
Handy, S. L. and D. A. Niemeier. 1997. "Measuring Accessibility: An Exploration of Issues and Alternatives."
Environment and Planning A 29 (7):1175–1194.
Heuvel, Frank P. van den, Peter W. de Langen, Karel H. van Donselaar, and Jan C. Fransoo. 2013. "Proximity
Matters: Synergies through Co-Location of Logistics Establishments." International Journal of Logistics
Research and Applications 17 (5): 377–395.
Holl, Adelheid. 2004. "Transport Infrastructure, Agglomeration Economies, and Firm Birth: Empirical Evidence
from Portugal." Journal of Regional Science 44 (4):693–712.
Hymel, Kent. 2009. "Does Traffic Congestion Reduce Employment Growth?" Journal of Urban Economics 65
(2):127–135.

117
Ihlanfeldt, Keith R. and Michael D Raper. 1990. "The Intrametropolitan Location of New Office Firms." Land
Economics 66 (2):182–198.
Khalili, Amir, Vijay K Mathur, and Diran Bodenhorn. 1974. "Location and the Theory of Production: A
Generalization." Journal of Economic Theory 9 (4):467–475.
Kolko, Jed and David Neumark. 2008. "Changes in the Location of Employment and Ownership: Evidence from
California." Journal of Regional Science 48 (4):717–744.
Lomax, Tim, David Schrank, and Bill Eisele. 2012. 2012 Annual Urban Mobility Report. College Station: Texas
A&M Transportation Institute.
Lomax, Tim, Shawn Turner, Gordon Shunk, Herbert S. Levinson, Richard H. Pratt, Paul N. Bay, and G. Bruce
Douglas. 1997. Quantifying Congestion. Volume 1: Final Report (National Cooperative Highway Research
Program Report 398). Washington, D.C.: National Academy Press.
Louveaux, François, Jacques-François Thisse, and Hubert Beguin. 1982. "Location Theory and Transportation
Costs." Regional Science and Urban Economics 12 (4):529–545.
Maoh, Hanna and Pavlos Kanaroglou. 2009. "Intrametropolitan Location of Business Establishments."
Transportation Research Record: Journal of the Transportation Research Board 2133 (1):33–45.
McCann, Philip and Stephen Sheppard. 2003. "The Rise, Fall and Rise Again of Industrial Location Theory."
Regional Studies 37 (6–7):649–663.
McFadden, Daniel. 1974. "Conditional Logit Analysis of Qualitative Choice Behavior." In Frontiers in
Econometrics, edited by Paul Zarembka, 105–142. New York: Academic Press.
McFadden, Daniel. 1978. "Modeling the Choice of Residential Location." In Spatial Interaction Theory and
Planning Models, edited by Anders Karlqvist, Kungl. Tekniska högskolan. Matematiska institutionen, 75–
96. Amsterdam: Elsevier.
McFadden, Daniel L. 1984. "Econometric analysis of qualitative response models." In Handbook of Econometrics,
edited by G. Zvi and D. I. Michael, 1395–1457. Amsterdam: Elsevier.
Melo, Patricia C., Daniel J. Graham, and Robert B. Noland. 2009. "A Meta-analysis of Estimates of Urban
Agglomeration Economies." Regional Science and Urban Economics 39 (3):332–342.
Melo, Patricia C., Daniel J. Graham, and Robert B. Noland. 2010. "Impact of Transport Infrastructure on Firm
Formation." Transportation Research Record: Journal of the Transportation Research Board 2163
(1):133–143.
Mills, Edwin S. and David M. De Ferranti. 1971. "Market Choices and Optimum City Size." The American
Economic Review 61 (2):340–345.
Mondschein, A., Brian D. Taylor, and Stephen Brumbaugh. 2011. "Congestion and Accessibility: What's the
Relationship? [online]." Los Angeles: University of California, University of California Transportation
Center (UCTC), [cited 16 September 2013]. Available from:
https://escholarship.org/uc/item/6bh2n9wx.pdf
Moomaw, Ronald L. 1985. "Firm Location and City Size: Reduced Productivity Advantages as a Factor in the
Decline of Manufacturing in Urban Areas." Journal of Urban Economics 17 (1):73–89.
Moses, Leon N. 1958. "Location and the Theory of Production." The Quarterly Journal of Economics 72 (2):259–
272.
Mun, Se-il and Kazuhiro Yoshikawa. 1993. "Communication Among Firms, Traffic Congestion and Office
Agglomeration." The Annals of Regional Science 27 (1):61–77.

118
Nerella, Sriharsha and Chandra R Bhat. 2004. "Numerical Analysis of Effect of Sampling of Alternatives in Discrete
Choice Models." Transportation Research Record: Journal of the Transportation Research Board 1894
(1):11–19.
Parsons, George R. and Mary Jo Kealy. 1992. "Randomly Drawn Opportunity Sets in a Random Utility Model of
Lake Recreation." Land Economics 68 (1):93–106.
Redfearn, Christian L. 2009. "Persistence in Urban Form: The Long-Run Durability of Employment Centers in
Metropolitan Areas." Regional Science and Urban Economics 39 (2):224–232.
Redfearn, Christian L. 2014. "Reconsidering the Impact of Access to Transit on Local Land Markets." Paper
presented at the 53rd annual meeting of the Western Regional Science Association, San Diego, Feburary
16–19.
Richardson, Harry W. 1995. "Economies and Diseconomies of Agglomeration." In Urban Agglomeration and
Economic Growth, edited by Herbert Giersch, 123–155. Berlin-Heidelberg: Springer.
Shearmur, Richard and William Coffey. 2002. "Urban Employment Subcenters and Sectoral Clustering in Montreal:
Complementary Approaches to the Study of Urban Form." Urban Geography 23 (2):103–130.
Shukla, Vibhooti and Paul Waddell. 1991. "Firm Location and Land Use in Discrete Urban Space: A Study of the
Spatial Structure of Dallas–Fort Worth." Regional Science and Urban Economics 21 (2):225–253.
Solow, Robert M. 1972. "Congestion, Density and the Use of Land in Transportation." The Swedish Journal of
Economics 74 (1):161–173.
Stahl, Konrad. 1987. "Therories of Urban Business Location." In Handbook of Regional and Urban Economics, vol.
2, edited by Edwin S Mills, 759–820. Amsterdam: Elsevier.
Sweet, Matthias. 2011. "Does Traffic Congestion Slow the Economy?" Journal of Planning Literature 26 (4):391-
404.
Tauchen, Helen and Ann D. Witte. 1983. "An Equilibrium Model of Office Location and Contact Patterns."
Environment and Planning A 15 (10): 1311–1326.
Waddell, Paul and Gudmundur F Ulfarsson. 2003. "Accessibility and Agglomeration: Discrete-Choice Models of
Employment Location by Industry Sector." Paper presented at the 82nd annual meeting of the
Transportation Research Board, Washington, D.C., January 12–16.
Waddell, Paul and Vibhooti Shukla. 1993. "Manufacturing Location in a Polycentric Urban Area: A Study in the
Composition and Attractiveness of Employment Subcenters." Urban Geography 14 (3):277–296.
Walls and Associates. 2008. "National Establishment Time-Series (NETS) Database: Database Description[online]."
Oakland: Wall & Associates, Inc., [cited 16 September 2013]. Available from:
https://msbfile03.usc.edu/digitalmeasures/cswenson/intellcont/NETS%20Database%20Description2008-
1.pdf.
Weber, Alfred. 1929. Theory of the location of industries. Chicago: University of Chicago Press.
Weisbrod, Glen, Don Vary, and George Treyz. 2001. "Economic Implications of Congestion." In National
Cooperative Highway Research Program (NCHRP) Report 463. Washington: Transportation Research
Board, American Association of State Highway and Transportation Officials.
Weisbrod, Glen, Don Vary, and George Treyz. 2003. "Measuring Economic Costs of Urban Traffic Congestion to
Business." Transportation Research Record: Journal of the Transportation Research Board 1839:98–106.
Wheaton, William C. 2004. "Commuting, Congestion, and Employment Dispersal in Cities with Mixed Land Use."
Journal of Urban Economics 55 (3):417–438.

119
White, Michelle J. 1976. "Firm Suburbanization and Urban Subcenters." Journal of Urban Economics 3 (4):323–
343.

120
Appendix for Chapter 3
Table 3-a1 Spatial Concentration pattern of different industries (2001-2005)
NAICS
Code
Industry Title
Center Non-center location
Est Emp LQ Est Emp LQ
Utilities (NAICS2=22)
221 Utilities 80 1,699 0.81 218 2,306 1.09
Construction (NAICS2==23)
236 Construction of Buildings 958 7,564 0.68 3,303 18,464 1.16
237
Heavy and Civil Engineering
Construction
1,562 5,563 1.09 2,759 9,707 0.95
238 Specialty Trade Contractors 2,752 13,341 0.67 9,628 37,152 1.16
Manufacturing (NAICS2=31-33)
311 Food Manufacturing 497 4,382 0.91 1,160 6,144 1.05
312
Beverage and Tobacco Product
Manufacturing
50 395 0.95 109 1,163 1.02
313 Textile Mills 158 2,373 1.82 104 1,218 0.59
314 Textile Product Mills 103 1,098 1.23 149 887 0.88
315 Apparel Manufacturing 860 9,668 1.73 638 3,131 0.64
316
Leather and Allied Product
Manufacturing
70 388 1.36 86 1,886 0.82
321 Wood Product Manufacturing 110 671 0.75 335 2,373 1.13
322 Paper Manufacturing 55 1,667 0.98 115 1,838 1.01
323
Printing and Related Support
Activities (Printing)
1,132 6,094 1.14 1,863 6,559 0.93
324
Petroleum and Coal Products
Manufacturing
26 272 1.06 48 2,410 0.97
325 Chemical Manufacturing 289 4,896 1.26 401 3,697 0.87
326
Plastics and Rubber Products
Manufacturing
131 3,960 1.08 234 4,779 0.96
327
Nonmetallic Mineral Product
Manufacturing
95 1,379 0.76 281 3,353 1.12
331 Primary Metal Manufacturing 67 1,190 1.19 103 1,134 0.91
332
Fabricated Metal Product
Manufacturing
649 6,058 1.06 1,200 7,217 0.97
333 Machinery Manufacturing 380 4,338 1.06 704 4,982 0.97
334
Computer and Electronic
Product Manufacturing
671 16,337 1.28 915 23,544 0.86
335
Electrical Equipment,
Appliance, and Component
Manufacturing
128 1,461 0.95 277 3,710 1.02
336
Transportation Equipment
Manufacturing
175 6,114 0.93 391 7,945 1.03

121
337
Furniture and Related Product
Manufacturing
263 2,440 0.95 573 3,320 1.02
339 Miscellaneous Manufacturing 788 6,174 0.99 1,609 9,224 1.00
Wholesale Trade (NAICS2=42)
423
Merchant Wholesalers,
Durable Goods
5,963 32,464 1.08 10,751 47,686 0.96
424
Merchant Wholesalers,
Nondurable Goods
5,735 25,323 1.33 7,306 28,225 0.84
Retail Trade (NAICS2=44-45)
441
Motor Vehicle and Parts
Dealers
1,450 6,622 0.68 4,957 23,512 1.16
442
Furniture and Home
Furnishings Stores
1,459 5,177 0.89 3,470 11,209 1.05
443
Electronics and Appliance
Stores
1,562 6,455 0.99 3,225 11,940 1.01
444
Building Material and Garden
Equipment and Supplies
Dealers
769 4,472 0.87 1,892 10,762 1.06
445 Food and Beverage Stores 2,154 9,140 0.76 6,390 31,267 1.12
446
Health and Personal Care
Stores
1,438 5,906 0.93 3,224 15,419 1.03
447 Gasoline Stations 319 1,339 0.55 1,438 6,569 1.22
448
Clothing and Clothing
Accessories Stores (Clothing
stores)
6,085 20,340 1.24 8,701 28,689 0.88
451
Sporting Goods, Hobby, Book,
and Music Stores
1,723 6,960 0.97 3,647 12,885 1.02
452 General Merchandise Stores 261 1,941 0.64 965 14,222 1.18
453 Miscellaneous Store Retailers 2,896 9,180 0.82 7,736 23,550 1.09
454 Nonstore Retailers 242 2,498 0.85 618 4,206 1.07
Transportation and Warehousing (NAICS2=48-49)
481 Air Transportation 43 357 1.24 62 769 0.88
482 Rail Transportation 13 235 1.03 25 118 0.98
483 Water Transportation 17 249 1.09 30 372 0.95
484 Truck Transportation 610 3,331 0.70 2,030 10,021 1.15
485
Transit and Ground Passenger
Transportation
320 3,100 0.91 739 5,071 1.04
486 Pipeline Transportation 5 342 0.94 11 292 1.03
487
Scenic and Sightseeing
Transportation
7 143 0.81 19 59 1.09
488
Support Activities for
Transportation
1,514 8,296 0.95 3,293 13,825 1.02
491 Postal Service 17 961 0.52 81 6,512 1.24
492 Couriers and Messengers 209 7,992 0.98 437 7,229 1.01

122
493 Warehousing and Storage 301 4,450 1.22 444 5,653 0.89
Information (NAICS2=51)
511
Publishing Industries (except
Internet) (Publishing)
1,059 7,900 1.31 1,385 6,156 0.85
512
Motion Picture and Sound
Recording Industries (Motion
pict.)
4,711 27,588 1.63 3,998 12,419 0.69
515 Broadcasting (except Internet) 190 6,298 1.32 245 3,736 0.84
517
Telecommunications
(Telecommnic.)
1,643 12,059 1.12 2,783 12,171 0.94
518
Internet Service Providers,
Web Search Portals, and Data
Processing Services
540 5,898 1.23 787 6,806 0.89
519 Other Information Services 126 426 1.15 204 1,198 0.92
Finance and Insurance (NAICS2=52)
522
Credit Intermediation and
Related Activities
3,743 26,082 1.06 6,960 40,780 0.97
523
Securities, Commodity
Contracts, and Other Financial
Investments and Related
Activities (Securities)
6,989 19,566 1.27 9,659 23,963 0.87
524
Insurance Carriers and Related
Activities
2,393 10,996 0.98 4,980 18,387 1.01
525
Funds, Trusts, and Other
Financial Vehicles
294 2,447 1.10 511 1,991 0.95
Real Estate and Rental and Leasing (NAICS2=53)
531 Real Estate 8,256 26,592 1.09 14,719 47,789 0.96
532 Rental and Leasing Services 1,242 4,956 0.88 3,029 9,692 1.06
533
Lessors of Nonfinancial
Intangible Assets (except
Copyrighted Works)
46 869 1.62 40 221 0.70
Professional, Scientific, and Technical Services (NAICS2=54)
5411 Legal Services 3,476 14,828 1.70 2,705 13,002 0.65
5412
Accounting, Tax Preparation,
Bookkeeping, and Payroll
Services (Accounting)
1,551 8,019 1.18 2,416 9,390 0.91
5413
Architectural, Engineering, and
Related Services
1,228 12,311 1.09 2,168 11,584 0.95
5414 Specialized Design Services 698 3,003 1.14 1,148 5,581 0.93
5415
Computer Systems Design and
Related Services (Computer
Syst. Design)
1,323 11,002 1.15 2,146 12,013 0.92
5416
Management, Scientific, and
Technical Consulting Services
(Managem.)
2,217 18,702 1.11 3,793 22,819 0.94
5417
Scientific Research and
Development Services
472 3,221 1.18 732 6,035 0.91

123
5418
Advertising and Related
Services (Advertising)
1,528 8,159 1.38 1,813 7,139 0.81
5419
Other Professional, Scientific,
and Technical Services
492 4,564 1.09 869 5,776 0.95
Management of Companies and Enterprises (NAICS2=55)
551
Management of Companies
and Enterprises
1,268 3,126 1.55 1,201 4,173 0.73
Administrative and Support and Waste Management and Remediation Services (NAICS2=56)
561
Administrative and Support
Services
9,634 48,463 0.83 25,429 104,083 1.08
562
Waste Management and
Remediation Services
202 1,426 0.77 586 3,043 1.11
Educational Services (NAICS2=61)
611 Educational Services 948 11,808 0.82 2,528 42,666 1.09
Health Care and Social Assistance (NAICS2=62)
621
Ambulatory Health Care
Services
10,128 47,666 1.01 20,160 86,664 1.00
622 Hospitals 177 4,934 1.23 259 5,184 0.89
623
Nursing and Residential Care
Facilities
163 3,797 0.51 802 13,683 1.24
624 Social Assistance 1,740 9,869 0.74 5,373 27,533 1.13
Arts, Entertainment, and Recreation (NAICS2=71)
711
Performing Arts, Spectator
Sports, and Related Industries
1,612 9,332 1.28 2,183 11,060 0.86
712
Museums, Historical Sites, and
Similar Institutions
120 440 0.94 264 723 1.03
713
Amusement, Gambling, and
Recreation Industries
1,081 5,539 0.89 2,588 14,701 1.05
Accommodation and Food Services (NAICS2=72)
721 Accommodation 360 9,866 0.94 791 12,672 1.03
722
Food Services and Drinking
Places
2,844 33,248 0.82 7,612 84,055 1.09
Other Services (except Public Administration) (NAICS2=72)
811 Repair and Maintenance 2,754 12,211 0.67 9,625 28,695 1.16
812 Personal and Laundry Services 3,331 11,903 0.74 10,280 30,215 1.13
813
Religious, Grantmaking, Civic,
Professional, and Similar
Organizations
3,617 12,604 0.80 10,053 39,019 1.10
Public Administration (NAICS2=92)
921
Executive, Legislative, and
Other General Government
Support
120 7,804 1.14 198 9,483 0.93

124
922
Justice, Public Order, and
Safety Activities
117 11,081 0.71 382 20,175 1.14
923
Administration of Human
Resource Programs
79 3,320 1.06 147 9,306 0.97
924
Administration of
Environmental Quality
Programs
22 1,410 0.57 94 4,630 1.21
925
Administration of Housing
Programs, Urban Planning, and
Community Development
32 2,743 0.85 82 1,827 1.08
926
Administration of Economic
Programs
50 5,360 0.96 108 2,341 1.02
927
Space Research and
Technology
8 89 1.34 10 35 0.83
928
National Security and
International Affairs
49 512 0.85 126 4,015 1.08

125
Table 3-a2 Spatial Concentration pattern of selected NAICS3 industries across centers (2000,
10/10 centers)
NAICS
Code
Industry Title
Center Non-center location
Est Emp LQ Est Emp LQ
Utilities (NAICS2=22)
221 Utilities 147 16,858 0.66 505 20,414 1.18
Construction (NAICS2==23)
236 Construction of Buildings 2,077 32,058 0.72 6,411 58,260 1.15
237
Heavy and Civil Engineering
Construction
1,263 19,606 1.06 2,224 29,286 0.97
238 Specialty Trade Contractors 4,213 58,068 0.66 14,533 143,097 1.18
Manufacturing (NAICS2=31-33)
311 Food Manufacturing 1,033 40,898 0.94 2,202 43,567 1.03
312
Beverage and Tobacco Product
Manufacturing
73 4,913 0.95 152 5,289 1.02
313 Textile Mills 474 13,554 1.81 295 5,810 0.58
314 Textile Product Mills 325 8,708 1.21 465 5,498 0.89
315 Apparel Manufacturing 2,150 50,835 1.76 1,436 22,037 0.61
316
Leather and Allied Product
Manufacturing
129 2,633 1.33 155 3,410 0.83
321 Wood Product Manufacturing 260 4,482 0.89 600 13,821 1.06
322 Paper Manufacturing 268 11,636 1.34 320 11,401 0.83
323
Printing and Related Support
Activities (Printing)
2,233 35,281 1.25 3,024 28,266 0.87
324
Petroleum and Coal Products
Manufacturing
63 3,297 0.99 123 6,995 1.00
325 Chemical Manufacturing 811 37,042 1.30 1,020 31,780 0.85
326
Plastics and Rubber Products
Manufacturing
598 27,074 1.09 1,009 32,956 0.95
327
Nonmetallic Mineral Product
Manufacturing
364 8,486 1.00 707 21,129 1.00
331 Primary Metal Manufacturing 257 17,316 1.24 352 12,397 0.88
332
Fabricated Metal Product
Manufacturing
2,529 58,805 1.12 4,096 67,223 0.94
333 Machinery Manufacturing 1,161 29,707 1.10 1,948 36,724 0.95
334
Computer and Electronic
Product Manufacturing
1,751 116,428 1.30 2,185 72,255 0.84
335
Electrical Equipment, Appliance,
and Component Manufacturing
529 28,497 1.29 674 17,678 0.85
336
Transportation Equipment
Manufacturing
716 67,563 1.14 1,134 44,504 0.93
337
Furniture and Related Product
Manufacturing
798 29,277 1.10 1,334 29,575 0.95

126
339 Miscellaneous Manufacturing 2,053 50,392 1.20 2,956 45,795 0.90
Wholesale Trade (NAICS2=42)
423
Merchant Wholesalers, Durable
Goods
12,494 181,231 1.22 17,537 162,596 0.89
424
Merchant Wholesalers,
Nondurable Goods (Merchant
Wholesalers)
8,763 115,593 1.46 8,884 92,954 0.76
Retail Trade (NAICS2=44-45)
441 Motor Vehicle and Parts Dealers 1,654 27,031 0.62 6,168 76,475 1.20
442
Furniture and Home Furnishings
Stores
1,658 12,875 0.87 3,912 22,610 1.07
443 Electronics and Appliance Stores 1,987 26,424 0.98 3,978 26,244 1.01
444
Building Material and Garden
Equipment and Supplies Dealers
764 11,067 0.66 2,620 31,554 1.17
445 Food and Beverage Stores 2,236 27,286 0.58 9,153 118,171 1.22
446 Health and Personal Care Stores 1,718 16,331 0.85 4,202 36,461 1.08
447 Gasoline Stations 550 4,779 0.53 2,492 17,617 1.24
448
Clothing and Clothing
Accessories Stores (Clothing
stores)
5,932 44,730 1.17 8,934 52,696 0.91
451
Sporting Goods, Hobby, Book,
and Music Stores
2,049 23,977 0.79 5,555 38,480 1.11
452 General Merchandise Stores 470 31,896 0.69 1,520 69,661 1.16
453 Miscellaneous Store Retailers 4,085 29,969 0.80 10,857 55,767 1.10
454 Nonstore Retailers 608 10,589 0.79 1,643 14,441 1.11
Transportation and Warehousing (NAICS2=48-49)
481 Air Transportation 158 3,619 1.56 140 5,545 0.71
482 Rail Transportation 7 1,728 0.64 25 1,999 1.19
483 Water Transportation 51 2,090 1.54 46 1,716 0.72
484 Truck Transportation 1,000 18,990 0.77 2,826 38,440 1.12
485
Transit and Ground Passenger
Transportation
322 9,587 0.94 682 18,716 1.03
486 Pipeline Transportation 10 570 1.09 17 826 0.96
487
Scenic and Sightseeing
Transportation
18 277 0.89 41 996 1.05
488
Support Activities for
Transportation
1,422 19,562 1.11 2,338 37,298 0.94
491 Postal Service 63 15,754 0.54 281 14,879 1.24
492 Couriers and Messengers 461 15,781 1.30 583 9,298 0.85
493 Warehousing and Storage 589 22,038 1.24 806 21,522 0.88

127
Information (NAICS2=51)
511
Publishing Industries (except
Internet) (Publishing)
1,600 46,260 1.28 2,073 25,967 0.86
512
Motion Picture and Sound
Recording Industries (Motion
pict.)
3,614 94,432 1.65 2,815 27,334 0.66
515 Broadcasting (except Internet) 365 26,841 1.39 406 15,189 0.80
517
Telecommunications
(Telecommunic.)
785 26,944 1.27 1,034 15,529 0.86
518
Internet Service Providers, Web
Search Portals, and Data
Processing Services
1,327 34,622 1.33 1,603 21,399 0.83
519 Other Information Services 155 3,303 0.76 446 8,223 1.13
Finance and Insurance (NAICS2=52)
522
Credit Intermediation and
Related Activities
4,298 93,210 1.07 7,526 82,560 0.97
523
Securities, Commodity
Contracts, and Other Financial
Investments and Related
Activities (Securities)
2,328 36,613 1.43 2,443 16,618 0.78
524
Insurance Carriers and Related
Activities
3,857 93,069 1.04 6,978 58,244 0.98
525
Funds, Trusts, and Other
Financial Vehicles
430 8,111 1.33 521 3,970 0.83
Real Estate and Rental and Leasing (NAICS2=53)
531 Real Estate 7,188 70,845 0.92 15,737 107,859 1.04
532 Rental and Leasing Services 1,596 19,245 0.86 3,837 28,513 1.07
533
Lessors of Nonfinancial
Intangible Assets (except
Copyrighted Works)
106 2,922 1.57 92 1,602 0.70
Professional, Scientific, and Technical Services (NAICS2=54)
5411 Legal Services 9,427 91,426 1.77 6,198 31,131 0.60
5412
Accounting, Tax Preparation,
Bookkeeping, and Payroll
Services (Accounting)
2,979 38,289 1.16 4,575 24,419 0.92
5,413
Architectural, Engineering, and
Related Services
2,646 49,152 1.20 3,838 46,882 0.90
5,414 Specialized Design Services 1,574 13,791 1.20 2,264 12,362 0.90
5415
Computer Systems Design and
Related Services
2,622 57,976 1.18 3,907 42,843 0.91
5416
Management, Scientific, and
Technical Consulting Services
(Managem.)
5,148 65,863 1.18 7,683 56,665 0.91
5,417
Scientific Research and
Development Services
651 20,234 1.34 772 14,221 0.82
5418
Advertising and Related
Services (Advertising)
2,418 33,492 1.41 2,606 22,935 0.79

128
5,419
Other Professional, Scientific,
and Technical Services
998 10,293 1.09 1,688 11,698 0.95
Management of Companies and Enterprises (NAICS2=55)
551
Management of Companies and
Enterprises
267 5,575 1.72 188 3,204 0.63
Administrative and Support and Waste Management and Remediation Services (NAICS2=56)
561
Administrative and Support
Services
9,040 189,366 1.04 16,467 195,243 0.98
562
Waste Management and
Remediation Services
361 10,222 0.85 879 16,035 1.08
Educational Services (NAICS2=61)
611 Educational Services 2,050 130,726 0.62 7,686 386,399 1.20
Health Care and Social Assistance (NAICS2=62)
621
Ambulatory Health Care
Services
12,248 115,398 1.01 23,457 179,473 1.00
622 Hospitals 293 91,430 0.85 724 146,116 1.08
623
Nursing and Residential Care
Facilities
335 15,393 0.44 1,897 65,627 1.29
624 Social Assistance 2,365 29,145 0.73 7,135 78,216 1.14
Arts, Entertainment, and Recreation (NAICS2=71)
711
Performing Arts, Spectator
Sports, and Related Industries
2,973 40,873 1.24 4,041 36,415 0.87
712
Museums, Historical Sites, and
Similar Institutions
144 2,724 1.00 279 4,084 1.00
713
Amusement, Gambling, and
Recreation Industries
764 21,011 0.68 2,544 53,696 1.17
Accommodation and Food Services (NAICS2=72)
721 Accommodation 749 42,748 0.70 2,397 66,110 1.16
722
Food Services and Drinking
Places
6,736 138,687 0.78 18,713 290,378 1.12
Other Services (except Public Administration) (NAICS2=72)
811 Repair and Maintenance 5,336 41,915 0.72 16,414 92,154 1.15
812 Personal and Laundry Services 5,221 38,952 0.71 16,408 87,366 1.15
813
Religious, Grantmaking, Civic,
Professional, and Similar
Organizations
3,293 53,583 0.65 11,609 115,105 1.18
Public Administration (NAICS2=92)
921
Executive, Legislative, and
Other General Government
Support
198 42,047 1.03 368 32,513 0.99
922
Justice, Public Order, and Safety
Activities
191 30,787 0.91 424 39,583 1.05

129
923
Administration of Human
Resource Programs
117 17,210 1.33 141 10,392 0.83
924
Administration of Environmental
Quality Programs
61 4,542 0.63 222 10,927 1.19
925
Administration of Housing
Programs, Urban Planning, and
Community Development
56 4,848 1.00 108 2,745 1.00
926
Administration of Economic
Programs
143 13,402 1.25 192 8,515 0.87
927 Space Research and Technology 5 109 0.81 13 2,527 1.10
928
National Security and
International Affairs
81 2,566 1.09 137 15,967 0.95

130
Table 3-a3 T-tests on the equality of the means of LQs in center locations

N Mean
Std.
Err.
Std.
Dev.
[95% Conf.
Interval]
LQ for new
establishments (2001-
2005)
96 1.03 0.03 0.27 0.97 1.08
LQ for existing
establishments (2000)
96 1.06 0.03 0.31 0.99 1.12
combined 192 1.04 0.02 0.29 1.00 1.08
diff -0.03 0.04 -0.11 0.05
diff = mean(LQ for new establishments) - mean(LQ for existing establishments) t = -0.7065
Ho: diff = 0 degrees of freedom = 190
Ha: diff < 0 Ha: diff != 0 Ha: diff > 0
Pr(T < t) = 0.2404 Pr(|T| > |t|) = 0.4808 Pr(T > t) = 0.7596

Table 3-a4 T-tests on the equality of the means of LQs in non-center locations

N Mean
Std.
Err.
Std.
Dev.
[95% Conf.
Interval]
LQ for new
establishments (2001-
2005)
96 0.99 0.01 0.13 0.96 1.01
LQ for existing
establishments (2000)
96 0.97 0.02 0.16 0.94 1.00
combined 192 0.98 0.01 0.15 0.96 1.00
diff 0.01 0.02 -0.03 0.06
diff = mean(LQ for new establishments) - mean(LQ for existing establishments) t = -0.665
Ho: diff = 0 degrees of freedom = 190
Ha: diff < 0 Ha: diff != 0 Ha: diff > 0
Pr(T < t) = 0.7466 Pr(|T| > |t|) = 0.5069 Pr(T > t) = 0.2534

131
Table 3-a5 Correspondences between broad economic sectors and land use types
Broad Economic
Sector
Detailed sectors + (NAICS2 code) Land Use Type
Detailed land use
type
Production-related
industry
Utilities (22);
Industrial and
transportation land use
Industrial;
Transportation,
communications,
and utilities
Manufacturing (31-34);
Wholesale Trade (42);
Transportation and warehousing (48)
Commercial
activities and
private service
Construction (23);
Office and commercial
land use
General office;
Commercial and
services
Retail Trade (44);
Information (51);
Finance and Insurance (52);
Real Estate and Leasing (53);
Professional Service (54);
Management (55)
Public services
and
administration
Educational Services (61);
Facilities and
education use
Facilities;
Education
Health Care and Social Assistance
(62);
Public administration (92)

132
Table 3-a6 Summary statistics of explanatory variables

Variable Description
Expected
Sign
Mean
Std.
Dev.
Min Max
Urbanization
effects
EmpSize
Total employment of each center (,000
jobs)
(+) 74.29 173.55711.407 1150.264
EmpDen
Total employment per land area (jobs per
acre) for each center
(+/-) 19.53 9.22 10.86 59.18
Localization
effects
InduShare31
Percent of center employment in the
manufacturing sectors (NAICS2=31)
(+)
2.11 1.96 0.02 8.43
InduShare42
Percent of center employment in the
wholesale trade sectors (NAICS2=42)
7.61 5.97 0.1 22.26
InduShare51
Percent of center employment in the
information sectors (NAICS2=51)
5.62 9.17 0.07 60.6
Indushare52
Percent of center employment in the
finance and insurance sectors
(NAICS2=52)
5.38 5.24 0.46 25.32
InduShare54
Percent of center employment in the
professional services sectors (NAICS2=54)
8.59 6.03 0.78 26.31
Traffic
Congestion
Delay Rate
Average of travel time delays (in minutes)
per mile for all possible trips within 12
miles of each center's peak tract
(-) 0.47 0.12 0.16 0.68
Delay
Rate_inner
Average of travel time delays (in minutes)
per mile for all possible trips within each
center
(-) 0.30 0.16 0 0.69
Accessible
opportunities
within 12 miles
BizAcc31
Cumulative sum of employment in the
wholesale trade sector (NAICS2=31) (,000
jobs)
(+) 168.997 69.749 29.181 350.436
BizAcc42
Cumulative sum of employees in the
wholesale trade sector (NAICS2=42) (,000
jobs)
(+) 81.258 35.569 15.5 178.708

133
BizAcc51
Cumulative sum of employees in the
information sector (NAICS2=51) (,000
jobs)
(+) 52.776 47.716 6.775 179.372
BizAcc52
Cumulative sum of employees in the
finance and insurance sector (NAICS2=52)
(,000 jobs)
(+) 56.040 28.011 10.007 117.335
BizAcc54
Cumulative sum of employees in the
professional service sector (NAICS2=54)
(,000 jobs)
(+) 94.581 53.057 15.263 228.472
AccLF
Cumulative sum of potential workers (,000
workers)
(+) 1250.8 482.904287.738 2396.28
RAccLF Relative labor accessibility (+) 71.736 144.1347.652 970.774
Transport
Access
DistHwy
Distance to the nearest highway ramp
(miles)
(+/-) 0.51 0.38 0.003 1.62
DistAP
Distance to the nearest commercial airport
(miles)
(-) 8.56 5.08 0.44 19.4
Land Use
Per_indutrans
Percent of center area developed into
industrial or transportation use
(+/-) 26.16 20.14 0.13 67.35
Per_officeco
mm
Percent of center area developed into office
or commercial use
(+/-) 18.64 10.87 5.14 68.73
Per_pubedu
Percent of center area developed into
public facilities or education use
(+/-) 6.88 7.18 0 42.41
Labor quality Per_baplus
Percent of center population over 25 and
with at least bachelor's degree
(+/-) 17.93 11.33 0 51.65

134
Table 3-a7 Summary statistics of percentage of centers employment in NAICS2 sectors by 3-
quantiles

Manufacturing,
NAICS2=31
Wholesale
Trade,
NAICS2=42
Retail Trade,
NACIS2=44
Information,
NACIS2=51
Finance &
Insurance,
NACIS2=52
Professional
Serv.,
NACIS2=54
N Min Max N Min Max N Min Max N Min Max N Min Max N Min Max
Low 15 0.4 8.8 16 0.1 3.6 16 1.9 5.6 16 0.1 2.5 16 0.5 2.6 16 0.8 5.2
Median 16 11.5 24.3 16 3.6 8.5 16 5.8 10.6 15 2.5 5.2 16 2.7 6.4 16 5.2 10.7
High 16 26.3 60.0 15 9.3 22.3 15 10.7 39.7 15 5.3 60.6 15 6.5 25.3 15 11.0 26.3

Figure 3-a1 Spatial pattern of congestion delays (measured by Delay Rate, 12-mile boundary used)

135

Figure 3-a2 Spatial pattern of access to workers, 12-mile boundary

136

Figure 3-a3 Spatial pattern of access to manufacturing sectors (NAICS2=31-33), 12-mile boundary

137

Figure 3-a4 Spatial pattern of access to wholesale trade sectors (NAICS2=42), 12-mile boundary

138

Figure 3-a5 Spatial pattern of access to retail trade sectors (NAICS2=44-45), 12-mile boundary

139

Figure 3-a6 Spatial pattern of access to information sectors (NAICS2=51), 12-mile boundary

140

Figure 3-a7 Spatial pattern of access to FI services (NAICS2=52), 12-mile boundary

141

Figure 3-a8 Spatial pattern of access to professional services (NAICS2=54), 12-mile boundary

142
Table 3-a8 Summary statistics of accessibility measures at different locations (12-mile boundary)
location mean sd min p10 p25 p50 p75 p90 p95 max
AccLF center tracts (n=378) 1,627,517 563,634 287,738 920,080 1,158,586 1,588,966 2,161,056 2,391,144 2,432,043 2,472,605
non-center (n=2972) 1,006,393 721,046 704 102,443 322,424 930,207 1,556,899 2,135,739 2,278,456 2,449,573
BizAcc31 center tracts (n=378)
193,645 68,965 29,181 99,676 135,599 193,306 251,433 277,973 302,561 350,436
non-center (n=2986)
134,178103,859 3 5,375 32,843 118,267219,143283,194319,848377,527
BizAcc42 center tracts (n=378) 101,161 37,455 15,500 51,382 69,536 98,458 139,118 146,510 158,062 178,708
non-center (n=2972) 65,885 52,905 5 2,229 14,698 57,173 104,697 146,332 162,910 182,309
BizAcc44 center tracts (n=378)
135,896 39,193 27,732 84,969 105,523 132,836 177,367 186,259 188,896 194,398
non-center (n=2986)
82,689 56,626 8 9,197 27,643 80,519 128,775 161,007 180,622 198,686
BizAcc51 center tracts (n=378) 101,873 62,971 6,775 21,903 36,513 109,886 165,224 183,217 184,892 190,879
non-center (n=2971) 40,844 48,975 5 1,492 5,459 21,893 47,382 128,625 160,545 188,669
BizAcc52 center tracts (n=378) 81,241 32,480 10,007 33,911 51,166 81,680 113,456 120,184 122,976 126,638
non-center (n=2971) 42,572 36,268 2 2,157 9,372 33,562 75,141 99,910 108,139 128,134
BizAcc54 center tracts (n=378) 148,320 65,654 15,263 53,693 93,328 150,459 216,250 229,574 234,430 248,149
non-center (n=2972) 72,283 66,568 3 3,073 13,790 53,804 120,777 171,076 217,590 244,004
Note*: All tracts with zero values of accessibility are removed; measured in number of workers (for AccLF) or jobs.

143

Table 3-a9 Correlation Matrix of agglomeration variables, accessibility and congestion variables

EmpSize EmpDen
Indu
Share31
Indu
Share42
Indu
Share44
Indu
Share51
Indu
Share52
Indu
Share54
BizAcc
31
BizAcc
42
BizAcc
44
BizAcc
51
BizAcc
52
BizAcc
54
AccLF RAccLF
Delay
Rate
EmpSize 1
EmpDen 0.14 1
InduShare31 -0.04 -0.38 1
InduShare42 0.06 -0.35 0.67 1
InduShare44 -0.09 -0.33 -0.24 -0.16 1
InduShare51 0.05 0.61 -0.22 -0.18 -0.11 1
InduShare52 0.10 0.26 -0.41 -0.34 0.10 0.003 1
InduShare54 0.17 0.12 -0.33 -0.36 -0.06 0.06 0.48 1
BizAcc31 0.20 0.003 0.25 0.22 -0.15 -0.14 -0.38 -0.29 1
BizAcc42 0.26 0.13 0.14 0.16 -0.17 0.00 -0.34 -0.24 0.95 1
BizAcc44 0.30 0.25 0.10 0.10 -0.24 0.18 -0.35 -0.10 0.81 0.87 1
BizAcc51 0.31 0.35 -0.12 -0.03 -0.21 0.49 -0.12 0.11 0.22 0.40 0.70 1
BizAcc52 0.36 0.42 -0.18 -0.11 -0.25 0.37 0.01 0.16 0.39 0.56 0.74 0.82 1
BizAcc54 0.31 0.38 -0.15 -0.09 -0.23 0.36 -0.09 0.18 0.42 0.56 0.77 0.89 0.92 1
AccLF 0.32 0.15 0.11 0.09 -0.20 0.12 -0.38 -0.12 0.84 0.92 0.95 0.60 0.65 0.65 1
RAccLF 0.99 0.27 -0.03 0.09 -0.10 0.05 0.08 0.14 0.23 0.28 0.33 0.33 0.36 0.31 0.35 1
Delay Rate 0.22 0.25 -0.07 0.02 -0.11 0.27 -0.17 -0.02 0.53 0.65 0.76 0.60 0.60 0.59 0.77 0.24 1
Delay Rate_inner
0.47 -0.06 0.20 0.41 -0.05 -0.20 -0.12 -0.02 0.26 0.28 0.31 0.27 0.25 0.25 0.32 0.50 0.39

144
CHAPTER 4 TRAFFIC CONGESTION, ACCESSIBILITY TO EMPLOYMENT AND
HOUSING PRICES

A study of single-family housing market in Los Angeles County

1 Introduction
Residential property has long been considered a bundle of attributes, including structural and
location characteristics. In turn, a house sale price reflects the price of each structural attribute and the
value of locational externalities (Rosen 1974). In standard urban economic theory, transport access to jobs
is usually considered one of the most important locational determinants of housing price (Alonso 1964;
Brueckner 1987; Muth 1969). Previous empirical studies extensively explored the relative importance of
job accessibility in the determination of residential property values. For example, early studies directly
tested the "access-space tradeoff" based on the monocentric assumption, and found negative price
gradients in some areas (e.g. Anderson and Crocker 1971; Coulson and Engle 1987). With the increasing
decentralization of employment within metropolitan areas, some studies included access to non-CBD
centers and found improved predictions of house prices (e.g. Heikkila et al. 1989; Waddell, Berry, and
Hoch 1993). Using more complex measures of job accessibility, some recent studies also prove that job
opportunities dispersed across the whole urban area, not just within employment centers, may also be
valued by home buyers (e.g. Giuliano et al. 2010; Srour, Kockelman, and Dunn 2002).
The core concept in the measures of accessibility to jobs is travel time, which determines the
actual cost and relative ease of reaching desired activities (such as jobs) at the destination (Lomax et al.
1997; Mondschein, Taylor, and Brumbaugh 2011). Another factor that is also fundamentally defined by

145
travel time and closely related to accessibility is traffic congestion (Lomax et al. 1997).
18
However, this
factor has not been often examined in previous empirical studies on residential structure. By slowing
travel speeds on the road system, households face increasing travel times. Traffic congestion may then
change the accessibility advantages of housing units and may influence house prices. These changes may
occur even as the transportation network remains largely unchanged. Thus, estimating to what extent the
accessibility premium of locations is constrained by traffic congestion will provide some useful
implications for evaluating the benefits of congestion mitigation policies, such as those associated with
travel time delay reductions and system efficiency improvements. Other localized negative externalities
associated with traffic congestion such as high volumes of through traffic, noise and pollution may also
reduce the neighborhood amenities and negatively affect residential property values (Steiner, Chuang, and
Kim 2012). I do not measure these independently at this point, but will discuss these correlated effects in
the conclusion.
Most previous empirical studies of accessibility that use hedonics pricing consider a metropolitan
area as a single market, so that the estimated prices of locational attributes are assumed to be the same
across the area. This assumption would be problematic if households in different neighborhoods value
various locational attributes differently (Freeman 1979; Habib and Miller 2008; McMillen and Redfearn
2010; Straszheim 1974). Specifically, the effects of accessibility and traffic congestion could be valued
differently by households of different income levels, given their differences in the value of time and
desire for housing space. This is consistent with sorting models in an urban housing market (e.g. Bayer,
Figlio, and Lucas 2004; McMillan and Rueben 2004; Schill and Wachter 1995; Voith 1991). Thus,
accounting for the heterogeneous effects of accessibility/congestion across income clusters may yield
more meaningful estimates of the spatial variation in housing prices within a metropolitan area. It may
also help us to better understand the economic impacts of traffic congestion beyond the transportation
system from the perspective of households’ residential location responses.

18
As suggested by Lomax et al. (1997, 19–22), congestion measures reflect travel time delays “relative to an
observable benchmark,” while accessibility measures reflect “absolute travel time.”

146
This study examines the effects of accessibility to employment and traffic congestion on housing
prices, based on the hedonic price model, using the single-family housing market in Los Angeles County
as an example. Los Angeles is an excellent choice of study area because of the widespread sense that its
residents understand congestion.
19
At issue is the empirical question of whether or not that knowledge is
reflected in their willingness to pay for houses. The purpose of this study is to add to the empirical
evidence on whether and to what extent costs of travel time delays are valued by households in their
residential location decisions and capitalized into property values. Another contribution of this study is to
examine whether or not the access-space tradeoff is valued differently by households of different income
groups.
The remainder of the chapter is structured as follows. Section 2 reviews the theoretical basis and
empirical evidence on the role of accessibility in determining housing prices and the possible impacts of
traffic congestion. Section 3 describes the research strategy and methods, and section 4 describes the data
source. The estimation results are presented in section 5. The chapter concludes with a summary of
findings and discusses the remaining questions for future work.
2 Literature review
2.1 Theoretical basis
The theoretical framework for this study can be drawn on the bid-rent theory, the fundamental
theory in urban economics (Alonso 1964, 1967). Starting with the monocentric model, access to
employment was central to the explanation of urban residential structure and property values (Alonso
1964; Mills 1967; Muth 1969). The classic monocentric model assumes that all the employment and
production activities take place at a single center (CBD) and that households bid-up living spaces around
the market center to maximize their utility under their income constraints; in equilibrium, all households
maintain the same level of utility, and at each household’s location, the marginal savings in housing costs
from moving slightly further from CBD is exactly offset by the marginal increases in commuting costs

19
Source: http://miovision.com/blog/north-americas-most-congested-cities/

147
(Brueckner 1987). Thus, the price per unit of housing and rent per unit of land is expected to decrease
with distance to the market center at a decreasing rate (Brueckner 1987). Later development of the
standard model introduce multiple employment centers so that the rent or price gradient may slope up
near subcenters instead of declining monotonically from the central market (e.g. White 1976; Yinger
1992). Despite the various residential structures generated in different models, the logic behind these
theoretical models is not changed in that the spatial pattern of residential location is an aggregated result
of individual households' location behavior by trading off commuting costs and housing consumption.
2.1.1 Introduction of congestion costs
Although the commuting cost is the key factor in the standard urban economic theory that
explains spatial patterns of household locations and population density, travel time—the biggest element
of commuting costs (Hamilton 1989; Hamilton and Röell 1982; Small and Song 1992)—is usually not
specified in the theoretical modeling. Instead, commuting costs are usually measured by physical distance,
which implies that the transport rate is constant and that the relative ease of access to jobs is fixed and not
subject to the level of congestion on the transportation network.
Since the seminal work of Solow (1972), several studies have modeled the influence of traffic
congestion on urban residential structure through affecting commuting costs (Anas and Buyukeren 2013;
Anas and Liu 2007; Segal and Steinmeier 1982; Sullivan 1983; Wheaton 2004). In general, the
introduction of congestion costs allows the per unit commuting cost to vary by location and be dependent
on the traffic flows on the existing transport network, which in turn results from the pattern of residential
and employment locations and the allocation of land for transportation use within a city (Solow 1972;
Wheaton 1998, 2004).
For example, Solow (1972, 1973) includes congestion costs by extending the monocentric model,
which assumes that all commuting trips are radial inward and that commuting costs at any distance from
CBD are simply a function of the ratio of traffic flows, which amount to the number of households living

148
between the distance and the city boundary, and the amount of land devoted to roads. Assuming that the
allocation of land to roads is a decreasing function of distance to CBD, the numerical approximation
shows that the presence of congestion costs generates more convex bid-rent functions that the rent
gradient would be steeper near the CBD where commuting costs increase sharply due to heavy congestion,
while the rent gradient near the periphery would be flatter (Solow 1972, 1973). Other theoretical models
also apply the "Solow-type" function to describe the endogenous congestion costs and the mutual
dependence between land rents and commuting costs. Although some later models are more complex by
allowing the residential and employment location to be simultaneously determined (Anas and Kim 1996;
Segal and Steinmeier 1982; Wheaton 2004), it is usually demonstrated that endogenous congestion will
reinforce the importance of commuting costs in determining residential location and shift the bid-rent
curve to reflect the altered accessibility pattern.
2.1.2 Location pattern by income
The classic monocentric model has also been extended to explain how and why households of
different incomes may have different patterns of residential location. When the assumption that all
households are identical is removed, the bid-rent gradient differs across different income groups, because
their time costs of commuting and demands for housing differ. If the marginal commuting cost increases
more quickly than housing consumption as income increases, the rich will have a steeper bid-rent curve
and will favor the center location within a monocentric city, while the opposite condition implies that the
rich will favor the suburban location (Brueckner, Thisse, and Zenou 1999; Franklin and Waddell 2003;
Glaeser, Kahn, and Rappaport 2008; LeRoy and Sonstelie 1983). Some theoretical studies extend the
basic model by allowing for two modes of commuting with different fixed and variable costs and suggest
that income distribution within a city depends on the "comparative advantage" of the rich in commuting a
longer distance, which is attributable to the relative lower variable material cost of driving that might
offset the relative higher time costs for the rich (Lave 1970; LeRoy and Sonstelie 1983). However, the
basic assumption is maintained that the relative values of income elasticity of marginal commuting cost

149
and income elasticity of housing consumption determines the relative steepness of bid-rent slopes for
different income groups at particular locations.
Other studies provided an alternative explanation of the difference in bid-rent function across
income groups by introducing into households’ utility function the amenity advantage of locations, such
as historical or natural advantage, that may or may not be located in the CBD. In other words, besides the
effects of commuting cost and housing consumption, the resultant location by income is influenced by the
spatial pattern of amenities and the marginal valuation of amenities by different income groups. For
example, Brueckner (1999) predicts that in a monocentric city where the amenities effects dominate, the
bid-rent curve for the rich will be flatter, either if the amenity function is upward sloping or if it is
downward sloping but the center’s amenity advantage over the suburb is small. Ng (2008) extends the
basic model to a duocentric city with amenities in the CBD and again proves that the bid-rent curves
would be steeper for those households who value the CBD amenities more (e.g. the rich) and are more
willing to commute longer distances if their job location is the subcenter.
One more interesting question is how the bid-rent gradient by income changes with the presence
of congestion costs. This question is not explicitly addressed in Solow’s modeling (1972, 1973), as the
time values are not added to commuting costs. Lave’s (1970) modeling considers both the time and
distance cost of commuting and implies that adding a congestion factor to the model would change the
boundary between the rich and the poor but would not change the general pattern of income distribution.
The possible reason is that the effects of (endogenous) congestion on housing prices are considered to
move in the same direction as the commuting effects. However, traffic congestion may also generate
other localized negative externalities such as noise and pollution that could be more averted by the rich
than by the poor, so that its effects on the location pattern by income are more complicated.
Based on the monocentric model and several of its extensions, the main predictions from the
standard model can be summarized as follows: 1) the tradeoff between commuting costs and desire for

150
housing space determines the urban residential location pattern; 2) the introduction of congestion costs
would reinforce the effects of commuting costs and change the bid-rent gradient; 3) assuming that
commuting costs and housing consumption increase with income, different bid-rent functions would be
generated for different income groups, while the location pattern by income depends on the relative rates
at which the two effects increase with income; 4) the introduction of locational amenity effects may
weaken the access-space tradeoff in determining the bid-rent function of different income groups and alter
the income distribution.
2.2 Empirical evidence
Most house price studies are built within the context of the hedonic regression model, which
derives from Rosen’s (1974) hedonic price theory that any goods are considered as a bundle of
characteristics and that the "implicit price" of each characteristic is estimated from the "observed price of
differentiated goods and the specific amount of characteristics associated with them" (34). The
determinants of house prices usually include housing structural attributes and spatial attributes, while the
relative importance of each attribute can be examined through the estimated coefficients of hedonic
regression (Rosen 1974). Accessibility to employment is one of the most examined spatial attributes that
contribute to housing prices. Other spatial attributes include a vector of neighborhood characteristics,
such as tax rate, school quality and proximity to transport facilities, as well as disamenities such as crime,
noise and air pollution. Housing structural characteristics such as lot size, building square footage and age
are usually controlled in a hedonic model. The following discussion focuses on the empirical evidence on
the relative importance of accessibility to employment and traffic congestion in determining housing
prices.
2.2.1 On the effects of accessibility to employment
Following the monocentric model, early empirical studies examined the role of access to CBD
with respect to housing price, expecting to find a downward sloping housing price curve with respect to
CBD. Access to CBD is simply measured by straight-line distance, its log form (Anderson and Crocker

151
1971) or network travel time (Coulson and Engle 1987). However, results on the "cost-distance
relationship" are mixed. As suggested by Richardson (1977) and Coulson (1991), the existence of a
positive CBD-price gradient is attributable to the increasingly decentralized employment opportunities
outside the CBD that are also valued by households, so that the estimated single CBD gradient reflects the
confounding effects of access to decentralized employment. To account for the possible polycentric urban
structure, some empirical studies include the influence of access to other non-CBD centers, and the
estimation results suggest that these effects are substantial and that the explanatory power of the hedonic
models is increased significantly (e.g. Heikkila et al. 1989; Waddell, Berry, and Hoch 1993).
Some studies apply alternative measures of location to employment—the "generalized measures"
of accessibility, which imposes no prior assumption on the distribution of employment and allows the
effects of many job locations to be accounted for (e.g. Adair et al. 2000; Brigham 1965; Franklin and
Waddell 2003; Giuliano et al. 2010; Nelson 1977). The gravity-type accessibility measure is usually
applied, which is constructed as the sum of job opportunities surrounding each location weighted by the
travel costs between pairs of locations (complete reviews are done by Vickerman (1974); Handy and
Niemeier (1997); Vickerman, Spiekermann, and Wegener (1999); etc.). Some studies run different
hedonic models, using measures of distance/travel time to CBD and accessibility-type measures
separately. Their results usually indicate that the accessibility-type measure is a better predictor of
housing price and yields a higher goodness of fit (Brigham 1965; Burnell 1985; Nelson 1977; Ottensmann,
Payton, and Man 2008). Ottensmann, Payton, and Man (2008) also compare the effect of distance or
travel time to multiple employment centers with the effects of generalized accessibility to employment
and find that they are equally good predictors of housing prices.
The non-monotonic or non-linear effects of accessibility to employment are also explored in
some studies. For example, Waddell et al. (1993) measure access to centers at dummy intervals to allow
the magnitude of price effects and the shape of the bid-rent function to vary over the study area. Their
results show that both the CBD price gradient and the subcenter price gradient are non-linear, usually

152
positive within a short distance (1–2 miles) and then negative and diminished to zero (Waddell, Berry,
and Hoch 1993). As suggested by Osland and Pryce (2012), this may be attributable to the negative
effects of employment locations. Another reason for the non-monotonic effects of access to employment
is the existence of housing submarkets and "differential price structure of property" within an area (Adair
et al. 2000, 708). For example, Adair, et al.’s (2000) study find no significant role of gravity-based
accessibility for the Belfast urban area as a whole, but identify that accessibility is a significant variable in
some spatial submarket and tends to be more important in some low income neighborhoods. However, the
spatial variation of accessibility effect across neighborhoods may also be explained by the difference in
bid-rent function across households of different income levels (Osland and Pryce 2012).
Despite the considerable evidence on the importance of accessibility to employment in predicting
house price, most studies suggest that the effects of location relative to employment are significant but
small. For example, Ottensmann et al. (2008) suggest that housing structure and other neighborhood
socioeconomic attributes explain most variations in the house prices of Marion County, while the
inclusion of access to employment only slightly increases the goodness of fit. Giuliano et al. (2011) also
find that housing structural characteristics and distance to the coast explain most variation in the house
prices of the Los Angeles metropolitan area. However, it is also suggested that since accessibility to
employment plays a significant role in housing price determination, elimination of this effect would
generate biased estimates of some neighborhood externalities, such as the noise and pollution associated
with some type of concentrated employment opportunities, while the inclusion of simple measures of
distance to CBD is far from enough to account for this effect (Ottensmann, Payton, and Man 2008).
2.2.2 On the effects of traffic congestion
Unlike the accessibility effect, the congestion effect is not explicitly addressed in house price
studies. According to our limited knowledge, the only study that focuses explicitly on the congestion
effect is by Steiner, Chuang and Kim (2012), in which congestion costs are measured by the weighted
sum of the differences in peak and off-peak travel time to all employment centers, weighted by the

153
employment size of each center. The localized congestion effect is measured by the road congestion index,
which accounts for the number of traffic counts relative to the road capacity at the local scale (Steiner,
Chuang, and Kim 2012). Their estimation results, however, show that regional congestion contributes
positively to housing price in both the Orlando and Jacksonville Metropolitan Statistical Areas, while
regional accessibility to employment plays a significant negative role. They explained the counterintuitive
effects of traffic congestion as a result of the co-location of high regional job accessibility and high
congestion levels. However, it might also be possible that their results are compromised by the relative
high correlation between the two measures at the regional level, which would virtually render the signs
and significance of the estimated coefficients random. Similarly, the insignificant role of localized
congestion effects might also be attributable to the high correlation between access and congestion
measures at the neighborhood scale.
By contrast, some studies that are not focused on examining congestion costs suggest that
measures of accessibility associated with congested travel time are better predictors of house price, thus
providing some evidence that congestion-induced travel time delays are valued by households in their
residential location choices (Franklin and Waddell 2003; Giuliano et al. 2010; Nelson 1977; Ottensmann,
Payton, and Man 2008). For example, Nelson (1977) examines the performance of different measures of
accessibility in a hedonic price model and find that compared with a straight-line distance to CBD or
other travel-time-based measures, the peak-period travel time required to reach 75 percent of total
employment is the best predictor of house price. Similarly, Ottensmann, Payton, and Man (2008) also
systematically compare the effectiveness of alternative measures of accessibility in the prediction of
house price, including distance or travel time to the CBD, distance/travel time to employment centers, and
gravity-based employment accessibility weighted by distance or travel time, in which both the free-flow
travel time and congested travel time are used. Their results show that travel time to CBD/subcenters
performs better than distance to CBD/subcenter, while congested travel time performs even better than
free-flow travel time.

154
Other studies explore the negative effects of localized congestion by accounting for the impact of
traffic counts in their estimation of house price. The estimation results usually show that, as expected,
areas with high levels of vehicle throughput have lower levels of house price (Hughes and Sirmans 1992;
Kawamura and Mahajan 2005).
In sum, most empirical analysis based on the hedonic price model confirms the positive effect of
accessibility to employment on housing prices. There is also some evidence on the negative effects of
traffic congestion on housing prices by affecting commuting costs or generating negative externalities at
the neighborhood level. What these papers do not address, however, is the identification of negative
congestion effects distinguished from the positive effects of job access on housing prices. Moreover, the
extent to which location response to congestion costs varies across income groups is not explicitly
addressed. This study aims to fill these gaps and to add to the empirical evidence on the effects of
accessibility and congestion on housing prices.
3 Research approach
Based on the previous theoretical and empirical research, two basic hypotheses are developed
here:
H1: Traffic congestion will negatively impact housing price by reducing the regional accessibility
advantage of locations and increasing local traffic density, all else equal.
H2: Higher income neighborhoods will be more sensitive to accessibility to employment,
especially under congested conditions, due to the higher opportunity cost of commuting time by higher
income households, all else equal.
To test these hypotheses, the relationship between members of a set of potential explanatory
variables is explored through the descriptive analysis. The results of preliminary exploration indicate that
there are two major methodological issues, which are discussed below.

155
The hierarchical data structure problem
Housing prices vary within a region because of differences in structural characteristics as well as
locational characteristics. While data on structural characteristics are disaggregated, locational
characteristics are usually aggregated at different geographic scales (such as census blocks, census tracts
and counties). This implies that there is an inherent spatial autocorrelation problem among houses located
within the same neighborhood (Giuliano et al. 2010). Previous studies have shown that when data are
aggregated at different geographic scales, a single-stage estimation generates downward biases in
standard errors (e.g. Giuliano et al. 2010; Habib and Miller 2008; Jones 1991; Moulton 1990). Thus, there
is a need to apply an appropriate framework to account for the within-neighborhood correlation.
Identification of congestion effects
Transportation studies suggest that more accessible locations are also more attractive and will
generate more local traffic and increase travel time delay to other locations. Thus, the correlation between
measures of accessibility to employment and measures of traffic congestion could be high. This would
increase the risk of multicollinearity within a hedonic price model. Thus, crucial to this study is to the
ability to disentangle the positive effects of accessibility from the negative effects of congestion in the
hedonic regression analysis.
3.1 Basic model development
Based on the above discussion, the basic hedonic price model specifications include structural
characteristics, accessibility/congestion levels and other neighborhood characteristics. To deal with the
spatial autocorrelation problem, two approaches are usually applied in previous studies: the spatial
econometric approach and the multi-level or hierarchical linear modeling approach (Giuliano et al. 2010;
Habib and Miller 2008). For example, some studies apply locally weighted regression techniques to allow
the effects of housing and locational characteristics to vary over space and to "reduce spatial
autocorrelation without imposing arbitrary contiguity matrices or distributional assumptions on the data"

156
(McMillen and Redfearn 2010, 712). Other studies apply the multi-level modeling technique to account
for the hierarchical data structure and to consider housing structural effects and locational effects
simultaneously (Giuliano et al. 2010; Habib and Miller 2008; Jones 1991; Jones and Bullen 1993). To
deal with the hypothesized hierarchical data structure, this study applies the latter approach.
The hierarchical linear model includes the simpler random intercept model and the more general
random slope model. In modeling the intra-urban house price variation, the former model allows different
areas to have different averages but the same relationships between house price and all predictors, while
the latter model allows the price effect of some variables to be random. This study starts with a simple
two-level random intercept model, assuming that houses are nested within neighborhoods and that the
average house price varies across neighborhoods. In the next part, the two-level spatial model is extended
to a three-level spatial model with the introduction of random slopes to examine to what extent the effects
of accessibility/congestion on house prices vary across neighborhoods of different income or economic
status.
Within the hierarchical linear framework, the basic two-level spatial model is structured by the
following two equations. The first-level is given by
lnP

β

∑ β
X

ε

, ε

~N0, σ
(4-1)
where lnP

= natural log of housing transaction price for unit i in census tract j,
β

and β

= parameters,
X

= pth structural attribute of house unit i in census tract j,
ε

= normally distributed random term at level 1.
Here, census tracts are chosen as the "neighborhoods" because: 1) they are the basic units where
socioeconomic characteristics are readily available from the census data; 2) compared with larger spatial
units such as municipalities or zipcode areas, they are relatively small and homogeneous (Giuliano et al.

157
2010); and 3) the boundaries of census tracts are usually defined by arterials and highways, which implies
that the estimated travel time to access other locations would not vary that much for housing units within
a census tract but would be more likely to differ across housing units in different census tracts. Equation
(4-1) captures the "within-tract" variation of house price. The second-level equation is structured to
represent variations across neighborhoods such that:

γ

∑ γ
Z

u

,u

~N0, τ
(4-2)
where γ

= fixed intercept
Z

= the qth independent variable for census tract j (see discussions in 3.3)
γ

= fixed parameters for Z

u

= normally districted random term at level 2
The complete model for housing price would be:
lnP

γ

∑ γ
Z

∑ β
X

u

ε

, Covε

,u

0 (4-3)
The model implies fixed price effects of all housing structure characteristics and locational
attributes but varying intercepts γ

u

for the census tract average price. All residuals are mutually
independent. Compared with the "separate regression approach," the hierarchical linear modeling allows
for decomposing the variance at each level (Jones 1991). The variance component of level 2 (τ
) is also
equal to the covariance between any two housing units (i and i’) in a census tract j, and the "intraclass
correlation coefficient" can be specified as (Snijders and Bosker 2012, 18, 49):
ρ

(4-4)
The model expects housing units within a neighborhood to be more alike than those drawn
randomly from the study area (Jones and Bullen 1993). A zero value of ρ implies that intraclass
correlation does not exist, so that the multi-level modeling technique performs no better than the ordinary

158
least squares (OLS) estimation (Snijders and Bosker 2012). This parameter is also analogous to the
contribution of between-tract variation to the total variation of house price across the study area (Snijders
and Bosker 2012).
To deal with the possible temporal dimension, the two-level model could be expanded to a three-
level model to capture the spatial and temporal variation of house prices. For example, some studies
suggest a hierarchical structure where a separate level 2, representing each time period, is added so that
the sale price of houses (level 1) is nested within each transaction time period (level 2) for each census
tract (level 3) (Jones and Bullen 1993). This formulation enables us to examine whether the trend of the
house-price annual inflation varies by different districts (Jones 1991; Jones and Bullen 1993). However,
this formation also assumes that the random variations for a given year within a neighborhood are
independent of the random effects from other years within the same neighborhood (and independent of
those in other neighborhoods). Since temporal data are likely to be correlated across time periods, this
assumption may be doubted easily.
An alternative way to account for the house price variation over time is to maintain the two-level
spatial model and include the time control variable by fixed effects, using, for example, dummy variables
for each transaction year. This approach may be appropriate when the housing price inflation of different
neighborhoods does not vary that much from the regional trend (Jones and Bullen 1993). This study uses
the latter approach, because I find in the preliminary data analysis stage that few neighborhoods diverge
from the general trend of housing price inflation in Los Angeles County. Moreover, transaction time
periods can be regarded as unique categories, and the inclusion of time periods as fixed effects allows us
to directly examine the changes in average house prices for each different period (Snijders and Bosker
2012). Since this study mainly concentrates on spatial variations of house prices, while the temporal
variation is examined only because we want to avoid the issues of omitted variables and biased estimates
of parameters, I apply the easy solution and kept the two-level hierarchical structure.

159
3.2 Identification of (regional) congestion effects
A main purpose of this research is to examine whether traffic congestion negatively impacts
housing prices through affecting commuting costs. However, as discussed, measures of accessibility and
measures of congestion could be highly correlated, generating the multicollinearity problem. Thus,
instead of generating a separate measure of traffic congestion, this study identifies its effect indirectly by
constructing two groups of accessibility measures, one associated with free-flow travel time and the other
with congested travel time, estimating two separate price models using the two measures and comparing
the estimated coefficients of the two measures from the two equivalent regressions. This method follows
previous studies on congestion effects by Graham (2007). The basic hypothesis is that if congestion does
have an impact on house prices, the congested travel-time-based measure would be a better proxy of
location to employment experienced by households and should perform significantly better in predicting
house price, while the free-time-based measures would overestimate the accessibility advantage of
locations.
To demonstrate this, simplified versions of complete equations of the hierarchical linear model
based on the two measures are illustrated as follows:
lnP
∑ X
(4-5)
lnP
∑ X
(4-6)
where and represent the peak hour (congested) travel time based access measure and the
free-flow travel time based access measure, respectively (using, for example, the gravity-based measures),
X
represents any other control variable, ,
and are parameters for the fixed component of the
two equations, and and are the random component of the two equations (all subscripts are suppressed
for simplicity). Since is assumed to be a better measure of location to employment, equation (4-5)
is considered as the "true" model and can be modified as equation (4-7), using the AccFF variable:

160
lnP
X


X

X

X

(4-7)
Therefore, the random term in equation (4-6) can be decomposed into two parts: the random
term µ from the true model and the term , which represents the "loss" of
accessibility advantage and is correlated with . Thus, the estimated coefficient α
obtained from
(4-6) is biased, with the direction of bias depending on the expected signs on the omitted variable
as well as the correlation of AccFF and the omitted variable. On the one hand, the
term is expected to have a negative effect on house prices, because higher levels of
accessibility loss imply higher levels of congestion delays and a lower "accessibility premiums" for
locations. On the other hand, is expected to be positively correlated with ,
because the gap between accessibility measured under congested condition and that measured under free-
flow condition becomes wider as the level of accessibility increases (Graham 2007). Thus, the estimated
congestion effects based on AccFF in equation (4-7) should be downward biased, making the finding of
statistical significance more difficult.
3.3 Housing market segmentation
Another purpose of this study is to test whether the marginal price of accessibility varies over
neighborhoods of different income or economic status. This would imply whether and to what extent
different income groups value accessibility to employment and travel time savings differently. Of course,
a more appropriate way to account for the differential accessibility effects is to partition the sample based

161
on home buyers’ economic status, which, however, is usually not available. This study uses an alternative
method to delineate income clusters by categorizing neighborhoods into different groups based on the
aggregate information on their residents, which is more readily available from census data. Since the
standard urban economic theory also predicts that households of different income levels will cluster at
different locations, we believe this method is appropriate in delineating income clusters or groups within a
region.
As suggested by previous studies, one direct approach to account for the spatial variability of
accessibility effects is to estimate the basic hedonic price model (two-level) separately for each category
of neighborhoods. All variables in the basic hedonic model except the one measuring neighborhood
income level could be used for the estimation of separate price models. One problem with the first
approach, however, is that the group size may not be consistent, and some types of neighborhood may
have fewer observations. Conceptually, it is likely that housing sales are more frequent in some types of
neighborhoods, such as high-income neighborhoods or peripheral areas (Giuliano et al. 2010). In terms of
this situation, a three-level random slope model might have some advantage over the separate estimation,
because those groups do not contain enough information "borrow strength" from other groups (Jones
1991; Snijders and Bosker 2012). The three-level random slope model for the price of house i in census
tract j in neighborhood type k is specified as:
Level 1: lnP

β

∑ β
X
, ε

,ε

~N0, σ
(4-8)
Level 2: for intercept β

δ

∑ δ
Z
,
u

,   u

~N 0, τ
(4-9)
for slopes of level 1 variables β
δ

u
,u
~N0, τ
(4-10)
Level 3: for intercept δ

γ

v

,v

~N0, φ
(4-11)
for slopes of level 1 variables δ

γ

v

,v

~N0, φ
(4-12)

162
for slopes of level 2 variables δ

γ

v

,v

~N0, φ
(4-13)
In the specification, each level 1 variable (X
, represents the pth level-1 variable) is allowed to
be random at both level 2 (equation 4-10) and level 3 (equation 4-12), while each level 2 variable (Z
,

represents the qth level-2 variable) is allowed to be random at level 3. In practice, however, a model as
complex as this could have problems in convergence if all the explanatory variables are allowed to be
random. To make the estimation efficient, only the accessibility variable is allowed to vary at level 3,
defined by income clusters. Thus, the complete model is specified as
lnP

γ

β
X
, δ

Z
,
δ

Acc

v

Acc

v

u

ε

,cov v

,v

;covε

, u

0; cov u

,v

0; cov u

,v

0
(4-14)
where X
, and Z
,
represent any other explanatory variables at level 1 and level 2, respectively, the
first term in the brackets represents random effects of the chosen explanatory variables, and the second
term in the brackets represents random intercepts in the three models. At level 3, the two random effects
v

,v

are correlated, and their covariance is defined as φ

. The multi-level model presumes that
random effects across different levels are independent, so that their covariance is zero (Snijders and
Bosker 2012). This study uses the latter approaches to examine the variation of accessibility effects by
income clusters. The categorization of neighborhoods by income is illustrated in the following discussion.
3.4 Construction of locational variables
The basic structural characteristics of houses used in the hedonic price model include lot size of
single-family housing ( , square foot), building area ( , square foot), age of house ( ),

163
owner-occupied ( ),
20
the presence of a pool and the presence of a fireplace. Other housing
characteristics such as number of bedrooms and number of bathrooms are not included, because they are
highly correlated with the building area of houses.
As for the locational attributes, typically, four types of variables are included in the price model:
measures of accessibility to employment, localized congestion (traffic density), Socio-demographic
characteristics of residents, characteristics of employment activities, and land use characteristics. Besides
that, distance to the coast ( ) is also included to account for the most important amenity feature of
neighborhoods in the Los Angeles metropolitan area (Giuliano et al. 2010).
21
Dummy variables are also
created to proxy for which school district each housing unit falls inside (00 ), which also captures the
variation in school quality and the public services mostly associated with it (Franklin and Waddell 2003).
The 2000 census tract boundary is used as the spatial reference for all the neighborhood-level variables.
All the accessibility/congestion measures and distance-based measures are constructed using the GIS
technique, and all the neighborhood-level measures are spatially joined to the individual properties.
3.4.1 Accessibility to employment
Since the investigation of regional congestion effects depends on the difference in the two
accessibility measures, the construction of accessibility variables capturing the difference is essential. The
most commonly used measure of accessibility is the gravity-based accessibility measure originally
developed by Hansen (1959). As discussed in some previous studies, this measure allows the effects of
dispersed employment across the study area to be accounted for and performes as well as measures of
distance/travel time to multiple employment centers, while the potential multicollinearity problem
associated with the multiple distance measure is avoided (Ottensmann, Payton, and Man 2008). Since this
study focuses on the general price effects of accessibility to employment, not the role played by specific

20
More than 10 percent of the samples are not owner-occupied. Since they occupy a significant amount of the
sample, I decide to include them and use the dummy variable to examine whether renter-occupied houses are
devalued or not.
21
It’s important to recognize that the coastal distance captures not only the amenity of the ocean front, but also more
moderate temperatures and better air quality than inland locations.

164
employment centers, the generalized measure of accessibility is applied. The basic form of the
accessibility to employment measure used here is:

∑
exp ·

(4-15)
where
represents the amount of employment in location j,
represents the travel time (transport
costs) between i and j, and is the impedance parameter. The travel time is measured using either free-
flow information or congested information, which is key to identifying to what extent the accessibility
advantage of a location is constrained by traffic congestion. The AM peak travel time is used as the
congested travel time, and the night period (NT) is used as the free-flow condition. The two measures of
accessibility are thus specified as follows:

_ ∑
exp ·

(4-16)

_ ∑
exp ·

(4-17)
Following Giuliano et al. (2011), the impedance parameter β is set as the inverse of the average
commuting distance, which is 12.1 miles, according to the 2001 NHTS.
3.4.2 Localized congestion effects
To account for the negative effects of traffic congestion at the neighborhood level, such as
pollution and noise generated by high volumes of traffic, this study uses a simple measure of traffic
density defined as the number of vehicles passing through each census tract during the AM peak period
normalized by the total land area each tract (_ , vehicles per acre).
22
It is expected that
neighborhoods with larger volumes of traffic flow on the existing road network are more likely to
experience negative externalities, which are capitalized into housing prices.

22
Here traffic flows are normalized by area, not by central road miles or lane miles, to get a measure of traffic
intensity. This is because a large tract could have a road with lots of traffic, but much of the tract would not be near
the road.

165
3.4.3 Highway access
Distance to the nearest highway/freeway ramp is used as a measure of highway access. As
suggested by Waddell, Berry, and Hoch (1993), the effect of highway access is non-linear in that
"immediate negative externalities override the gains from increased proximity" (129). Following their
study, highway access is dummied out into three intervals: 0–0.5 mile (05 ), 0.5–2 miles (dhwy2),
and more than 2 miles as the reference case. The expected sign for 05 is negative, given the
possible negative effects of adjacency to a highway, such as noise and localized congestion, while the
expected sign on 2 is positive, as within distance, the positive access effects would dominate. Given
this micro-neighborhood feature of highway proximity effects, highway access variables are the only
locational variables generated at the housing unit level instead of the census tract level in this study.
3.4.4 Socio-demographic characteristics
Socio-demographic characteristics include population density ( ), percentage of white
population (_ ), median household income ( ) and percentage of population between
18 and 24 who have less than high school education level (_ ). The impact of population
density on housing prices is ambiguous; it might be positively associated with housing price, because
higher population density implies higher residential density, which in turn implies that more capital is
substituted for land due to higher land rents (Kulish, Richards, and Gillitzer 2012; Song 1996). It could
also depress housing prices, as high density places may generate such disamenity effects as overcrowding
and high crime rates (Bender 1981; Langer and Winston 2008; Wabe 1971). Percentage of white
population is used to control for the minority concentration of the neighborhoods and is expected to
increase housing prices due to the "segregation premium" that whites are willing to pay (Waddell, Berry,
and Hoch 1993). Median household income represents the economic status of average households in a
neighborhood and is expected to increase housing prices. Other variables representing average economic
status of residents in neighborhoods, such as percentage of population over 25 with a bachelor’s degree or
higher (_ ), poverty rate (_ ) and percentage of owner-occupied housing units

166
(_ ) are not included in the hedonic regression analysis, because they are found to be highly
correlated with the median household income measures, with the lowest correlation coefficient more than
0.73. Thus, the variable used in this model is expected to confound the effects of the above
variables. Finally, the percentage of population between 18 and 24 without a high school degree is
included to roughly proxy for the crime rates for each neighborhood. Conceptually, neighborhoods with
larger number of high school dropouts are more vulnerable to vandalism or crime activities, which
decrease the housing prices (Li and Brown 1980).
3.4.5 Economic characteristics
Employment density ( ) is used as a measure of the intensity of local job opportunities.
Like population density, its effects on housing price could be either positive or negative; it may increase
housing price because of the availability of jobs at the local level, or it may depress housing price as
households may prefer low density neighborhoods and value less the proximity to economic activities at
the local level, especially those activities generating negative externalities. Another density measure of
the employment density in the arts, entertainment and recreational sector (71 ) is also constructed
using all employment in the NAICS2-71 sector to represent the desired cultural amenity of neighborhoods;
it is expected to increase residential property values.
23

3.4.6 Land use characteristics
A set of variables represents the percentage of land in each census tract developed into different
use types, including single-family residence (_ ), multifamily residence (_ ),
office _ , industry (_ ) and commercial uses (_ ). These variables are
constructed to measure the positive or negative externalities generated from different land uses. Since the
prices of single-family houses are focused on in this study, neighborhoods with larger areas of single-
family residential use are expected to have higher-level housing prices, because buyers of single-family

23
The negative effects of concentration in some types of employment activities, such as the industrial production
sectors, are captured by the percentage of land that is developed into those associated activities (e.g., industrial land
use). See the following discussion.

167
houses are expected to have a higher preference for those single-family dominated communities. The
larger areas for multi-family uses might also increase prices for single-family houses if the two residential
uses comply with each other. The effects of other land use types, however, are mixed. Like the effects of
employment density, land use for commercial or office purposes may or may not be compatible with
single-family residential use, while land use for industrial purposes is expected to depress housing prices
(Habib and Miller 2008). A dummy variable representing the existence of a park ( ) is also included
in the hedonic analysis to account for its desired amenity effects at the neighborhood level.
4 Data
This study uses Los Angeles County as the study area. Los Angeles County is the core of the Los
Angeles Metropolitan Statistical Area (MSA), which consists of four other counties, Orange, Riverside,
San Bernardino and Ventura, as defined in 2000. Based on the definition of the MSAs by the U.S. Census
Bureau, there is a "high degree of socio-economic integration" between Los Angeles County and the other
counties.
24
This means that such measures as accessibility to employment for households in Los Angeles
County should be constructed at the regional scale. This is discussed in detail in the following part.
4.1 House price data
The data on house prices and structural characteristics is from DataQuick Information Systems,
Inc., which compiles housing sales data from the Los Angeles County assessors’ office and includes the
structural characteristics of houses such as building size, lot size, number of bedrooms, bathrooms,
longitudes/latitudes and census tract IDs (2000 boundary), as well as housing transaction information
such as sale price, sale date and transaction year. I chose the 2001–2005 period because the economic
activities and housing market in this period were relatively stable before the 2007 bubble and recession.
25

The multi-period data is used because we expect that without an economic shock, households’ locational
responses to spatial attributes, as reflected in the sale prices, should not change that much. Dealing with

24
Source: http://quickfacts.census.gov/qfd/meta/long_metro.htm
25
This study period is also used for Chapter 3.

168
multi-period data might also avoid the bias that home sales may be more frequent in some areas than
others in a given year.
For the years 2001–2005, there are in total 621,218 sales, among which 435,226 of them are
single-family detached houses.
26
Some sales in the database are excluded because of missing information
on one or more variables used in the model or missing geographic information that means that they could
not be correctly geocoded. Thus, the remaining records include 403,332 sales of single-family detached
houses. A 10 percent random sample is drawn from the remaining records to conduct the data analysis,
which includes 40,333 sales. The chosen housing structural characteristics and their descriptive statistics
are shown in Table 4-a1 (in the appendix). The table shows that the mean and median values are both
reasonable. For example, the median sale price of single-family houses is $236,000 in 2001, with an
average annual inflation rate of about 20 percent. A typical single-family housing unit sold in 2001 would
be an owner-occupied housing unit aged about 43 years, with a building area of about 1514 square feet
and a land area of 6754.5 square feet, having fireplaces but no swimming pools. The average and median
values of structural characteristics of houses sold in other years in the study period (2002–2005) do not
seem to differ greatly from the values of those sold in 2001.
The spatial distribution of single-family housing prices for the study period is illustrated in Figure
4-1. The map shows that those housing units with the highest quintile of sales prices are mainly
concentrated in a few areas. The first area starts at the west boundary of Los Angeles County, covers the
mountainous area between the S-101 (Ventura) Freeway and the S-1 Freeway and extends south along the
coastline until the end of the I-110 Freeway. This area covers most of the places with a high value of
cultural or environmental amenities, such as the coastal line of Malibu, the corridor along the S-2
Freeway, including Hollywood, West Hollywood, Beverly Hills and Santa Monica, and the west coastline
of Los Angeles County. Another place showing a concentration of the highest quintile of housing sales

26
This study only chose those properties with price category coded as “A” in the record, which accounts for
approximately 27.5 percent of the records in the whole dataset for the 2001–2005 period.

169
prices is the City of Pasadena. On the other hand, the map also shows that samples belonging to the
lowest quintile of housing sale prices mainly concentrate in the central area of Los Angeles county, which
is mainly bounded by the I-10 Freeway on the north, the S-91 Freeway on the south, the I-710 Freeway
on the east and the I-110 Freeway on the west. There are also some samples of the lowest sale prices
located in the northern part of Los Angeles County and concentrating around Lancaster City. There are
very few sales samples in the Los Angeles downtown area, which shows on the map as a blank area in the
central part of Los Angeles County.

Figure 4-1 Spatial distribution of selected samples
Simply valued by distance to downtown Los Angeles, there are many places that are of almost
equal accessibility but with heterogeneous housing prices. This implies that many other factors, such as
access to other employment centers and access to environmental and cultural amenities, might be
influencing housing prices. However, as discussed, simply physical distance does not tell much about the

170
actual travel time and the relative ease of access to other locations. In the following, I will briefly describe
the spatial patterns of access to jobs under different scenarios and compare them with the distribution of
housing prices.
4.2 Employment data
The National Establishment Time Series (NETS) is used to construct the accessibility variables
and the local employment characteristics variables. The dataset provides establishment-level information,
including address, employment, establishment category and industrial sector.
27
Because employment
opportunities outside Los Angeles County are also relevant for households of the county, I extracted the
NETS dataset of the Los Angeles region from the California dataset for the year 2000, which is used as
the background year for measuring access to jobs and other opportunities at the regional and local scale.
For the purpose of this study, all self-employed standalone firms (employment size = 1) are excluded,
because these can hardly be considered as employment opportunities. Moreover, firms belonging to the
nonurban sectors, including the agriculture (SIC1 = 1, or NAICS2 = 11) and mining (SIC1 = 2, or
NAICS2 = 21) sectors, are also excluded. Based on the definition, the total number of jobs for the Los
Angeles metropolitan area in the year 2000 is 7,864,032, associated with 541,674 establishments. The
establishment-level NETS data are matched to the 2000 census tracts boundaries.
4.3 Transportation network data
The transportation network data of the Los Angeles region are obtained from the SCAG
(Southern California Associations of Governments) base year network files for the Regional
Transportation Plan of 2003.
28
The use of the regional transportation model allows accessibility measures
to be constructed based on congestion conditions, so that the effects of travel time delays on housing
prices can be examined. The free-flow travel time between each pair of locations is defined as the shortest
travel time on the road network during night periods (NT) plus the estimated travel time for the "network

27
See detailed discussion in Chapter 3.
28
See detailed discussion in Chapter 3.

171
collector" between the centroid of each census tract and the nearest road network intersection; the
congested travel time between pairs of locations is defined similarly, but the network travel time for AM
peak periods is used.
Using the transportation network data and the employment data, accessibility to jobs is calculated
at the census tract level. Table 4-a2 (in the appendix) shows that under free-flow conditions, tracts are on
average accessible to a larger number of jobs, with a ratio between free-flow time accessibility and AM-
peak time accessibility of more than 1.7. The spatial patterns of access to jobs under congested and free-
flow scenarios are illustrated in Figure 4-a1 and Figure 4-a2 (in the appendix), respectively. The two
maps indicate that the spatial patterns of the two job access measures differ from each other. Figure 4-a1
(in the appendix) shows that census tracts of the LA county belonging to the fifth quintile of job access
generally concentrate in an area starting with northeast and downtown Los Angeles, extending along the
I-110 Freeway in the form of a wide corridor that is bounded by the I-405 Freeway on the east, the I-710
Freeway on the west and the S-91 Freeway on the south. Figure 4-a2 (in the appendix) shows that one
area having the highest quintile of job access also starts with northeast and downtown Los Angeles,
extends south along the I-110 Freeway and I-710 Freeway corridor until it meets the S-91 Freeway, and
extends southeast along the I-5 Freeway until meeting the S-91 Freeway. The two maps indicate that
those places, which are the same physical distance to downtown Los Angeles, vary in terms of actual
access to jobs. Moreover, some areas have different levels of access to jobs in the AM peak and off-peak
hours. For example, the corridor along the I-5 Freeway belongs to the highest quintile of job access
during off-peak hours but not during the AM peak hours. This result indicates that congestion-induced
travel time delays will alter the accessibility pattern. It would be interesting to examine how that altered
accessibility pattern might influence the bid-rent curve and households’ locational responses.
Comparing the house price map and the two accessibility maps, it is found that higher-price
housing units vary in their levels of accessibility to employment, so do those lower-priced housing units.
For example, some places in which are concentrated the housing units belonging to the highest quintile of

172
sale prices, such as Hollywood, West Hollywood, Beverly Hills and Pasadena, also have the highest level
of accessibility to employment under either congested or free-flow conditions, while other places such as
the west coastline along the S-1 Freeway have the least accessible level. Similarly, some places
concentrated with the lowest-priced samples, such as the I-110 corridor, are highly accessible to jobs,
while other places also heavily concentrated with the lowest-priced samples, such as Lancaster, are least
accessible. This might imply that except for access to jobs and commuting time and costs, other location
factors also influence the spatial structure of housing prices.
4.4 Population data
The model also includes other socio-demographic characteristics at the census tract level, derived
from the 2000 U.S. census of population. The remaining locational characteristics include population
density, percentage of white population, percentage of population under poverty, median household
income and percentage of less educated population aged between 18 and 24. The unified school districts
as of 2000 are also used for creating dummy variables.
Table 4-a2 (in the appendix) shows the summary statistics of socioeconomic characteristics.
Census tracts on average have about 13 persons per acre, with about 60 percent of population that is white,
27 percent of the population belonging to the group prone to criminal activities and a median household
income of about $56,450. Comparing the mean and median values, it is found that these variables are
normally distributed. The standard deviation values show that the values of these variables vary largely
across census tracts.
Figures 4-a3 to 4-a5 plot the spatial distribution pattern of three key socioeconomic variables:
population density, median household income and poverty rates. Comparing the three maps, it is
interesting to find that the area starting from northeast and downtown Los Angeles and extending south
along the I-110 Freeway until the S-91 Freeway is concentrated with census tracts having the highest
level of poverty rate or the lowest level of median household income as well as the highest level of

173
population density. As discussed above, this area also has the highest level of job access under either free-
flow or congested conditions. This result implies that lower income groups are more likely to be
constrained to the central part of Los Angeles County due to possible housing market discriminations.
4.5 Land use data
The land use information was obtained from SCAG’s region-wide land use (LU) dataset for the
survey year of 2001. The parcel-level data spatially joined to the 2000 census tract boundaries and the
areas of each land use type are aggregated for each census tract. Thus, the percentages of different land
uses are measured for each census tract, including single-family residential, multi-family residential,
industrial and commercial uses. Moreover, the LU dataset also includes detailed land use categories such
as regional parks, which provide useful information for measuring localized amenity effects.
Table 4-a2 (in the appendix) shows the summary statistics of land use characteristics. On average,
census tracts have about 40 percent of land area that is developed into single-family residential uses, 4.2
percent into multi-family residential uses, less than 0.5 percent into office uses, 4.8 percent into
commercial uses and 3.2 percent into industrial uses. Moreover, about half of the census tracts have at
least one park. The minimum and maximum values and the standard deviations of those variables show
that these land use characteristics also vary greatly across census tracts.
5 Results
The results of the hierarchical linear models are presented for the different levels. The two-level
spatial model is used as the basic model. Then the model is extended to examine the differential effects of
accessibility by income clusters, using the three-level spatial model allowing for random slopes for
accessibility. The dependent variable in all regressions is the natural log of housing price. The natural log
form is also used for some explanatory variables, including lot size and building area at level 1 and
locational characteristics, including distance to the coastline, measures of population density, employment
densities and traffic flow density at level 2, because of their skewed distribution. Moreover, the four

174
dummy variables representing the transaction years of 2002 to 2005 are controlled in all regressions,
while the transaction year 2001 is used as the reference. The maximum likelihood (ML) estimation is
used for all estimations.
5.1 Basic model estimation
Table 4-1 shows the general findings of the basic price model. To examine the relative
contribution of housing structural variables and locational variables to the total variance of housing prices,
a "step-wise modeling" process is applied. Column (1) in Table 4-1 shows the results of an "empty
model" without any explanatory variables. The purpose of running this regression is to test whether there
exists autocorrelation within a census tract. The degree of autocorrelation ρ is defined as the proportion of
level 2 variance to the total variance of the dependent variable. The estimated parameter τ = 0.23 and σ
= 0.16 yields an intraclass coefficient of ρ = 0.58. This implies that there exists an important similarity
between prices of different housing units in the same census tract, and level 2 accounts for a larger
proportion of variance. Thus, the multi-level model is supported by the data.

175
Table 4-1 Results of Two-level spatial model
29

(1) (2) (3) (4) (5) (6)
Fixed effects
Transaction
years
2002 0.156*** 0.150*** 0.150*** 0.150*** 0.150***
(33.06) (39.37) (39.52) (39.58) (39.58)
2003 0.341*** 0.338*** 0.337*** 0.338*** 0.338***
(72.97) (89.06) (89.28) (89.40) (89.40)
2004 0.592*** 0.587*** 0.587*** 0.588*** 0.588***
(126.56) (154.67) (154.97) (155.19) (155.17)
2005 0.805*** 0.801*** 0.801*** 0.802*** 0.802***
(170.01) (208.00) (208.61) (208.88) (208.86)
Structural characteristics
lnbldg 0.471*** 0.471*** 0.467*** 0.468***
(94.39) (96.01) (95.15) (95.42)
lnland 0.128*** 0.125*** 0.125*** 0.125***
(36.10) (35.40) (35.71) (35.62)
age -0.000278**-0.000105 -0.000306** -0.000284**
(-2.68) (-1.03) (-3.00) (-2.78)
ownocc 0.0157*** 0.0168*** 0.0161*** 0.0162***
(4.05) (4.35) (4.17) (4.18)
pool 0.0637*** 0.0607*** 0.0617*** 0.0615***
(18.09) (17.27) (17.56) (17.51)
fireplace -0.00685* -0.00963** -0.00891** -0.00884**
(-2.30) (-3.25) (-3.01) (-2.99)
Locational characteristics
lncoast -0.120*** -0.121*** -0.111***
(-19.19) (-20.94) (-19.06)
dhwy05 -0.0254** -0.0327*** -0.0360***
(-3.23) (-4.23) (-4.65)
dhwy2 -0.00203 -0.00917 -0.011
(-0.30) (-1.38) (-1.65)
lnpopden 0.02 -0.0161* -0.0133
(1.90) (-2.10) (-1.73)
white 0.00317*** 0.00536*** 0.00504***
(11.14) (18.49) (17.73)
medhhinc 0.00655*** 0.00637*** 0.00631***
(22.40) (23.69) (23.37)
per_lessedu -0.00436***-0.00349*** -0.00358***

29
Dummy variables for the unified school districts of 2000 (00 ) are included in the regression, but the estimated
coefficients are not reported here.

176
(-12.24) (-10.53) (-10.77)
lnempden 0.0360*** 0.0364*** 0.0370***
(5.68) (6.25) (6.33)
lnempden71 0.178*** 0.147*** 0.152***
(7.09) (6.28) (6.47)
per_sfam -0.000259 0.000345 0.000263
(-0.92) (1.32) (1.01)
per_mfam 0.00470*** 0.00633*** 0.00627***
(8.06) (11.48) (11.33)
per_office -0.00467 -0.00349 -0.0039
(-1.71) (-1.38) (-1.53)
per_com -0.00198* -0.00202* -0.00188*
(-2.12) (-2.34) (-2.18)
per_indu -0.00146** -0.00172*** -0.00189***
(-2.67) (-3.41) (-3.75)
park -0.00368 -0.00191 -0.000433
(-0.42) (-0.23) (-0.05)
lnFlowden_AM 0.0312*** 0.00442 0.00158
(6.59) (0.96) (0.34)
Accessibility
AccE_AM 0.000739***
(17.55)
AccE_NT 0.000425***
(17.16)
_cons 12.79*** 12.40***7.837*** 7.433*** 7.243*** 7.246***
(1128.08) (1054.35)(186.70) (124.73) (125.23) (125.03)
Random effects
level 2
Var(cons) τ
0.225 0.236 0.139 0.03 0.03 0.03
S.E. 0.008 0.008 0.005 0.001 0.001 0.001
level 1
Var(cons) σ
0.165 0.083 0.054 0.054 0.054 0.054
S.E. 0.001 0.001 0.0004 0.0004 0.0004 0.0004
N 40333 40333 40333 40333 40333 40333
Group size(level
2)
1909 1909 1909 1909 1909 1909
ll -23752.2 -10528.9 -1886.4 -428.5 -287.6 -293.3
aic 47510.4 21071.9 3798.7 1006.9 727.1 738.6
chi2 . 38059.4 79732.6 91625.2 94315.8 94199.9
R12 0.18 0.51 0.81 0.81 0.81
t statistics in parentheses;
Bold and Italic p<0.1, * p<0.05, **p<0.01, *** p<0.001

177
The second model includes time fixed effects. Column (2) shows that all the time variables are
positively significant, showing a strong trend of housing price increase during the study period. To
examine the contribution of added variables, a pseudo R-square is constructed based on the "level-one
explained proportion of variance" as (Snijders and Bosker 2012, 110–113):
1
∑ X

  (4-18)
which is defined as "the proportional reduction in mean squared prediction error." In model (2), the R

amounts to the proportional reduction in the value of τ +σ and is estimated to be 1-(0.32/0.39) = 0.18.
The third model adds housing structural variables, which greatly increase the level-1 explained
proportion of variance to 0.516. This results show that the two size variables—building areas and lot size
in natural log forms—are strong predictors of housing prices. A 1 percent change in building area
increases housing prices by 0.47 percent, and a 1 percent change in lot size is associated with a 0.13
percent change in housing prices. The age of houses shows a significant negative impact on housing price.
A dummy variable presenting owner occupancy status contributes significantly to housing prices, so does
the dummy variable representing the presence of a pool. The presence of a fireplace, however, reduces the
premium on housing prices.
Column (4) adds all the locational variables except the accessibility measures. Columns (5) and
(6) show the full model with accessibility measures based on congested travel time and free-flow time,
respectively. The results show that adding locational variables does not affect the estimated coefficients
for structural characteristics very much, but adds greatly to the explanatory power of the model. The
regression coefficients for both of the accessibility measures are highly significant, with expected positive
signs. Moreover, accessibility to employment based on congested travel time has a higher t value than
accessibility based on free-flow travel time, while the regression coefficient for the congested time
accessibility is about 1.7 times as large as that for free-flow time accessibility. To further test whether

178
there exists a significant difference in the estimated coefficients between the two measures, the
asymptotic t-test is used to compare the difference between the two coefficients (Allison 1999, 188):
t

AM NT
.. AM
.. NT
(4-19)
where s.e.(
•
) is the estimated standard error, and the degree of freedom is 1. The estimated t-value is
41.31, highly significant at the 0.0001 level. This again suggests that congested time accessibility
performs significantly better in terms of predicting housing prices. Given the significant difference in the
estimated coefficients, however, the magnitude of the price effects of job accessibility is relatively small.
Using the transformation 100(e
– 1), each one unit increase of accessibility based on congested travel
time only yields an increase in housing price of about 0.07 percent, while each one unit increase in free-
flow-based accessibility yields an increase of only 0.04 percent in housing price. This is consistent with
findings of previous studies that access to employment has a small effect on housing prices.
Despite the small effect of accessibility, eliminating this effect may cause biased estimates of
other coefficients. One most relevant effect is the localized congestion externality, measured by traffic
density in natural log forms. The results in column (4) indicate that when accessibility to employment is
not controlled in the model, the estimated coefficient for the localized congestion effects is positive and
significant, contradictory to the common wisdom. This might be attributable to the fact that more
congested areas have more traffic flows because they are also more accessible from/to other places, so
that the localized congestion measure may take over the accessibility effects. This variable becomes
insignificant once the accessibility effect is controlled.
The regression results for the full model also show that most of the control variables show
significant price effects with expected signs. Distance to the coastline is found to be a strong predictor of
housing prices such that the sale price decreases by 0.11 to 0.12 percent for each 1 percent increase in
distance away from the coast. Highway access has non-linear effects on housing prices, in that proximity
to a highway ramp within 0.5 miles reduces housing prices by 2.5 percent to 3.6 percent, implying that

179
negative externalities at this distance are not enough to offset the positive access benefits. Moreover,
proximity to a highway of between 0.5 and 2 miles shows no significant effects on housing price.
In terms of socio-demographic characteristics of neighborhoods, the estimated coefficients for all
the four variables included are significant. The effect of population density is positively significant in
model (4), but it becomes negative in the full model. The possible explanation is that neighborhoods of
higher population density might also have higher levels of accessibility to jobs; when the accessibility
variable is omitted, the positive effects of population density dominate, and the negative effects of
population density, such as overcrowding and high crime rates, are masked. The percentage of white
population has a positive effect on housing price in that every additional 1 percentage point increase in
the variable adds approximately a 0.5 percent premium to housing price. Similarly, median household
income reflects neighborhood economic status and has an expected positive effect on housing price. The
percentage of population aged 18–24 who are high school dropouts within a neighborhood significantly
decreases housing price, as expected.
The two employment density measures also contribute positively to housing prices. The results of
models (5) and (6) suggest that whereas the sale price increases by 0.04 percent for every 1 percent
increase in aggregate employment density, it increases by 0.15 percent for every 1 percent increase in the
density of arts and entertainment employment. This implies that there is a premium for being in
neighborhoods concentrated with entertainment and recreational activities.
With regard to the land use characteristics, housing prices increase with higher percentages of
census tract land that is for multi-family use but decrease with a higher percentage of commercial and
industrial use. This suggests that multi-family use is compatible with single-family residential use, while
the latter two uses show incompatibility. However, contrary to our expectation, estimation results also
show that neighborhoods with a larger percentage of single-family use do not have significantly higher
housing prices. The presence of parks in a neighborhood does not add a premium to housing prices either.

180
5.2 Differential effects of accessibility by income
This section addresses a different question of whether price effects of accessibility vary among
income clusters. Specifically, all census tracts within Los Angeles County are categorized into five
income clusters based on the quintiles of neighborhood income variables. Two census tract-level variables
are used in this study to define income clusters: median household income ( ) and percentage of
population under poverty (_ ). The summary statistics of independent variables within each
quintile of the two variables are listed in Tables 4-2 and 4-3. The basic statistics suggest that higher-
income neighborhoods have higher average housing prices, while the level of accessibility is greater in
the lowest-income neighborhoods but does not show much variation across other income clusters. The
correlation between accessibility index and housing prices is higher in the three middle-income clusters
but is much lower in the lowest and highest income groups.
Table 4-2 Summary statistics of housing price and Accessibility by quintiles of median household
income

Range ($)

Ln( pr
ice)
AccE_A
M(,000
jobs)
Cor(Acc
E_AM,
lnpr)
AccE_NT(
,000 jobs)
Cor(Acc
E_NT,
lnpr)
lowest
income
0—29,861 mean 12.34 554.30 0.03 928.17 0.03
(n=3840) sd 0.52 132.65 216.84
lower-
middle
income
29,862—40,196 mean 12.50 397.06 0.27 670.38 0.27
(n=7404) sd 0.54 201.06 348.52
middle
income
40,197—50,183 mean 12.69 395.20 0.34 679.90 0.33
(n=9092) sd 0.52 168.60 304.96
upper-
middle
income
50,184—66,731 mean 12.98 362.16 0.22 624.20 0.22
(n=10775) sd 0.58 167.15 296.13
highest
income
66,732—200,001 mean 13.22 307.31 0.16 539.40 0.20
(n=9222) sd 0.59 148.57 263.92

181
Table 4-3 Summary statistics of housing price and Accessibility, income clusters defined by quintiles
of poverty rate

Range (%)

Ln(pri
ce)
AccE_AM
(,000 jobs)
Cor(Acc
E_AM,
lnpr)
AccE_
NT(,00
0 jobs)
Cor(Acc
E_NT,
lnpr)
lowest
income
25.4—100 mean 12.34 557.42 -0.02 926.74 -0.01
(n=3360) sd 0.52 147.65 241.36
lower-
middle
income
16—25.3 mean 12.52 449.46 0.29 762.30 0.30
(n=6243) sd 0.53 170.50 296.58
middle
income
9.5—15.9 mean 12.61 383.03 0.26 658.26 0.26
(n=9715) sd 0.51 188.86 339.73
upper-
middle
income
5.2—9.5 mean 12.84 363.12 0.18 624.40 0.18
(n=10726) sd 0.50 158.60 287.74
highest
income
0—5.2 mean 13.33 301.57 0.04 529.13 0.08
(n=10289) sd 0.60 149.40 259.47

Table 4-4 shows the results of three-level modeling, with level 3 defined by and
_ . All the explanatory variables in the two-level spatial model are also included in the three-level
model, while accessibility measures based on congested time and that based on free-flow travel time are
entered into the regression separately. The estimates of fixed terms are almost the same as those in the
two-level spatial model and are not discussed further. The most interesting issue here is whether or not the
price effects of accessibility are stronger in some income groups than in others. As defined in equation (4-
14), the estimated coefficient for the income group k is defined as δ

v

, which is in fact a
normally distributed variable with the mean δ

and a standard deviation of φ
.

182
Table 4-4 Results of three-level spatial model
30

Level3 identified by quintiles
of median household income
Level3 identified by
quintiles of poverty rate

(1) (2) (3) (4)
Fixed effects
Transaction years
2002 0.150*** 0.150*** 0.150*** 0.150***
(39.57) (39.57) (39.55) (39.55)
2003 0.337*** 0.337*** 0.338*** 0.338***
(89.31) (89.31) (89.31) (89.31)
2004 0.588*** 0.588*** 0.588*** 0.588***
(155.18) (155.16) (155.14) (155.12)
2005 0.802*** 0.802*** 0.802*** 0.802***
(208.78) (208.77) (208.75) (208.74)
Structural characteristics
lnbldg 0.470*** 0.471*** 0.471*** 0.472***
(95.58) (95.77) (95.74) (95.92)
lnland 0.127*** 0.126*** 0.127*** 0.127***
(35.99) (35.89) (36.08) (35.99)
age -0.000271**-0.000253* -0.000293** -0.000276**
(-2.64) (-2.47) (-2.85) (-2.69)
ownocc 0.0156*** 0.0156*** 0.0154*** 0.0155***
(4.03) (4.05) (3.99) (4.00)
pool 0.0628*** 0.0626*** 0.0634*** 0.0632***
(17.87) (17.82) (18.05) (18.00)
fireplace -0.00897** -0.00893** -0.00856** -0.00853**
(-3.03) (-3.02) (-2.89) (-2.88)
Locational characteristics
lncoast -0.128*** -0.118*** -0.138*** -0.127***
(-21.04) (-19.10) (-22.01) (-20.06)
dhwy05 -0.0342*** -0.0374*** -0.0347*** -0.0379***
(-4.38) (-4.76) (-4.41) (-4.79)
dhwy2 -0.00991 -0.0116 -0.0108 -0.0126
(-1.47) (-1.72) (-1.59) (-1.85)
lnpopden -0.0256** -0.0235** -0.0429*** -0.0395***
(-3.10) (-2.83) (-5.07) (-4.65)
white 0.00613*** 0.00574*** 0.00695*** 0.00655***
(19.99) (19.08) (21.81) (20.98)

30
Dummy variables for the unified school districts of 2000 (00 ) are included in the regression, but the estimated
coefficients are not reported here.

183
per_lessedu -0.00336***-0.00346*** -0.00422*** -0.00423***
(-8.92) (-9.16) (-10.78) (-10.71)
lnempden 0.0282*** 0.0296*** 0.0259*** 0.0275***
(4.55) (4.76) (4.01) (4.24)
lnempden71 0.128*** 0.133*** 0.129*** 0.136***
(5.21) (5.38) (5.08) (5.32)
per_sfam 0.000355 0.00036 0.00101** 0.000950**
(1.16) (1.18) (3.29) (3.08)
per_mfam 0.00567*** 0.00573*** 0.00480*** 0.00492***
(9.61) (9.70) (7.91) (8.07)
per_office -0.00421 -0.00498 -0.00146 -0.0022
(-1.58) (-1.86) (-0.53) (-0.79)
per_com -0.00180* -0.00156 -0.00316*** -0.00294**
(-1.97) (-1.71) (-3.34) (-3.10)
per_indu -0.00206***-0.00222*** -0.00222***-0.00240***
(-3.85) (-4.14) (-4.01) (-4.30)
park -0.00579 -0.00384 -0.00197 -0.000824
(-0.67) (-0.44) (-0.22) (-0.09)
lnFlowden_AM 0.00162 -0.00172 0.00334 -0.000572
(0.33) (-0.34) (0.65) (-0.11)
Accessibility
AccE_AM 0.0008*** 0.000779***
(7.77) (5.48)
AccE_NT 0.000457*** 0.000449***
(8.57) (6.16)
_cons 7.504*** 7.510*** 7.529*** 7.532***
(85.59) (87.89) (86.22) (90.94)
Random effects
level 3
Var(AccE) 4.28E-08 8.97E-08 1.07E-08 2.27E-08
S.E. 2.99E-08 5.99E-08 7.79E-09 1.54E-08
Var(_cons) 0.02 0.02 0.02 0.02
S.E. 0.01 0.01 0.01 0.01
Cov(AccE,_cons)

-1.98E-05 -4.11E-05 -8.34E-06 -1.80E-05
S.E. 1.76E-05 2.80E-05 8.26E-06 1.26E-05
level 2
Var (_cons) τ
0.02 0.03 0.02 0.03
S.E. 0.0009 0.001 0.001 0.001
level 1
Var (Res) σ
0.05 0.05 0.05 0.05
S.E. 0.0004 0.0004 0.0004 0.0004
N 40333 40333 40333 40333

184
Group size(level
3)
5 5 5 5
Group size(level
2)
1909 1909 1909 1909
ll -395.00 -403.1 -464.80 -473.80
aic 946 962.2 1085.5 1103.5
chi2 85236.2 85143.4 85871.9 85753.6
t statistics in parentheses;
Bold and Italic p<0.1, * p<0.05, **p<0.01, *** p<0.001

Figures 4-1 and 4-2 illustrate the variation around the average price effect of accessibility (v

)
across different income clusters defined in two ways. The interesting story here is that regardless of the
definition of income groups, accessibility advantage is most valued in middle income neighborhoods and
least valued by the lowest and the highest income neighborhoods. For example, every unit improvement
in accessibility based on congested time adds a 0.12 percent premium in housing prices in neighborhoods
of the third quintile of poverty rate (middle-income neighborhood), while the premium reduces to 0.05
percent in neighborhoods belong to the first quintile and the fifth quintile of the poverty rate, which
correspond to the highest income neighborhoods and lowest income neighborhoods, respectively. The gap
in the highest and lowest price effects is more than 2.7 times. For the other two types of neighborhoods,
the lower-middle-income neighborhoods (corresponding to the second quintile of the poverty rate or the
fourth quintile of median household income) usually have a price effect of accessibility slightly higher
than the average value, while the upper-middle income neighborhoods have a price effect of accessibility
very close to the overall effect.

185

Figure 4-2 Random component of estimated coefficients for accessibility (
), income
clusters defined by quintiles of median household income ( )

Figure 4-3 Random component of estimated coefficients for accessibility (
), income
clusters defined by quintiles of poverty rate (_ )
12 345
V01k_AM ‐0.0003 0.0002 0.0002 0.0000 ‐0.0002
V01k_NT ‐0.0001 0.0001 0.0001 0.0000 ‐0.0001
‐0.0004
‐0.0003
‐0.0002
‐0.0001
0.0000
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
Random components of coefficients for accessibility effects
5th 4th 3rd 2nd 1st
V01k_AM ‐0.0003 0.0001 0.0005 0.0000 ‐0.0003
V01k_NT ‐0.0002 0.0001 0.0002 0.0000 ‐0.0001
‐0.0004
‐0.0003
‐0.0002
‐0.0001
0.0000
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
Random components of coefficients for accessibility effects

186
The possible explanation for the smaller price effects of accessibility in low income
neighborhoods is that many of those neighborhoods are located at central locations that are already
endowed with a "natural advantage" of accessibility. This is consistent with what we learned from the
maps (see Figures 4-5 to 4-6), that the most accessible areas overlap greatly with those areas having the
highest poverty rate or the lowest median household income, especially in the central location. An
alternative explanation would be that lower income groups value access to jobs less because their choice
sets are much narrower compared with other income groups, because of housing market discrimination,
and that they are more likely to be constrained to those places having the oldest and the cheapest housing
stock, which is more likely to be concentrated in the central location.
Different from lower income neighborhoods, the highest-income neighborhoods do not
necessarily overlap well with the least accessible areas. Thus, the smaller price effects of accessibility
might suggest that compared with middle-income households, highest-income households place less
value on job access. This might be explained by the more flexible work schedule the richest households
have and the lower share commuting time takes up in those households' time budgets.
Moreover, the estimated coefficients for the accessibility measures based on free-flow travel time
also vary across neighborhoods of different income groups, and their pattern of variation follows a similar
trend. The ratio of estimated coefficients for the two accessibility measures is very similar across
neighborhoods of different income clusters. This suggests that all income groups are more responsive to
accessibility under congested conditions. Given the differential effects of accessibility to jobs, it may be
inferred that commuting costs and congestion costs are valued most by the middle-income group and least
by the lowest and the highest income groups.
6 Conclusions
This study examines the effects of traffic congestion on single-family housing prices by looking
at the extent to which the congestion affects commuting costs and constrains accessibility to jobs, using

187
Los Angeles County as a case study. The study may also add to the understanding of how the effects of
congestion and accessibility on housing price vary over neighborhoods of different income clusters. To
identify congestion costs, accessibility measures based on AM peak and off-peak travel time are
constructed, and their implicit prices estimated separately from two regressions are compared. A
hierarchical linear model is used to control for the similarities between housing units within the same
neighborhood. Specifically, a two-level random intercept model is used to account for the spatial variation,
and the results indicate that there exist high levels of within-neighborhood correlation, so that the multi-
level modeling generates more accurate estimates of the influence of accessibility. A three-level random
slope model is also used to capture the spatial variability of accessibility effects across neighborhoods of
different income levels. Several conclusions are drawn from the preliminary results.
1) Consistent with what previous studies find, accessibility to employment has small but
significant positive impacts on housing prices. A 1 percent increase in accessibility based on congested
travel time predicts a 0.08 percent increase in sale price, while a 1 percent increase in accessibility based
on free-flow travel time predicts a 0.04 percent increase in sale price. Given that the accessibility measure
based on congested travel time performs significantly better in terms of predicting housing prices, while
accessibility measured by free-flow travel time produces downward biased estimates, it could be inferred
that traffic congestion plays a significant role in constraining the accessibility premium in housing prices.
2) The effects of accessibility to employment on single-family housing prices vary across income
clusters. Contrary to the priori expectation, I find that accessibility to employment is most valued by
home buyers in middle income neighborhoods and is least valued by those in the lowest- or the highest-
income neighborhoods. The results suggest that low income households are more likely to locate in the
most accessible locations, such as the central area of Los Angeles County, while high income households
may place less value on access to jobs in their residential decision-making.

188
3) Localized congestion externality, measured by the density of traffic throughputs during the
AM peak periods, does not have significant effects on housing price once accessibility to employment is
controlled.
4) Housing structural characteristics are strong predictors of single-family housing price. The
inclusion of the four structural variables adds greatly to the explanatory power of the hedonic price model.
5) Considering other neighborhood variables, most of them have significant expected effects on
housing prices. Consistent with the results of previous studies, proximity to the coast is a strong predictor
of housing price, while proximity to highway ramps within 0.5 miles reduces housing prices. The results
also show that housing prices are higher in those neighborhoods with lower population density, a larger
percentage of white population, a higher level of median household income, and a lower percentage of a
less educated population prone to crimes. Locations with a higher local employment density and a higher
concentration of arts and entertainment-related activities also have higher housing prices. In terms of land
use characteristics, the results also show that multi-family residential land use seems to be compatible
with single-family residential land use, as reflected by the positive marginal price of the percentage of
land area developed into this use, while a higher percentage of commercial and industrial land use
decreases housing prices significantly.
In sum, this study adds to the empirical evidence that congestion-induced travel time delays are
valued by households in their location choices, as reflected in single-family housing prices. It also
indicates that congestion and accessibility effects vary across different income clusters. There are several
more issues that need to be further examined. For example, the accessibility to employment measure
could be further disaggregated by economic sectors and employment types to explore how different types
of jobs are valued differently by households (e.g. Franklin and Waddell 2003; Giuliano et al. 2010).
Moreover, instead of using neighborhood-level information to define income clusters, alternative ways
could be used examine the differential effect of accessibility by income. Also, alternative approaches such

189
as the nonparametric modeling techniques (e.g. McMillen and Redfearn 2010; Redfearn 2009) could be
used to deal with the spatial correlation issue and to examine heterogeneity effects of accessibility and
congestion. Hence, the next step of this research would be to conduct these tests.
Chapter 4 References
Adair, Alastair, Stanley McGreal, Austin Smyth, James Cooper, and Tim Ryley. 2000. "House Prices and
Accessibility: The Testing of Relationships within the Belfast Urban Area." Housing Studies 15 (5):699–
716.
Allison, PAUL D. 1999. "Comparing Logit and Probit Coefficient across Groups." Sociological Methods &
Research 28(2):186–208.
Alonso, William. 1964. Location and Land Use: Towards a General Theory of Land Rent. Harvard University Press.
Anas, Alex and Akin C. Buyukeren. 2013. "A Simple Model of Road Congestion with Job Sprawl." Paper presented
at the 60th annual North American meeting of the North American Regional Science Council, Atlanta,
November 13–16.
Anas, Alex, and Ikki Kim. 1996. "General Equilibrium Models of Polycentric Urban Land Use with Endogenous
Congestion and Job Agglomeration." Journal of Urban Economics 40 (2):232–256.
Anas, Alex, and Yu Liu. 2007. "A Regional Economy, Land Use, and Transportation Model (RELU‐TRAN©):
Formulation, Algorithm Design, and Testing." Journal of Regional Science 47 (3):415–455.
Anderson, Robert J., and Thomas D. Crocker. 1971. "Air Pollution and Property Values." Urban Studies 8 (3):171–
180.
Bayer, Patrick, Robert McMillan, and Kim Rueben. 2004. "An Equilibrium Model of Sorting in an Urban Housing
Market." National Bureau of Economic Research Working Paper No. 10865.
Bender, Bruce. 1981. "Urban Housing Density and the Price of Housing Services." Journal of Urban Economics 9
(1):80–84.
Brigham, Eugene F. 1965. "The Determinants of Residential Land Values." Land Economics 41 (4):325–334.
Brueckner, Jan K. 1987. "The Structure of Urban Equilibria: A Unified Treatment of the Muth-Mills Model.
Handbook of Regional and Urban Economics 2: 821–845.
Brueckner, Jan K., Jacques-Francois Thisse, and Yves Zenou. 1999. "Why is Central Paris Rich and Downtown
Detroit Poor? An Amenity-based Theory." European Economic Review 43 (1):91–107.
Burnell, James D. 1985. "Industrial Land Use, Externalities, and Residential Location." Urban Studies 22 (5):399–
408.
Coulson, N. Edward. 1991. "Really Useful Tests of the Monocentric Model." Land Economics 67 (3):299–307.
Coulson, N. Edward, and Robert F. Engle. 1987. "Transportation Costs and the Rent Gradient." Journal of Urban
Economics 21 (3):287–297.
Figlio, David N, and Maurice E Lucas. 2004. "What’s in a Grade? School Report Cards and the Housing Market."
American Economic Review 94(3):591–604.
Franklin, Joel P, and Paul Waddell. 2003. "A Hedonic Regression of Home Prices in King County, Washington,
Using Activity-specific Accessibility Measures." Paper presented at the 82nd annual meeting of the
Transportation Research Board, Washington, D.C., January 12–16.
Freeman, A. Myrick, III. 1979. "Hedonic Prices, Property Values and Measuring Environmental Benefits: A Survey
of the Issues." The Scandinavian Journal of Economics 81 (2):154–173.

190
Giuliano, Genevieve, Peter Gordon, Qisheng Pan, and JiYoung Park. 2010. "Accessibility and Residential Land
Values: Some Tests with New Measures." Urban Studies 47 (14):3103–3130.
Giuliano, Genevieve, Chris Redfearn, Ajay Agarwal, and Sylvia He. 2012. Network Accessibility and Employment
Centers. Urban Studies 49(1): 77–95.
Glaeser, Edward L., Matthew E. Kahn, and Jordan Rappaport. 2008. "Why do the Poor Live in Cities? The Role of
Public Transportation." Journal of Urban Economics 63 (1):1–24.
Graham, Daniel J. 2007. "Variable Returns to Agglomeration and the Effect of Road Traffic Congestion." Journal of
Urban Economics 62 (1):103–120.
Habib, Muhammad Ahsanul, and Eric J Miller. 2008. "Influence of Transportation Access and Market Dynamics on
Property Values: Multilevel Spatiotemporal Models of Housing Price." Transportation Research Record:
Journal of the Transportation Research Board 2076 (1):183–191.
Hamilton, Bruce W. 1989. "Wasteful Commuting Again." The Journal of Political Economy: 1497–1504.
Hamilton, Bruce W., and Ailsa Röell. 1982. "Wasteful Commuting." The Journal of Political Economy 90 (5):1035–
1053.
Handy, S. L., and D. A. Niemeier. 1997. "Measuring Accessibility: An Exploration of Issues and Alternatives."
Environment and Planning A 29 (7):1175–1194.
Hansen, Walter G. 1959. "How Accessibility Shapes Land Use." Journal of the American Institute of Planners 25
(2):73–76.
Heikkila, E., P. Gordon, J. I. Kim, R. B. Peiser, H. W. Richardson, and D. Dale-Johnson. 1989. "What Happened to
the CBD-distance Gradient? Land Values in a Policentric City." Environment and Planning A 21 (2):221–
232.
Hughes, William T., and C. F. Sirmans. 1992. "Traffic Externalities and Single-family House Prices." Journal of
Regional Science 32 (4):487–500.
Jon, P. Nelson. 1977. "Accessibility and the Value of Time in Commuting." Southern Economic Journal 43
(3):1321–1329.
Jones, Kelvyn. 1991. "Specifying and Estimating Multi-level Models for Geographical Research." Transactions of
the Institute of British Geographers: 148–159.
Jones, Kelvyn, and Nina Bullen. 1993. "A Multi-level Analysis of the Variations in Domestic Property Prices:
Southern England, 1980–87." Urban Studies 30 (8):1409–1426.
Kawamura, Kazuya, and Shruti Mahajan. 2005. "Hedonic Analysis of Impacts of Traffic Volumes on Property
Values." Transportation Research Record: Journal of the Transportation Research Board 1924 (1):69–75.
Kulish, Mariano, Anthony Richards, and Christian Gillitzer. 2012. "Urban Structure and Housing Prices: Some
Evidence from Australian Cities." Economic Record 88 (282):303–322.
Langer, Ashley, and Clifford Winston. 2008. "Toward a Comprehensive Assessment of Road Pricing Accounting
for Land Use." Brookings-Wharton Papers on Urban Affairs: 127–175.
Lave, Lester B. 1970. "Congestion and Urban Location." Papers in Regional Science 25 (1):133–150.
LeRoy, Stephen F, and Jon Sonstelie. 1983. "Paradise Lost and Regained: Transportation Innovation, Income, and
Residential Location." Journal of Urban Economics 13 (1):67–89.
Li, Mingche M, and H. James Brown. 1980. "Micro-neighborhood Externalities and Hedonic Housing Prices." Land
Economics: 125–141.
Lomax, T, S. Turner, G. Shunk, H. S. Levinson, R. H. Pratt, P. N. Bay, and G. B. Douglas. 1997. "Quantifying
Congestion." Vol. 1: Final Report. In National Cooperative Highway Research Program (NCHRP) Report
398. Washington, D.C.: Transportation Research Board; Texas Transportation Institute.
McMillen, Daniel P, and Christian L Redfearn. 2010. "Estimation and Hypothesis Testing for Nonparametric
Hedonic House Price Functions." Journal of Regional Science 50 (3):712–733.

191
Mondschein, A, Brian D. Taylor, and Stephen Brumbaugh. 2011. "Congestion and Accessibility: What's the
Relationship? [online]." Los Angeles: University of California, University of California Transportation
Center (UCTC), [cited 16 September 2013]. Available from:
https://escholarship.org/uc/item/6bh2n9wx.pdf
Moulton, Brent R. 1990. "An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro
Units." The Review of Economics and Statistics 72 (2):334–338.
Muth, Richard. 1969. Cities and Housing: The Spatial Patterns of Urban Residential Land Use. University of
Chicago, Chicago.
Nelson, Jon P. 1977. "Accessibility and the Value of Time in Commuting." Southern Economic Journal: 1321–1329.
Osland, Liv, and Gwilym Pryce. 2012. "Housing Prices and Multiple Employment Nodes: Is the Relationship
Nonmonotonic?" Housing Studies 27 (8):1182–1208.
Ottensmann, John R, Seth Payton, and Joyce Man. 2008. "Urban Location and Housing Prices within a Hedonic
Model." Journal of Regional Analysis and Policy 38 (1):19–35.
Redfearn, Christian L. 2009. "How Informative are Average Effects? Hedonic Regression and Amenity
Capitalization in Complex Urban Housing Markets." Regional Science and Urban Economics 39 (3):297–
306.
Richardson, Harry W. 1977. "On the Possibility of Positive Rent Gradients." Journal of Urban Economics 4 (1):60–
68.
Rosen, Sherwin. 1974. "Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition." The
Journal of Political Economy 82 (1): 34–55.
Schill, Michael H, and Susan M Wachter. 1995. "Housing Market Constraints and Spatial Stratification by Income
and Race." Housing Policy Debate 6 (1):141–167.
Segal, David, and Thomas L Steinmeier. 1982. "Location in a Congested City." Regional Science and Urban
Economics 12 (3):405–423.
Small, Kenneth A., and Shunfeng Song. 1992. “‘Wasteful’ Commuting: A Resolution." The Journal of Political
Economy 100 (4):888–898.
Snijders, Tom, and Roel Bosker. 2012. Multilevel Modeling: An Introduction to Basic and Advanced Multilevel
Modeling., 2nd ed. Los Angeles: Sage.
Solow, Robert M. 1972. "Congestion, Density and the Use of Land in Transportation." The Swedish Journal of
Economics 74 (1):161–173.
Solow, Robert M. 1973. "Congestion Cost and the Use of Land for Streets." The Bell Journal of Economics and
Management Science 4 (2):602–618.
Song, Shunfeng. 1996. "Some Tests of Alternative Accessibility Measures: A Population Density Approach." Land
Economics: 474–482.
Srour, Issam M., Kara M. Kockelman, and Travis P. Dunn. 2002. "Accessibility Indices: Connection to Residential
Land Prices and Location Choices." Transportation Research Record: Journal of the Transportation
Research Board 1805 (1):25–34.
Steiner, Ruth L., Hyungchul Chuang, and Jeongseob Kim. 2012. "A Tale of Two Cities: Impacts of Accessibility
and Congestion on Property Values in Jacksonville MSA and Orlando MSA." Paper presented at the 53rd
annual meeting of the Association of Collegiate Schools of Planning, Cincinnati, November 1–4.
Straszheim, Mahlon. 1974. "Hedonic Estimation of Housing Market Prices: A Further Comment." The Review of
Economics and Statistics: 404–406.
Sullivan, Arthur M. 1983. "The General Equilibrium Effects of Congestion Externalities." Journal of Urban
Economics 14 (1):80–104.

192
Vickerman, Roger, Klaus Spiekermann, and Michael Wegener. 1999. "Accessibility and Economic Development in
Europe." Regional Studies 33 (1):1–15.
Vickerman, Roger W. 1974. "Accessibility, Attraction, and Potential: A Review of Some Concepts and their Use in
Determining Mobility." Environment and Planning A 6 (6):675–691.
Voith, Richard. 1991. "Transportation, Sorting and House Values." Real Estate Economics 19 (2):117–137.
Wabe, J Stuart. 1971. "A Study of House Prices as a Means of Establishing the Value of Journey Time, the Rate of
Time Preference and the Valuation of Some Aspects of Environment in the London Metropolitan Region."
Applied Economics 3 (4):247–255.
Waddell, Paul, Brian J. L. Berry, and Irving Hoch. 1993. "Residential Property Values in a Multinodal Urban Area:
New Evidence on the Implicit Price of Location." The Journal of Real Estate Finance and Economics 7
(2):117–141.
Wheaton, William C. 1998. "Land Use and Density in Cities with Congestion." Journal of Urban Economics 43
(2):258–272.
Wheaton, William C. 2004. "Commuting, Congestion, and Employment Dispersal in Cities with Mixed Land Use."
Journal of Urban Economics 55 (3):417–438.
White, Michelle J. 1976. "Firm Suburbanization and Urban Subcenters." Journal of Urban Economics 3 (4):323–
343.
Yinger, John. 1992. "City and Suburb: Urban Models with More than One Employment Center." Journal of Urban
Economics 31 (2):181–205.

193
Appendix for Chapter 4
Table 4-a1 Descriptive statistics of housing price and structural characteristics
Variable
Transaction
year
N Mean Std. Dev.Median Min Max
saleprice($) 2001-2005 40,333 463,038 550,858 361,000 1,000 5.50E+07
2001 7645 312,778 287,415 236,000 1,000 1.02E+07
2002 8053 377,986 358,396 285,000 1,000 1.27E+07
2003 8341 440,129 400,570 342,000 5,000 1.35E+07
2004 8297 536,580 744,744 421,000 15,909 5.50E+07
2005 7997 639,924 716,328 500,000 3,500 3.60E+07
buildingsf 2001-2005 40,333 1750.6 869.38 1522 360 9756
2001 7645 1721.3 823.2 1514 360 9276
2002 8053 1778.2 880 1544 396 9151
2003 8341 1764 881.1 1531 412 9542
2004 8297 1745.8 879.6 1514 384 9756
2005 7997 1741.7 877.7 1508 384 9331
landsf 2001-2005 40,333 9510.4 15662.1 6754 619.7 493947.1
2001 7645 9375.4 15177.3 6754.5 1081.8 427755.2
2002 8053 9454.2 13752.1 6791 976 422123.2
2003 8341 9490.7 14877.2 6759 1085 434609
2004 8297 10054.3 19500.6 6756.3 619.7 493947.1
2005 7997 9152.3 14170.5 6724.4 1251.4 466787
age 2001-2005 40,333 41.42 21.69 45 0 115
2001 7645 40.3 20.7 43 0 106
2002 8053 40.8 21.3 44 0 115
2003 8341 41.5 21.7 45 0 113
2004 8297 42.1 21.9 46 0 114
2005 7997 42.4 22.6 46 0 109
ownocc 2001-2005 40,333 0.89 0.31 1 0 1
2001 7645 0.91 0.29 1 0 1
2002 8053 0.91 0.29 1 0 1
2003 8341 0.9 0.3 1 0 1
2004 8297 0.88 0.33 1 0 1
2005 7997 0.87 0.34 1 0 1
pool 2001-2005 40,333 0.17 0.37 0 0 1
2001 7645 0.17 0.38 0 0 1
2002 8053 0.18 0.38 0 0 1
2003 8341 0.17 0.38 0 0 1
2004 8297 0.16 0.37 0 0 1
2005 7997 0.16 0.36 0 0 1
fireplace 2001-2005 40,333 0.5 0.5 0 0 1
2001 7645 0.51 0.5 1 0 1
2002 8053 0.52 0.5 1 0 1
2003 8341 0.5 0.5 1 0 1
2004 8297 0.48 0.5 0 0 1
2005 7997 0.46 0.5 0 0 1

194
Table 4-a2 Descriptive statistics of locational attributes and data sources (n=1909)
Source Variable Description Mean
Std.
Dev.
Median Min Max
Expected
Sign
coast Distance to the coast (mi) 17.98 13.26 14.73 0.03 62.65 (-)
dhwyramp Distance to the nearest highway ramp (mi) 1.41 1.74 0.95 0 21.24 (-)
2000 U.S. Census
popden Population density (population per acre) 12.5 9.04 11.28 0 109.79 (+/-)
per_white Percentage of white population (%) 59.03 21.85 60.46 1.55 97.62 (+)
medhhinc median household income ($,000) 56.45 24.8 50.77 8.13 200.001 (+)
per_lessedu
Percentage of population within 18 and 24
who have less than high school education (%)
26.95 14.61 25.82 0 100 (-)
NETS (2000)
empden Employment density (jobs per acre) 3.87 5.95 2.16 2.47E-04 161.47 (+/-)
empden71
Employment density in the arts and
entertainment sector (jobs per acre)
0.08 0.32 0.02 0 14.61 (+)
SCAG land use
dataset (2001)
per_sfam
Percentage of land developed into single-
family residential use
39.59 19.06 43.01 0 80.05 (+)
per_mfam
Percentage of land developed into multi-
family residential use
4.24 6.27 1.97 0 67.42 (+)
per_office Percentage of land developed into office use 0.47 1.46 0 0 29.98 (-)
per_comm
Percentage of land developed into commercial
use
4.77 5.33 3.23 0 58.34 (-)
per_indu
Percentage of land developed into industrial
use
3.22 8.32 0 0 74.96 (-)
park Presence of parks within a census tract 0.5 0.5 1 0 1 (+)
SCAG's RTP
(2003)/NETS
Flowden_AM
Traffic flow density in AM peak hours
(vehicles per acre)
201.43 241.21 125.46 0.06 2781.52 (-)
AccE_AM Access to job in AM peak hour (,000 jobs) 381.77 179.72 408.83 6.49 858.96 (+)
AccE_NT Access to job in Night hours (,000 jobs) 654.78 312.54 699.63 10.55 1305.75 (+)

195

Figure 4-a1 Spatial distribution of Accessibility to employment (AM peak)

196

Figure 4-a2 Spatial distribution of Accessibility to employment (Night period)

197

Figure 4-a3 Spatial distribution of population density

198

Figure 4-a4 Spatial distribution of median household income

199

Figure 4-a5 Spatial distribution of poverty rate

200
CHAPTER 5 CONCLUSIONS

My dissertation seeks to add to the study on the impacts of traffic congestion on the urban
economy. Based on urban economic and agglomeration theories, my two empirical analyses focus on the
micro-economic responses of individual firms and households to congestion costs. Complementary to the
existing theoretical and empirical literature on congestion's economic impacts, my findings provide new
evidence on the trade-off between congestion costs and agglomeration benefits/accessibility advantages at
the intra-metropolitan scale. Since the major findings of each empirical analysis are discussed in each
chapter, I summarize in this chapter the results of the two empirical analyses and compare my results with
current evidence discussed in chapter 2.
1 Summary of findings
Similar to the findings of previous studies, my study finds that traffic congestion negatively
impacts the urban economy at the sub-metropolitan level by inducing travel time delays and altering the
accessibility pattern within them. In the second essay (in chapter 3), the effects of traffic congestion are
examined from the perspective of firms' location choices among employment centers within the Los
Angeles region. Consistent with the existing evidence that firms experience metro-wide agglomeration
benefits, such as regional access to suppliers, customers and workers, as well as those benefits that are
mainly effective within local economic clusters (e.g. employment centers), traffic congestion is also
hypothesized to reduce agglomeration benefits at the regional scale and local scale. I find that the spatial
pattern of travel delays to other locations within the region does not totally align with the regional
accessibility pattern, nor do the most congested locations always accommodate large and/or dense
employment centers. At the local scale, I also find that the level of congestion is only moderately
correlated with the employment size or density of centers, which is not totally consistent with the
predictions in urban economic theories that larger agglomerations are associated with higher costs of
congestion. These results imply that costs of traffic congestion are not perfectly associated with

201
agglomeration benefits or accessibility advantages at either the regional scale or local scale and might be
caused by other exogenous factors. However, the good news is that the negative effects of traffic
congestion can be separately evaluated from the potential benefits of agglomeration. In other words, there
exist employment centers with different bundles of agglomeration benefits and congestion costs so that
individual firms are able to evaluate the two effects separately in their location choices.
The results of discrete choice analysis demonstrate that there is some evidence that congestion
delays at the regional scale reduce the probability of centers being chosen by new firms in most of sectors,
which is consistent with the prior expectation that regional congestion delays diminish the metro-wide
agglomeration benefits experienced by firms. On the other hand, congestion delays within centers do not
reduce employment centers' attractiveness to new firms until reaching a threshold. This is consistent with
the results of inter-regional comparison of congestion costs in current studies that traffic congestion is "a
drag on otherwise high levels of economic growth" (Mondschein, Taylor, and Brumbaugh 2011, 17). My
results also show that the estimated congestion thresholds are higher for firms in those office-related
activities that may place more value on face-to-face interactions and agglomeration economies.
Moreover, my results prove that while the employment size of centers always increases the
likelihood of centers being chosen by firms, the employment density of centers plays a negative role in
the locations of firms in most sectors. This result complies with the findings of previous studies that
higher employment density at more localized scales seems to drive away business due to the possible
higher level of land use intensity and competition. My results also imply that localization effects are more
likely to be effective within centers rather than diffusing throughout a metropolitan area. This reflects an
earlier statement that different types of agglomeration economies may operate at different geographic
scales.
In the third essay (in chapter 4), the effects of traffic congestion are examined through the lens of
household location, as reflected by the single-family housing prices. Consistent with previous theoretical

202
and empirical studies that accessibility to employment is a key location determinant of urban residential
structure, the effect of traffic congestion is identified through its constraints on the accessibility premiums
of locations within the metropolitan area, which also reflects the metro-wide agglomeration benefits
experienced by individual households. Following Graham (2007), I estimate and compare the marginal
price of the two accessibility measures—one based on peak-hour congested travel time and the other
based on off-peak (free-flow) travel time. The results of hedonic price analysis demonstrate that
accessibility to employment do have a small but significant effect on housing prices, and the accessibility
measure based on congested travel time yields higher marginal price. This result is consistent with
previous findings and implies that households value job access and are more responsive to congested
commuting costs.
I also examine whether the price effects of peak-hour accessibility vary across neighborhoods of
different income levels. Contrary to the theoretical prediction that higher income groups would place
more value on accessibility as their value of time is higher, my results indicate that accessibility to
employment is mostly valued by home buyers in middle-income neighborhoods and least valued by those
in the lowest and the highest income neighborhoods. These results imply that low-income households are
more likely to be constrained to those places that have a natural advantage of accessibility to jobs in the
region and preserve most of the oldest housing stocks affordable to the poor, such as the central area of
Los Angeles County. On the other hand, high-income households are less willing to pay for those more
job-accessible places as they may place less value on commuting costs and delays in their time budgets.
In sum, my dissertation provides new empirical evidence on the economic impacts of traffic
congestion at the sub-metropolitan level. The empirics from my study show that congestion costs impact
the intra-metropolitan location behavior of firms and households by increasing their travel time and costs
and reducing the otherwise high levels of agglomeration economies and accessibility advantages they
would experience within the metropolitan area. Moreover, congestion effects vary across firms of
different industrial sectors and households of different income groups.

203
2 Policy implications
My results could have some implications for urban planning. First, my study indicates that the
costs of traffic congestion are balanced with agglomeration benefits and accessibility advantages within a
metropolitan area. This confirms that policies aiming at reducing the road network congestion should not
bring negative impacts on the economic concentration and accessibility levels of those locations
connected by those congested routes. For example, since firms concentrated in employment centers
benefits from better access to other business despite the higher levels of congestion delays, it might be a
mistake to reduce local congestion by reducing the agglomeration levels of the centers and diffusing jobs
to other less congested areas.
Moreover, my results also show that it is important to identify the geographic extent at which
congestion mitigation efforts might be effective. On the one hand, my study proves that metro-wide
congestion delays, especially those associated with commuting trips, negatively affect the location of
firms and households. Thus, my results supports congestion mitigation policies that aim at improving
metro-wide access to labor forces (for firms) and other employment activities (for firms and resident
workers) would be cost effective. For example, transportation investments aiming at improving the
connections among major employment centers and/or alleviating congestion on major commuting
corridors in a region would facilitate the interactions among firms and between firms and residents, thus
enhancing the metro-wide agglomeration benefits. On the other hand, the empirical results also indicate
that localized congestion delays would only negatively affect the location of firms in production-related
activities, while localized congestion effects in physical terms have no significant impact on housing
prices when regional accessibility is controlled. These results suggest that policies aiming at reducing
congestion at more localized scale, such as those reducing local traffic generations, would grant little
benefits for firms and households.
Finally, my study also indicates that firms of different economic sectors and households of
different income groups respond differently to congestion delays. If these results are confirmed in future

204
studies, urban policy makers may find them useful in transportation plans aiming at influencing the
spatial pattern of different groups of residents and economic activities. In other words, policy makers
should pay attention to the differential effects of congestion mitigation policies on different types of firms
and households. For example, my findings support public efforts to improve metro-wide commuting
access to increase the attractiveness of employment centers to office-related activities, especially those in
high-order service sectors. In terms of urban residential structure, my study suggests that it would be more
cost effective to improve commuting access for those places concentrated by middle-income and lower-
middle-income households within a metropolitan area. In other words, transportation investments are
more likely to increase the accessibility premiums and consequently the housing prices in those
neighborhoods.
3 Limitations and future research
My dissertation is just a start of empirical investigations on how traffic congestion impacts the
urban spatial structure. In the next steps, my study can be extended along two aspects discussed below.
1) Congestion costs and other costs (diseconomies) of agglomeration
As discussed, traffic congestion is only one source of agglomeration diseconomies. Other
important sources of agglomeration diseconomies include high land rents and wages, which can also
impact on the location behavior of firms and residents. For example, land rents are important inputs in
firms' production function and households' utility function and can affect the location of business
activities and residents. In the second essay (in chapter 3), land rents are not taken into account because of
data availability. Instead, I use employment density and other locational factors to summarize their
influences. In the third essay (in chapter 4), the effects of congestion costs on residential land values are
examined indirectly by looking at their constraints on accessibility. Future studies could be developed to
fully explore the relationship between congestion costs and industrial/commercial land rents to better
capture the interactions among various types and sources of diseconomies and their influences on the
urban economy.

205
2) Extension to other metro areas
My empirical analyses are all based on the Los Angeles region. Although this region is not an
outlier (Richardson 1995), previous studies found that many U.S. metropolitan areas have long been
characterized by monocentricity (e.g. New York and Boston) or dispersion (e.g. Portland and Philadelphia)
(Lee 2006). Thus, it would be interesting to examine how firms and households respond to congestion
costs in a typical monocentric city or dispersed city and compare the results with my findings based on
the Los Angeles region where the polycentric structure provides sufficient heterogeneity among
alternatives in the location choice set.

206
Chapter 5 Reference
Graham, Daniel J. 2007. "Variable Returns to Agglomeration and the Effect of Road Traffic Congestion." Journal of
Urban Economics 62 (1): 103–120.
Lee, Bumsoo. 2006. Urban spatial structure, commuting, and growth in US metropolitan area. Los Angeles:
University of Southern California. Ph.D. Dissertation
Mondschein, A., Brian D. Taylor, and Stephen Brumbaugh. 2011. "Congestion and Accessibility: What's the
Relationship? [online]." Los Angeles: University of California, University of California Transportation
Center (UCTC), [cited 16 September 2013]. Available from:
https://escholarship.org/uc/item/6bh2n9wx.pdf
Richardson, Harry W. 1995. "Economies and Diseconomies of Agglomeration." In Urban Agglomeration and
Economic Growth, edited by Herbert Giersch, 123–155. Berlin-Heidelberg: Springer.

Abstract (if available)

Abstract While traffic congestion is considered an important source of diseconomies in theories of agglomeration economies, only a few studies measure the effects congestion costs empirically. My study seeks to fill in this gap by directly estimating the economic impact of traffic congestion at the intra-metropolitan scale. Three major questions are addressed in my dissertation: 1) What is our current knowledge on agglomeration economies and congestion costs, in terms of their effective geographic extents, the methods for measuring them, and the methods for unraveling their separate effects? 2) Do congestion-induced travel time delays adversely affect locations' attractiveness to firms when potential benefits of agglomeration are controlled? 3) Does traffic congestion negatively affect single-family housing values by inducing commuting delays and reducing locations' accessibility premiums? ❧ To answer these questions, I review prior theoretical and empirical studies on agglomeration and congestion to summarize the mechanism by which traffic congestion influences the urban economy and urban spatial structure. Using the 2001‐2005 establishment-level data from the National Establishment Time Series (NETS) dataset and the 2001‐2005 disaggregate housing sales data from the DataQuick dataset, I examine the impacts of traffic congestion on intra-urban business and residential geography through the lens of firms’ location choices and households' bid-up prices for housing units, respectively. Specifically, I empirical test whether and to what extent congestion-induced travel time delays are costly to firms and households in their location decisions within a metropolitan area, how these influences differ at different geographic extents, and how these influences vary across firms of different industrial sectors and home buyers of different income groups. Focusing on the Los Angeles region, the results indicate that metro-wide congestion delays negatively influence firm location and housing price, while local congestion delays will not impose a drag on otherwise high levels of agglomeration benefits until reaching a threshold. I also find that firms valuing proximity (or agglomeration benefits), such as those specialized in the information, finance and insurance and services sectors, are more likely to endure congestion costs than those production-related firms and retail firms. Moreover, compared with those in lower-income and upper-income groups, households in middle-income neighborhoods are more responsive to commuting costs and congestion costs. In sum, my study provides new evidence that congestion costs matter to urban spatial structure by altering the accessibility patterns and inhibiting additional agglomeration benefits at the intra-metropolitan scale.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Unraveling decentralization of warehousing and distribution centers: three essays

PDF

Urban spatial transformation and job accessibility: spatial mismatch hypothesis revisited

PDF

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

PDF

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

PDF

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

PDF

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

PDF

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

PDF

Urban air pollution and environmental justice: three essays

PDF

Location of warehouses and environmental justice: Three essays

PDF

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

PDF

The built environment, tour complexity, and active travel

PDF

Active travel, outdoor leisure, and neighborhood environment: path analysis, Los Angeles County

PDF

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

PDF

The flexible workplace: regional tendencies and daily travel implications

PDF

Curating gastronomy: restaurants and social media in the cultural economy

PDF

Healthy mobility: untangling the relationships between the built environment, travel behavior, and environmental health

PDF

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

PDF

Urban universities' campus expansion projects in the 21st century: a case study of the University of Southern Calfornia's "Village at USC" project and its potential economic and social impacts on...

PDF

Three papers on dynamic externalities and regional growth

PDF

Essays on urban and real estate economics

Asset Metadata

Creator Hou, Yuting (author)

Core Title Essays on congestion, agglomeration, and urban spatial structure

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Policy, Planning, and Development

Publication Date 11/17/2014

Defense Date 10/08/2014

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

Tag accessibility,agglomeration economies,firm location choice,housing price,OAI-PMH Harvest,Traffic congestion,urban spatial structure

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Giuliano, Genevieve (committee chair), Redfearn, Christian (committee chair), Boarnet, Marlon (committee member), Moore, James Elliott, II (committee member)

Creator Email houyuting86@hotmail.com,yutingho@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-517269

Unique identifier UC11297561

Identifier etd-HouYuting-3090.pdf (filename),usctheses-c3-517269 (legacy record id)

Legacy Identifier etd-HouYuting-3090.pdf

Dmrecord 517269

Document Type Dissertation

Format application/pdf (imt)

Rights Hou, Yuting

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA