University of Southern California Dissertations and Theses

Unraveling decentralization of warehousing and distribution centers: three essays

(USC Thesis Other)

Unraveling decentralization of warehousing and distribution centers: three essays

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Unraveling Decentralization of Warehousing and Distribution Centers:

Three Essays

by

Sanggyun Kang

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
URBAN PLANNING AND DEVELOPMENT

August 2017

Copyright 2017 Sanggyun Kang
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)

Copyright © 2017 by Sanggyun Kang
All rights reserved. Printed in the United States of America. No part of this dissertation may
be reproduced or transmitted in any manner or by any means whatsoever without written
permission from the author, except in the case of brief quotations embodied in critical
articles or reviews.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)

I dedicate this dissertation to my wife, Jin Joo, and my son, Iejae.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
i

ACKNOWLEDGEMENT
I give thanks to God for leading the way and traveling together over the last ten years of my
life in New York, NY and Los Angeles, CA. I am grateful to Professor Genevieve Giuliano, my
advisor and dissertation committee chair for her support and patience through the Ph.D.
program. I wish to express sincere appreciation for being a mentor with keen insight, who
provided clear directions, critical feedback, and valuable suggestions. I would also like to
express my sincere gratitude to dissertation committee members – Professors Marlon
Boarnet and Petros Ioannou. I am indebted to Professor Sarah Williams at MIT for her help
in resolving critical data problems, which might have hampered my entire doctoral
research. I also thank my colleagues Jack Quan Yuan, Nathan Hutson, and Eun Jin Shin for
their effort to make warehousing datasets operational. My wife, Jin Joo, has always been
the primary source of creativity, energy, joy, and happiness. My son, Iejae, who was born
two months before the dissertation defense, provided the strongest motivation to finish my
dissertation on time. I am enormously grateful to my family, my father, Hee Sung Kang, my
mother, Kyung Mi Yu, my mother-in-law, Myoung Sook Jeong, and especially my father-in-
law in heaven, Soo Kyu Chae, for their endless support, love, and encouragement.

Sanggyun Kang
August 2017

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
ii

ABSTRACT OF THE DISSERTATION
This dissertation examines how and why warehousing and distribution centers have
decentralized from central urban areas to the periphery. The research verifies theoretical
discourses on the factors that explain warehouse location change and generates consistent
and robust empirical evidence with descriptive analysis and estimation of econometric
models. Through three independent yet interrelated empirical studies, this dissertation
investigates how the restructuring of the logistics industry has reshaped the spatial
distribution of warehousing facilities at the sub-metropolitan level. Findings suggest that
freight demand and land prices are two main factors for decentralization. To transport
large volumes of freight, the logistics industry has built large-scale warehousing facilities
on urban outskirts where land is readily available at relatively lower costs. This process of
decentralization involves tradeoffs of logistics costs between land and transport costs.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
iii

TABLE OF CONTENTS
Acknowledgement ................................................................................................................................................. i
Abstract of the Dissertation .............................................................................................................................. ii
List of Tables .......................................................................................................................................................... vi
List of Figures ..................................................................................................................................................... viii
CHAPTER 1 Introduction ................................................................................................................................... 1
1 Dissertation Overview ............................................................................................................................... 1
2 Contributions of this Research ............................................................................................................... 4
3 Organization of the Dissertation ............................................................................................................ 7
CHAPTER 2 Spatial Dynamics of the Logistics Industry: Evidence from California ................... 8
Abstract ..................................................................................................................................................................... 8
1 Introduction ................................................................................................................................................... 8
2 Literature Review ..................................................................................................................................... 10
2.1 Restructuring and Decentralization ........................................................................................ 10
2.2 Empirical Evidence ......................................................................................................................... 12
3 Research Approach .................................................................................................................................. 13
4 Data ................................................................................................................................................................ 15
5 Results ........................................................................................................................................................... 17
5.1 Changes in Warehousing and Distribution ........................................................................... 17
5.2 Decentralization and Concentration........................................................................................ 21
5.2.1 Absolute Decentralization ...................................................................................................... 21
5.2.2 Relative Decentralization ........................................................................................................ 23
5.2.3 Absolute Concentration ........................................................................................................... 24
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
iv

5.2.4 Relative Concentration ............................................................................................................. 25
6 Summary and Conclusions .................................................................................................................... 27
6.1 Explaining Results .......................................................................................................................... 28
6.2 Potential Implications for Truck VMT .................................................................................... 29
CHAPTER 3 Why do warehouses decentralize more in certain metropolitan areas? ............. 32
Abstract .................................................................................................................................................................. 32
1 Introduction ................................................................................................................................................ 32
2 Literature Review ..................................................................................................................................... 34
2.1 Why Should We Care? ................................................................................................................... 34
2.2 Logistics Restructuring and Warehousing Decentralization ......................................... 37
3 Research Approach .................................................................................................................................. 39
3.1 Conceptual Framework – Rationale behind Decentralization ...................................... 39
3.2 Measurement and Data ................................................................................................................. 45
3.2.1 Warehousing Decentralization: Changes in Distribution over Time ..................... 45
3.2.2 Spatial Distribution of Land Prices ...................................................................................... 47
3.2.3 Freight Flow ................................................................................................................................. 49
4 Results ........................................................................................................................................................... 50
4.1 Descriptive Statistics and Hypothesis Testing..................................................................... 50
4.1.1 W&D decentralization in 48 U.S. Metropolitan Areas .................................................. 54
4.1.2 Distribution and Decentralization of Large W&Ds ....................................................... 56
4.1.3 Decentralization and Freight Flow ...................................................................................... 60
4.2 Model Results ................................................................................................................................... 61
5 Conclusions And Future Research ..................................................................................................... 68
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
v

CHAPTER 4 Warehouse Location Choice: A Case Study in Los Angeles, CA ............................... 72
Abstract .................................................................................................................................................................. 72
1 Introduction ................................................................................................................................................ 73
1.1 Warehousing Decentralization .................................................................................................. 74
1.2 Location Determinants for Warehousing Facilities ........................................................... 76
2 Research Approach .................................................................................................................................. 80
2.1 Conceptual Framework ................................................................................................................ 80
2.2 Discrete Choice Model ................................................................................................................... 83
3 Data ................................................................................................................................................................ 86
3.1 Warehouse Data and Sample Distribution ............................................................................ 88
3.2 Choice Set Design and Location Attributes ........................................................................... 92
4 Results ........................................................................................................................................................... 96
5 Conclusions and Discussion ............................................................................................................... 102
CHAPTER 5 Conclusions ................................................................................................................................ 105
Appendix – Chapter 2 Mathematical formulas of spatial measures ............................................. 110
A1 Measure 1. Decentralization .......................................................................................................... 110
A2 Measure 2. Relative Decentralization ........................................................................................ 110
A3 Measure 3. Concentration .............................................................................................................. 110
Appendix – Chapter 3 ..................................................................................................................................... 111
A1 Summary of Hypothesis Tests ...................................................................................................... 111
A2 The number, distribution, and decentralization of W&Ds by metro area ................... 112
BIBLIOGRAPHY ................................................................................................................................................. 116

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
vi

LIST OF TABLES
Table 1 Identification of the research gap and the contribution of this research ....................... 5
Table 2 Four Categories of Spatial Structure Measures ...................................................................... 14
Table 3 Descriptive Statistics of W&D Establishments and Employment, 2003 - 2013 ........ 18
Table 4 Percentage change in absolute decentralization between 2003 and 2013 ................. 23
Table 5 Percentage change in relative decentralization between 2003 and 2013 .................. 24
Table 6 Percentage change in absolute concentration between 2003 and 2013 ..................... 25
Table 7 Definition and data source of variables..................................................................................... 50
Table 8 Large metro areas (N=22) and their population (2000) and employment (2003) . 52
Table 9 Small metro areas (N=26) and their population (2000) and employment (2003) .. 53
Table 10 Summary statistics of decentralization by metro size ...................................................... 54
Table 11 Decentralization by significance and by metro size .......................................................... 56
Table 12 Comparison of the distribution of large and small W&Ds ............................................... 58
Table 13 Summary statistics of decentralization by W&D size ....................................................... 59
Table 14 Decentralization of large W&Ds by significance by metro size ..................................... 60
Table 15 Decentralization by freight flow quartile groups ............................................................... 61
Table 16 Summary statistics of explanatory variables ....................................................................... 63
Table 17 Pairwise correlation of explanatory variables with decentralization ........................ 63
Table 18 Pairwise correlation between the explanatory variables ................................................ 64
Table 19 Results of model estimation: Model 1A and 2A (all warehouses) ................................ 67
Table 20 Results of model estimation: Model 1L and 2L (large warehouses) ........................... 68
Table 21 Distribution and share of W&D entries by built year and size ...................................... 91
Table 22 Summary statistics of location attributes (N=660) ........................................................... 95
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
vii

Table 23 Summary of variable definitions ............................................................................................... 95
Table 24 Estimation results of Models 1-1, 1-2, and 1-3 .................................................................. 100
Table 25 Estimation results of Models 2-1, 2-2, and 2-3 .................................................................. 101

Table A. 1 Testing results of null hypotheses #1-7 ............................................................................. 111
Table A. 2 Number of W&Ds and changes thereof, 2003-2013, Group 1 ................................... 112
Table A. 3 Number of W&Ds and changes thereof, 2003-2013, Group 2 ................................... 113
Table A. 4 Distribution of W&Ds and changes thereof, 2003-2013, Group 1 ........................... 114
Table A. 5 Distribution of W&Ds and changes thereof, 2003-2013, Group 2 ........................... 115

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
viii

LIST OF FIGURES
Figure 1 W&D locations, 2003, 2013, Los Angeles ............................................................................... 19
Figure 2 W&D locations, 2003, 2013, San Francisco ............................................................................ 20
Figure 3 W&D locations, 2003, 2013, Sacramento (left) and San Diego (right) ........................ 20
Figure 4 Average distance to CBD by size category, Los Angeles and San Francisco .............. 22
Figure 5 Share of W&D establishments in total employment density quartiles, 2003 and
2013 ........................................................................................................................................................................ 26
Figure 6 A schematic of the mechanism of W&D location change .................................................. 41
Figure 7 Illustration of two linear models ................................................................................................ 43
Figure 8 Illustration of density gradient and peak density measures ........................................... 49
Figure 9 Scatter plot of W&D decentralization and 2000 population ........................................... 55
Figure 10 Scatter plot of decentralization of large/small W&Ds and 2000 population ......... 57
Figure 11 Scatter plot of W&D decentralization and 2002 freight flow ....................................... 61
Figure 12 Spatial distribution of 5,364 warehouses and trade nodes .......................................... 89
Figure 13 Percent distribution of warehouses from the Los Angeles CBD by built year ....... 91
Figure 14 Percent distribution of warehouses from the Los Angeles CBD by facility size ... 92
Figure 15 Distribution of employment sub-centers and trade nodes (close-up of the region)
................................................................................................................................................................................... 94

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
1

CHAPTER 1
INTRODUCTION

1 DISSERTATION OVERVIEW
Over the last decade, warehousing and distribution centers (W&D) have been relocated
from central urban areas to the periphery (Dablanc and Ross, 2012). This spatial shift is
defined as W&D decentralization, which has been attributed to the restructuring of the
logistics industry to move large volumes of products as quickly, cheaply, and reliably as
possible (Cidell, 2011; Dablanc et al., 2014; Rodrigue, 2008). At the global scale, the system
of goods production and distribution has been geographically dispersed. Simultaneously,
at the sub-metropolitan level, larger and automated W&Ds have been built on the urban
outskirts, where land is cheaper and readily available (Bowen, 2008; Cidell, 2011; Dablanc
and Ross, 2012; Hesse and Rodrigue, 2004; Glaeser and Kohlhase, 2004; Rodrigue, 2008).
Researchers have been interested in this spatial shift not only because it influences
the geography of freight activity but also because it may result in more freight vehicle miles
traveled and associated negative externalities – increased fuel consumption, greenhouse
gas and criteria emissions, congestion, noise, infrastructure damage, and environmental
justice (Allen, Browne & Cherrett, 2012; Bowen, 2008; Cidell, 2010; Dablanc & Ross, 2012;
Dablanc & Rakotonarivo, 2010; Dablanc et al., 2014; Giuliano and Kang, 2016; Heitz and
Dablanc, 2015; Sakai et al., 2015; Van den Heuvel, et al., 2013; Woudsma et al., 2016).
However, the dearth of data on factors, such as warehousing location, zone-level truck flow,
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
2

or freight origin-destination data for US metropolitan areas, has significantly limited
researchers from looking for robust evidence on determining whether the current pattern
of decentralization is a problem worthy of policy intervention.
In that regard, I primarily focus on understanding the mechanism of W&D
decentralization and raise the research question: how and why have W&Ds changed
location over time? I aim to answer the question with three independent yet interrelated
essays, which evaluate the spatial shift with three different methodologies at three
different geographic levels. In the first essay, I define multi-dimensional spatial measures
(centrality or concentration; absolute distribution of W&Ds or relative to the shift of the
general population; warehouse establishment or employment; and large or small
warehouses). The purpose is 1) to assess whether W&D decentralization from 2003 to
2013 has been a consistent trend; 2) to compare the extent of decentralization across four
metropolitan areas in California; and 3) to evaluate whether the spatial shift has been more
pronounced than that of population or employment. I also identify several metro-level
factors that might contribute to the decentralization.
In the second essay, I quantify W&D decentralization over the same period in 48
major US metropolitan areas to examine the current theory that decentralization has been
mainly attributed to large distribution centers built on the urban periphery to transport
large volumes of freight more efficiently. Accordingly, I test several hypotheses to see
whether large warehouses have decentralized to a greater degree than smaller ones; and
whether warehouses have decentralized more in large metro areas (e.g. higher land prices)
than in small ones. I also run several linear models to evaluate whether the variance across
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
3

metro areas in freight activity and land prices explains the variance in decentralization. In
the first and second essays, I use ZIP Code level datasets.
In the third essay, I investigate the transition in warehouse location decisions using
5,364 facilities built over the last six decades in the Greater Los Angeles region to
determine whether the change in W&D facility characteristics from logistics restructuring
has contributed to the change in location decisions over time. I compare the location
decisions of W&Ds built in the recent period to those built in earlier periods. Similarly, I
compare location decisions of very large W&Ds to those of small facilities. I use detailed
W&D information retrieved from an industrial real estate listings company, which provides
a comprehensive view of the current warehousing market. Information on a facility’s
physical characteristics (address, rentable building area, and year of construction) allows a
model structure that separates the effect of built year from that of facility size, as well as
location characteristics (land prices, proximity to local markets, trade nodes, and labor
pool).
Results of the three essays provide robust and consistent empirical evidence on
recent trends in W&D decentralization. First, the extent of W&D decentralization varies
across US metro areas. W&Ds tend to decentralize more in metro areas where freight
demand is heavy, due to a large population or a significant trade gateway function; land
prices are high in central urban areas; and physical geography constrains where
development can occur. Second, decentralization is also attributed to very large W&Ds.
Indeed, the fundamental driver of W&D decentralization is the restructuring of the logistics
industry, which has established large-scale, automated facilities on the urban outskirts,
where land prices are relatively low to process large volumes of freight efficiently. Third,
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
4

over the past several decades in the Los Angeles region, there has been a substantial
change in warehouse location decisions. Warehouses built after 2000 have prioritized
lower land prices over access to local markets, trade nodes, and the labor force. It can also
be inferred that as newer, larger facilities are established on the outskirts farther from
central urban markets, the gains of lower land prices and lower per-unit inventory costs
from economies of scale offset the increase in transport costs. Fourth, in the Los Angeles
region, a considerable share of W&D development over the past decade has occurred in a
relatively small area around San Bernardino and Riverside, and most of the development
consists of very large distribution centers. Given that warehouses are major truck travel
attractors, it is important that the change in the geography of freight movement and
potential negative impacts on the local community and the region are carefully examined.

2 CONTRIBUTIONS OF THIS RESEARCH
W&D decentralization has drawn a fair amount of attention in the fields of urban planning,
transportation planning, and transportation geography. However, most studies have either
theoretically discussed why warehouses would decentralize, or at most quantified the
location change. Currently, there is no literature on empirical tests of the factors that
explain decentralization. The present research not only fills a large research gap but also
proposes several methodologies to test and evaluate factors that explain decentralization.
Table 1 identifies the research gap and the contributions of this research.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
5

Table 1 Identification of the research gap and the contribution of this research
Research Topic Methodology Literature Contribution
Logistics restructuring Business and operations research Robust discussion -
Decentralization of
the W&D industry
Use of multidimensional measures of
spatial distribution
Not available
Essay 1
Comparison of W&D distribution in two
time periods
Several mixed
results
Comparison among multiple metro areas
using ZIP Code level datasets
Limited
Comparison between large and small
warehouses
Limited
Formal testing of the difference in spatial
distribution
Not available
Factors for W&D
decentralization
at the national level
Theoretical foundation of the connection
between logistics restructuring and
decentralization
Robust discussion
Essay 2
Comparison among 48 US metro areas
using ZIP Code level datasets
Not available
Hypothesis testing of the nation-wide
trend of W&D decentralization
Not available
Linear models to evaluate the factors for
decentralization at the national level;
empirical evidence
Not available
Transition in W&D
location decisions and
tradeoffs among
logistics costs
Warehouse location choice factors
Several studies of
mixed methods
Essay 3
Comparison of W&D location choice by
facility size
Limited
Comparison of W&D location choice in
multiple time periods
Not available
Testing of logistics cost tradeoffs Not available

Logistics restructuring has been widely studied in business and operations research.
W&D decentralization, as one of the outcomes of the restructuring processes, has been
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
6

evaluated in multiple metro areas in North American, European, and Asian countries.
However, most studies have used one or two rudimentary measures of spatial distribution
or concentration; have focused on just one or two metro areas; and have not conducted
formal statistical testing of the spatial shifts. So far, only one paper, Dablanc et al. (2014),
has conducted a systematic comparison of the pattern of decentralization between two
metro areas (Los Angeles, CA, and Seattle, WA). Factors for warehouse location decisions
have been evaluated in multiple industrial firm location studies, which include several
studies using discrete choice models or firm surveys and interviews. However, the change
in effects of the choice factors, in conjunction with logistics restructuring and consequential
decentralization, has not been empirically verified. This dissertation aims to address an
important part of the above research gaps.
In Essay 1, the multiple measures describe the complex aspects of spatial shifts and
provide a clearer picture of how spatial shifts differ among the four metro areas. This
difference provides useful information to determine whether W&D decentralization may be
a problem worthy of policy intervention. In addition, with a systematic comparison across
the metro areas, I identify metro-level factors that influence the extent of decentralization.
In Essays 2 and 3, I contribute to substantiating the current theory that explains why W&Ds
have decentralized. In Essay 2, with a sample of 48 US metro areas, for the first time, I
verify two principal factors for decentralization (high freight demand and high land prices)
which have been frequently discussed in theoretical discourses. It is also the first study
which uses ZIP Code level data sets of warehouse location at the national level. Compared
to past studies that have relied on county-level data, the geographic precision in
quantifying the spatial shifts is much improved. In Essay 3, I verify heterogeneous
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
7

preferences in warehouse location choice by facility size and built year. For the first time, I
compare parameters of several discrete choice models of different time periods and
different facility size to obtain robust evidence on the shift in location determinants as well
as the tradeoffs between facility and transport costs. These factors have previously only
been discussed in theoretical discourses.

3 ORGANIZATION OF THE DISSERTATION
This dissertation consists of three independent essays. All three essays have been
submitted to peer-reviewed journals or presented at academic conferences. Each essay has
an independent structure of a research paper; hence introduction and literature review
chapters partially overlap. The dissertation closes with conclusions and a brief discussion
on the implications of my research for planning scholarship and practice.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
8

CHAPTER 2
SPATIAL DYNAMICS OF THE LOGISTICS INDUSTRY: EVIDENCE FROM CALIFORNIA
*This essay is co-authored by Dr. Genevieve Giuliano and Sanggyun Kang

ABSTRACT
Is logistics decentralization a consistent trend across metropolitan areas? If so, is the trend
more pronounced than population or employment decentralization? This paper examines
logistics industry location trends from 2003 to 2013 in the four largest California
metropolitan areas: Los Angeles, San Francisco, San Diego, and Sacramento. We define
measures of both decentralization and de-concentration and compare logistics location
trends with those of population and employment. Decentralization with respect to logistics
establishments and employment is confirmed for Los Angeles; the other metro areas show
mixed results. Possible explanatory factors include metropolitan size, share of non-local
trade, and local geography.

1 INTRODUCTION
The purpose of this research is to examine changes in the spatial pattern of warehousing
and distribution (W&D) activities and consider the implications of these changes on freight
flows. W&D activities may be decentralizing in response to rising land values and scale
economies. Ultimately, we seek to understand whether these spatial shifts result in more
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
9

truck VMT, or whether the efficiencies gained by larger scale operations allow offsetting
savings, such as enabling the use of larger trucks or achieving higher average load factors.
However, there is no good source for tract- or zone-level truck flow or origin-destination
data for US metropolitan areas. We, therefore, focus on changes in the spatial distribution
of W&D activities and discuss potential implications of these changes on truck travel.
One of the most notable trends in metropolitan areas is the rapid growth in
warehousing and distribution activity. In the US, the number of warehousing
establishments increased 15%, and warehousing employment increased 33% between
2003 and 2013. In contrast, total establishments and employment increased by 3% and 4%
respectively. Explanations for this growth include continued globalization, changes in
consumer demand, advances in information, communication, and transportation
technology, just-in-time production, and restructuring of the logistics industry (Hesse and
Rodrigue, 2004; Cidell, 2011).
An understanding of W&D distribution and decentralization trends is a first step in
determining the extent to which decentralization may be a problem worthy of policy
intervention. This paper examines recent trends in W&D location and decentralization in
four metropolitan areas in California. W&Ds are part of goods supply chains. They are
strategically located, considering land and transport costs, access to market, labor, and
major freight nodes. Even if W&Ds are decentralizing, the question is whether they are
decentralizing more than the markets they serve. We use measures of centrality and
concentration and consider the distributions of population, employment, and freight
infrastructure. Our results are mixed. In general, we find decentralization with respect to
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
10

employment, but not with respect to the number of establishments. Only Los Angeles
shows decentralization across all measures.
The remainder of this paper is organized as follows. Section 2 provides a brief
review of the literature. Our research approach is presented in Section 3, and data is
described in Section 4. Section 5 presents results, and the report closes with some
conclusions and suggestions for future research.

2 LITERATURE REVIEW
As production and distribution systems have reorganized, so has their spatial structure.
The reorganization of production and distribution systems is well documented, but their
spatial reorganization is not (Hesse and Rodrigue, 2004). Restructuring has been
attributed to 1) globalized market- and customer-driven goods production systems; 2)
integrated management of information; 3) e-commerce, and consumer preference changes;
4) an increasing share of high value/low weight goods; and 5) increased competition due to
1970s and 1980s deregulation and liberalization in the US, and integration of European
markets in the 1990s (Hesse and Rodrigue, 2004; Castells, 1996; Knowles and Hall, 1998;
Dablanc et al., 2011).

2.1 Restructuring and Decentralization
Restructuring has resulted in geographically fragmented supply chains, which imply
geographically separated locations of suppliers, producers, distributors and consumers
(Rodrigue, 2008). The concurrent spatial reorganization is attributable to pressure for
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
11

economies of scale in goods production and distribution systems. Decreased freight
transport costs and expanded freight transport capacity, due to technology advancement
and infrastructure improvements, have eased spatial reorganization processes (Hall et al.,
2006). These factors have facilitated the emergence of a logistics industry that puts
emphasis on reliability and high throughput of goods transportation, which, rather than
storage, has become the main goal of logistics (Hesse and Rodrigue, 2004). Moreover,
demand for a centralized goods distribution system (e.g. logistics consolidation) increased
significantly (Cidell, 2011; Rodrigue, 2008).
This systematic reorganization of logistics has generated a spatial reorganization of
facility locations, termed the “new distribution economy” (Hesse and Rodrigue, 2004, p.
178). It requires efficient goods distribution chains that have become more and more
sensitive to the spatial configuration of logistics facilities rather than direct transportation
cost itself (Movahedi et al., 2009). Location decisions are based on securing proper access
to international and intercontinental economies (Bowen, 2008).
Metropolitan population is the main driver for location of goods distribution
activities in the conventional model (McKinnon, 1983). The new logistics system selects
physical locations based on real estate costs (Hesse, 2006), access to highways and rail
facilities (Rodrigue, 2008), access to low-skilled and low-wage labor, and reasonable
business costs (Cidell, 2011). In particular, the rebalance on tradeoffs between transport
and inventory costs play a significant role (McKinnon, 2009). Thus, optimal scale becomes
a major factor in location choice (Dablanc and Ross, 2012). In addition, global supply
chains prioritize access to major links in the national or international network (Hesse,
2002). Given the emphasis on scale and velocity, we would expect spatial shifts away from
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
12

the urban core, due to development density, land constraints and arterial congestion
(Hesse and Rodrigue, 2004). Agglomeration economies associated with the urban core are
less valuable given the land requirements of large-scale facilities (Hesse and Rodrigue,
2004). In search of alternative locations, increased distance from the urban core offers
cheaper land, larger parcels, access to congestion-free transportation infrastructure, and a
supporting environment for logistics operations. The result is logistics decentralization
and clustering of freight facilities in large metropolitan areas (Dablanc and Ross, 2012).

2.2 Empirical Evidence
Empirical studies of W&D location are limited. Two aspects of spatial structure changes
have been of particular interest: 1) movement of facilities from the urban core to
peripheral places (decentralization) and 2) clustering of logistics functions (concentration).
An expansion of warehousing activities and associated W&D decentralization have
been documented in three major US metropolitan areas, Atlanta, Los Angeles, and Chicago,
during the 2000s (Dablanc and Ross, 2012; Dablanc et al., 2014; Goodchild and Dubie,
2016). These studies calculate the average distance of each W&D from the geographic
centroid of all W&Ds (centrography). This measures the geographic spread but does not
provide a comparison to population or employment shifts. Suburbanization of W&D
activities was observed in metro areas in Sweden, the UK, and Japan as well (Heitz et al.,
2016; Allen et al., 2012; Sakai et al., 2015). The decline of W&Ds in traditional port cities
has been documented in six Canadian metro areas (Woudsma and Jakubicek, 2016). In
contrast, W&D concentration is observed for Seattle, again using the same centrography
measure (Dablanc et al., 2014). The authors surmise that W&D decentralization may occur
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
13

only in very large metropolitan areas, in which the functions of major trade nodes and
major consumer markets coexist. Cidell (2010) used Gini coefficients and observed
decentralization in US metropolitan areas (CBSA, Core-based Statistical Areas) 1986-2009.
Van den Heuvel et al. (2013), also using the Gini coefficient, but at the establishment level,
observed increased spatial concentration in a province of the Netherlands 1996-2009.
Thus, the empirical evidence on W&D decentralization is mixed.

3 RESEARCH APPROACH
Anas, Arnott, and Small (1998) conceptualize urban spatial structure in two dimensions:
centrality and concentration. Centrality is the degree to which activities are located around
a single center. Urban structure may be centralized (activities located close to the center)
or decentralized (activities located further from the center, but still spatially oriented to the
center). Concentration is the degree to which activities are located within close proximity
to one another and ranges from clustered to dispersed. Concentration can take many forms;
there may be one or a few clusters, or many clusters. The share of activity that is clustered
may also vary. The extreme case of no clustering is dispersion, a uniform distribution
across space.
We use these concepts of spatial organization to characterize W&D locations and
measure changes over time. We use both absolute and relative measures of centrality and
concentration. Absolute measures provide information on changes in W&D spatial
patterns with respect to a fixed point, such as the city center. Relative measures provide
information on changes in W&D spatial patterns with respect to changes in other spatial
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
14

patterns, such as the population distribution. Relative measures indicate where goods may
be coming from or going to and hence may provide some indication of how these changes
could affect transport to and from markets. We generate four categories of measures:
absolute and relative measures of centrality, and absolute and relative measures of
concentration. Measures are listed in Table 2. There are many possible ways to generate
these measures. For example, we could measure centrality by the average distance of all
W&Ds to the city center, or to the geographic centroid of all W&Ds, as in the Dablanc and
co-author studies (Dablanc and Ross, 2012; Dablanc et al., 2014). We, therefore, generate
several different measures and compare results. Distance is calculated as Euclidean
distance. We compared network and Euclidean distances; they are highly correlated and
do not generate different results. We, therefore, used the simpler measure. Distance is
calculated from the ZIP Code centroid and weighted by the number of W&Ds or
employment.

Table 2 Four Categories of Spatial Structure Measures
Spatial Structure Absolute Relative
Measure of
Centrality

Measure 1. Decentralization

1.1 Average distance to CBD
1.2 Average distance to freight nodes
1.3 Average distance to W&D
geographic center

Measure 2. Relative Decentralization

2.1 Average distance to all employment
2.2 Average distance to all population

Measure of
Concentration

Measure 3. Concentration

3.1 Gini coefficient for W&Ds

Measure 4. Relative Concentration

4.1 W&D concentration by total employment
density quartiles

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
15

We use measures based on both establishments and employment of W&Ds for two
reasons. First, location choices of firms underlie changes in spatial distribution; hence, the
establishment is an appropriate unit of analysis. Second, a measure of business size is also
appropriate, because the research goal is to understand the effect of W&D location changes.
In the case of W&Ds, the physical size of the facility (square footage) is likely the best
measure for size, but such data are not available for prior time periods. We use
employment as a second-best proxy.

4 DATA
Our main data source is ZIP Code Business Patterns (ZBP) for 2003 and 2013. ZBP, a
subset of County Business Patterns (CBP), includes the number of establishments,
employment, and payroll for all establishments at the 6-digit industry code level. The
Business Register is the data source. Every establishment (a single physical location where
business is conducted) with at least one employee is included. The spatial unit of ZBP is
United States Postal Service ZIP Codes.
There are some significant limitations to the ZBP data. ZIP Codes are relatively large
spatial units, and we have no information on the location of businesses within the ZIP Code.
We must assume a distribution or use centroids published by Tele-Atlas. We compared
these centroids with the distribution of establishment level data within the ZIP Codes from
another data source for the Los Angeles region. We found that the centroids generally
represent the locations with the highest concentration of establishments. We, therefore,
choose to use centroids as the basis of our measures. Also, like census tracts, ZIP Code size
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
16

is correlated with development density. Thus, our distance measures (based on centroid to
centroid) become more approximate as ZIP Code size increases.
Employment counts below a threshold are suppressed for confidentiality and are
available only at County- or State-level. ZBP provides the number of establishments by
nine establishment size classes. To identify employment counts, we used quadratic
programming to find an establishment-size vector that minimizes the difference between
the county-level employment count and the ZIP Code-level sum of the number of
establishments in each size class multiplied by the size vector. Although there are various
sources of establishment data, for example, NETS or InfoUSA, our previous work has shown
that these data sets are not consistent over time or with more aggregate data, such as CBP.
We, therefore, trade off geographic specificity for temporal and spatial consistency.
To identify W&Ds, we use establishments within NAICS 493, warehousing and
storage, which includes facilities that store goods, and/or provide logistics services. In
order to generate our spatial measures, additional data were drawn from the US Census,
World Port Index, FAA, and Intermodal Association of North America.
We use the four largest metropolitan areas in California – Los Angeles CSA
(Combined Statistical Area), San Francisco CSA, Sacramento CSA, and San Diego MSA
(Metropolitan Statistical Area) – as our case study areas. They vary in size, industry mix
and role in the global economy. The extent to which each metropolitan area participates in
freight activities varies widely. Los Angeles CSA (LA) is the largest metro area in California
and has almost half of the population (17.9 million in 2010) and employment (6.5 million in
2013) of the State. It is a major international trade node. It has the largest container
seaport complex in the US, which handles 37% of all containerized trade (Strocko et al.,
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
17

2013), and it has the seventh largest air cargo volumes in the US (FAA, 2015). San
Francisco CSA (SF) is the second largest in terms of population (8.2 million in 2010) and
employment (3.4 million in 2013) size. It is well known as the largest high-tech center in
the US. San Diego MSA (SD) (3.1 million population in 2010) shares national borders with
Mexico and is second only to Texas in trans-border trade. Sacramento CSA (SC) is the
smallest in size (2.4 million population in 2010) and is not a major national or international
trade node.
The Los Angeles metro area is well-known for its polycentric urban form (Giuliano
and Small, 1991; Agarwal et al., 2012). Similarly, recent studies of the three other metro
areas show all of them to be polycentric to some degree (Strocko et al., 2013). In all cases,
however, the CBD remains the largest center and has a significant influence on the overall
employment distribution, holding effects of other centers constant. Thus, using the CBD as
the central point for distance measurement is appropriate.

5 RESULTS
We present results in two parts. First, we describe changes in W&D activity and its spatial
distribution. Second, we present results for our spatial location measures.

5.1 Changes in Warehousing and Distribution
Table 3 shows trends in the number of W&D establishments and employment for each case
study area. Table 3 shows different growth patterns for the four metropolitan areas. In
terms of the number of establishments, growth ranges from 2% (San Diego) to 79%
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
18

(Sacramento). In terms of employment, growth ranges from 4% (San Diego) to 52%
(Sacramento). There is no consistent relationship between establishment and employment
growth. It is possible that these differences are due to the type of W&Ds that were added
over the period. Larger warehouses are consistent with increased import/export trade.
Figures 1 through 3 present changes in spatial distribution of warehousing
establishments between 2003 and 2013. The size of the bubbles corresponds to the
number of W&D establishments by ZIP Code as presented in the legend. Note that the map
and legend are scaled to each metro area (e.g. the largest bubble for Los Angeles is ‘32 or
more’, whereas it is ‘7 or more’ for Sacramento/San Diego). The comparative size
difference of bubbles between the two-year periods shows where the W&D industry
expanded or shrunk. ZIP Code centroids that contain at least one W&D establishment in
either of the two periods are shown.

Table 3 Descriptive Statistics of W&D Establishments and Employment, 2003 - 2013
Year Los Angeles CSA San Francisco CSA Sacramento CSA San Diego MSA

Est. Emp. Est. Emp. Est. Emp. Est. Emp.
2003 775 34,333 257 9,603 80 3,699 84 1,650
2013 1,001 49,266 311 11,476 143 5,641 86 1,720
%∆ 29% 43% 21% 20% 79% 52% 2% 4%

The Los Angeles map (Figure 1) shows that there are many W&Ds in the core of the
region and along major highway corridors to the east. Although the number of
establishments grew in many places (for example around the ports), new growth is
particularly evident to the east around San Bernardino and Moreno Valley. San Francisco
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
19

(Figure 2) has a different spatial pattern, with many W&Ds clustered around the Bay, and
many others located many miles away to the east (Tracy, Stockton) and north (Vallejo,
Santa Rosa). In Sacramento (Figure 3, left), W&Ds are clustered around the CBD, and along
the major highway corridors to the south and northeast. The number of W&Ds greatly
increased, but the spatial pattern remains approximately the same: most of the expansion
has taken place within a 10-30 mile distance range from the CBD. In San Diego (Figure 3,
right), W&Ds are mainly distributed along the coast, to the south and north of the CBD. San
Diego is the one metro area that did not have a large increase in the number of W&Ds over
the period, and it can be seen that the distribution shifted. New warehousing emerged to
the north (again along a major highway), while clusters in the south declined.

Figure 1 W&D locations, 2003, 2013, Los Angeles

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
20

Figure 2 W&D locations, 2003, 2013, San Francisco

Figure 3 W&D locations, 2003, 2013, Sacramento (left) and San Diego (right)
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
21

5.2 Decentralization and Concentration
In this section, we present results of the four groups of spatial measures. In order to
determine whether differences are statistically significant, we conduct several tests. For
the average and relative distance measures 1-1 (CBD), 1-2 (freight nodes), 1-3 (geographic
center), 2-1 (population) and 2-2 (employment), we conduct Welch’s t-tests (unpaired,
unequal-variance t-tests) between 2003 and 2013, which do not require sample
independence. For the Gini Coefficient (3-1), we use the jackknife standard error to test for
statistically significant differences.

5.2.1 Absolute Decentralization
Table 4 gives results for the absolute decentralization measures. We show distances and
percent change where differences are statistically significant. Los Angeles stands out as the
only metro area for which all measures are significant, and in every case in the direction of
decentralization. For average distance to CBD, to airports, and to geo-center with respect
to employment, differences are significant for all four metro areas, and again in the
direction of decentralization. Looking at average distances, although the degree of change
is greatest for Los Angeles, San Francisco has the greatest average distance in both years
(about 35 miles for establishments and up to nearly 45 miles for employment). The two
smaller metro areas have much lower average distances across all our measures, with San
Diego having the shortest average differences. These results are consistent with their
smaller population size and likely greater availability of land closer to the CBD than in the
much larger metro areas. It would appear that physical geography plays a role; San
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
22

Francisco imposes more constraints on W&D with its large bay in the center and hilly
topography, while the expanse of flat land has the opposite effect in Sacramento.
Differences in establishments vs. employment measure results suggest that
decentralization is more pronounced for larger firms, as would be expected from the
literature. We compared the change in distance from CBD for two groups of W&Ds: those
with less than 100 employees, and those with 100 or more. Results for Los Angeles and
San Francisco are shown in Figure 4 (Sacramento and San Diego have a very small number
of large W&Ds). It can be seen that in both cases the large W&Ds account for most of the
increase in decentralization.

Figure 4 Average distance to CBD by size category, Los Angeles and San Francisco

25.0
27.8
25.3
35.9
0.0
10.0
20.0
30.0
40.0
50.0
60.0
2003 2013
Average Distance (miles)
Los Angeles
Less than 100 More than 100
34.2 34.7
42.7
47.7
0.0
10.0
20.0
30.0
40.0
50.0
60.0
2003 2013
Average Distance (miles)
San Francisco
Less than 100 More than 100
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
23

Table 4 Percentage change in absolute decentralization between 2003 and 2013

Los Angeles San Francisco Sacramento San Diego
1-1 Ave dist.
to CBD
W&Ds 14.2% n/s n/s n/s
(miles) 25.1 – 28.6
W&D Emp. 43.0% 8.3% 4 6% 21.0%
(miles) 25.3 – 36.1 41.4 – 44.8 13.2 – 13.8 8.6 – 10.4
1-2a Ave
dist. to
airports
W&Ds 7.1% n/s n/s n/s
(miles) 27.4 – 29.3
W&D Emp. 25.6% 11.1% 6.8% 22.9%
(miles) 26.5 – 33.3 37.7 – 41.9 15.5 – 16.6 8.9 – 10.9
1-2b Ave
dist. to
freight nodes
W&Ds 7.8% n/s n/a n/a
(miles) 26.3 – 28.4
W&D Emp. 27.0% n/s n/a n/a
(miles) 25.5 – 32.4
1-2c Ave dist.
to seaports
W&Ds 10.5% n/s n/a n/s
(miles) 28.1 – 31.0
W&D Emp. 34.5% 8.4% n/a 20.3%
(miles) 28.4 – 38.2 38.7 – 41.9 9.0 – 10.9
1-3 Ave dist.
from W&D
geo-center
W&Ds 9.7% n/s n/s n/s
(miles) 20.7 – 22.7
W&D Emp. 19.2% 4.8% 19.8% 12.0%
(miles) 19.3 – 23.0 25.1 – 26.3 11.4 – 13.7 8.8 – 9.8
(n/s: not significant; n/a: not applicable)

5.2.2 Relative Decentralization
Measure 2 considers W&D location changes relative to employment and population. Los
Angeles again stands out; relative decentralization is significant in every case, whether
measuring with respect to total employment or population. For establishments, the change
is about 2 miles, and for employment, the change is about 7 miles. Although all but one
change for total employment are significant, the magnitude of change is much smaller in
the other metro areas.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
24

Comparing the average distances in Measure 1-1 to Measure 2-1/2-2 shows
different patterns across the four metro areas. For Los Angeles, the average distances for
relative measures are greater than for the absolute measure, likely because of the
dispersion of employment across a vast area. For San Francisco, the absolute and relative
measures are similar. Patterns for Sacramento and San Diego are similar to Los Angeles,
with Sacramento the most extreme case. These long distances may be an artifact of the
geographic size of the metro area. With population and employment spread over hundreds
of square miles, any warehouse location will be far from some of the population or
employment.

Table 5 Percentage change in relative decentralization between 2003 and 2013

Los Angeles San Francisco Sacramento San Diego
2-1 Ave dist.
to all emp.
W&Ds 6.8% n/s n/s n/s
(miles) 33.2 – 35.5
W&D Emp. 19.9% 6.0% 3.0% 4.1%
(miles) 32.7 – 39.2 41.0 – 43.5 23.0 – 23.7 15.5 – 16.2
2-2 Ave dist.
to all pop.
W&Ds 7.7% n/s n/s n/s
(miles) 34.7 – 37.3
W&D Emp. 17.8% 4.1% n/s 2.6%
(miles) 34.0 – 40.0 41.8 – 43.5 17.2 – 17.7
(n/s: not significant)

5.2.3 Absolute Concentration
Results for the Gini Coefficient are given in Table 6. W&D activity is relatively more
concentrated with respect to employment than to establishments. The value of the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
25

coefficient is similar across metro areas. This could be a result of land use regulations or
agglomeration benefits. In 6 of the 8 cases, the Gini Coefficient increases significantly,
suggesting increased concentration. However, the Gini Coefficient has little spatial
meaning; we do not know if W&Ds are concentrated in adjacent ZIP Codes, or concentrated
in many dispersed ZIP Codes. Our maps suggest the latter.

Table 6 Percentage change in absolute concentration between 2003 and 2013

Los Angeles San Francisco Sacramento San Diego
3-1
Gini
coefficient
W&Ds 8.3% 0.7% n/s 32.3%
(Gini) 0.56 – 0.61 0.48 – 0.49 0.39 – 0.51
W&D Emp. 2.7% n/s 13.1% 10.4%
(Gini) 0.78 – 0.80 0.79 – 0.90 0.68 – 0.75
(n/s: not significant)

5.2.4 Relative Concentration
Finally, we compare the distribution of W&Ds to the distribution of employment density.
For each metro area, we generate quartiles of employment density and then calculate the
share of W&Ds in each quartile. The first quartile has the lowest employment density. The
density quartile is also non-spatial, but it provides a hint of urban structure with respect to
its density. The W&D shares, 2003 and 2013, are shown in Figure 5 for establishments.
Several observations may be drawn. First, the patterns for Los Angeles and San Francisco
are similar to each other. In both cases, W&Ds are shifting out of the highest quartile and
into the lower quartiles. The pattern is particularly pronounced for San Francisco. Our
distance measures do not capture the nature of the new locations of W&Ds. Although the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
26

distance measures for decentralization were consistently significant only for Los Angeles,
Figure 5 suggests that de-concentration is much greater for San Francisco, with nearly 50%
of all W&Ds located in the lowest two employment density quartiles, compared to about 25%
for Los Angeles. That is, Los Angeles decentralized more from 2003 to 2013, but San
Francisco has a far greater proportion of decentralized W&Ds. The pattern when we use
W&D employment (not presented here) is analogous except for much smaller shares of
employment in the highest quartile. Clearly, W&D facilities located in higher density areas
are smaller, as would be expected, given land prices in these metro areas. Using average
employment per establishment as a proxy for W&D size, we find that average size declined
in the highest density quartile for Los Angeles and San Francisco but increased for
Sacramento and San Diego. We surmise that land availability and price explain these
differences.

Figure 5 Share of W&D establishments in total employment density quartiles, 2003 and 2013

3%
7%
16%
20%
0%
1% 1% 0%
12%
17%
22%
29%
6% 7%
17%
29%
46%
44%
31%
23%
48%
36%
32%
20%
39%
31% 32%
28%
46%
56%
50% 51%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
LA-2003 LA-2013 SF-2003 SF-2013 SC-2003 SC-2013 SD-2003 SD-2013
1st Quartile 2nd Quartile 3rd Quartile 4th Quartile
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
27

6 SUMMARY AND CONCLUSIONS
Our results present a mixed picture of W&D location changes. When measured with
respect to establishments, only Los Angeles consistently shows decentralization over the
period. However, when measured with respect to employment, decentralization is
confirmed for all metro areas across most of our measures. Although Los Angeles showed
the most decentralization, San Francisco has the most decentralized distribution of W&Ds.
The unique geography of the metro area (together with exceptionally high land prices in
the core) has likely forced W&Ds to the periphery. W&D spatial distribution in the two
smaller metro areas is far less decentralized by all measures, likely because of greater
availability of land. Decentralization from airports and seaports are explained by their
location; only Sacramento has its major airport in a low-density area, and the Los Angeles
and San Francisco seaports are located in the regions’ core. The Gini Coefficient indicates
that W&Ds are relatively concentrated, and concentration is increasing. Our maps (Figures
1 – 3) suggest that this may simply reflect that most warehousing is located in populated
areas. W&D location with respect to employment indicates a form of spatial de-
concentration for Los Angeles and San Francisco, with W&Ds shifting to lower employment
density locations, but the opposite trend for the smaller metro areas.
Regarding our measures, distance to the CBD, to airports, and to the W&D geo-
center generate similar results. Although the distance values are different, the direction of
change and the rank order of magnitude of change are the same. Our relative
decentralization measures also give similar results. Thus, the overall trends are robust
across several different measures. The Gini Coefficient indicates that the magnitude of
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
28

concentration is similar across the metro areas, but give little information on concentration
locations. Our relative concentration measure does a better job of indicating where shifts
are taking place.

6.1 Explaining Results
What are the possible explanations for our results? We identify three factors:
metropolitan size, economic structure, and physical geography. Metropolitan size
(population) is correlated with density. In general, the largest metro areas have the highest
peak and average/median density. Density is a proxy for demand, and high density implies
high land prices. Thus as large metropolitan areas continue to grow, the more land-
intensive activities (manufacturing, trade, and transport) seek cheaper land away from the
center. Today’s metropolitan areas are polycentric; they have multiple activity clusters,
and thus multiple density “peaks.” The shift is away from the largest activity clusters. Also,
as competition for land grows, land use regulation may “zone out” less preferred activities.
For example, older industrial zones may be targeted for redevelopment to residential and
commercial mixed use. These types of pressures are apparent in Los Angeles and San
Francisco, but not yet for Sacramento and San Diego. Finally, in the smaller metro areas
labor force access is likely a consideration. There would be little marginal gain in land
price reduction to locating at the periphery, but a large loss in labor force access.
A second consideration is economic structure. Large metropolitan areas are the
hubs of global commerce and serve as national and regional distribution centers. W&Ds
thus serve both local and non-local markets. For W&Ds oriented to non-local markets,
location within the metropolitan area makes little difference. If location near the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
29

import/export node (airport, seaport) is prohibitive, an alternative near a major interstate
highway may be a good substitute. We used the Freight Analysis Framework 2007 (FAF)
data to compare internal, domestic and foreign flows across the four metro areas. As
expected, foreign flows for Los Angeles are about twice that of San Francisco, and several
times larger than those for the smaller metro areas.
The third factor is physical geography. The physical constraints of San Francisco
contribute to high land prices and limit where development can occur. In contrast, Los
Angeles has been able to spread across over 5,400 square miles. Because population and
employment are distributed across such a vast area, W&Ds are relatively closer to local
markets, even as they decentralize. Sacramento’s geography allows a similar spread as the
region grows, but low density and plentiful land availability relatively near the center does
not provide the push factors for W&D decentralization. San Diego also has some physical
constraints (the coast, borders with Mexico, and hilly terrain). Thus, development is more
concentrated, and at this point in the metro area’s development, W&D decentralization is
exhibited only with respect to employment.

6.2 Potential Implications for Truck VMT
Does changing spatial organization of the warehousing and distribution industry have
implications for truck VMT? First, with respect to W&D employment, there is considerable
evidence of decentralization, especially in the largest metro areas. The difference in
pattern between establishments and employment is consistent with larger facilities being
built where land is cheaper and more available. If all W&D activity were locally oriented,
this would imply more truck VMT. However, from the FAF data, import/export flows by
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
30

tonnage account for between 50 and 60 percent of total flows. It is possible that much of
the decentralization is driven by the growth in domestic and international trade,
particularly in the case of the largest W&Ds. With regional or national markets, location
with respect to serving the entire market is the critical factor. As these markets expand in
both scale and geographic size, we should expect W&Ds to locate in low land price areas
with good access to the interstate highway system. If larger, more distant W&Ds are
oriented to external trade, we cannot conclude that decentralization leads to more truck
VMT.
Second, W&Ds are located throughout the populated areas of each metro area. This
is logical, both for market and labor force access. Only in San Francisco do we observe a
large and growing portion (over 20% in both years) of W&Ds in areas with the lowest
employment density quartiles. We also note that W&Ds located within the core areas are
smaller, which likely implies more local distribution. We, therefore, cannot rule out that
local serving W&Ds continue to seek locations near their markets. The growth in e-
shopping and same day delivery should reinforce the demand for near market locations.
Indeed, in the Los Angeles area, W&D rents in the center of the region have risen
dramatically in the past few years.
Third, truck VMT could change without any change in the spatial locations of W&Ds.
Supply chains and product markets are constantly changing; therefore, shipment patterns
are constantly changing as well. One need only think about the rise in instant deliveries
available for online shoppers and its impact on consumer behavior to understand how
quickly such patterns can change.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
31

Our results lead to an obvious question: how might the changes in truck VMT
associated with these changes be examined? In an ideal world, we would have full
information on the type of W&D facility and the associated truck trips (e.g. all the forward
and backward links in the associated supply chain that occur within the metro area) for at
least two time periods. Sakai et al. (2015) had such data from Tokyo. They find that W&D
decentralization is associated with increased truck travel, because of the increased
distances between shipment origins and destinations. In Paris, a similar survey has been
conducted, but a comprehensive analysis has not yet been performed (Dablanc and Gardrat,
2015). No such data exists within the US. Absent such data, simulations based on partial
data may be the best approach.
More research is needed to understand why spatial patterns vary across
metropolitan areas, and to document the extent to which decentralization is taking place in
US metropolitan areas. More data on freight patterns at the sub-metropolitan level is
needed if we are to develop a better understanding of the relationship between spatial
organization, shipment patterns, and truck VMT.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
32

CHAPTER 3
WHY DO WAREHOUSES DECENTRALIZE MORE IN CERTAIN METROPOLITAN AREAS?

ABSTRACT
Over the last decade, warehousing and distribution centers have decentralized to the urban
peripheries where land is cheaper and readily available. This change in location patterns
has been driven by the demand to build more modernized and larger facilities to
accommodate an ever-increasing influx of freight. Since efficient freight movement is
essential for the smooth functioning of metropolitan areas, decentralization should occur
everywhere. However, this is not necessarily true. I hypothesize that depending on the
volume of goods movement and the spatial distribution of land prices, the extent of
decentralization varies across metropolitan areas. I test this hypothesis using 48 US
metropolitan areas. Results provide robust evidence that high land prices push large
warehouses away from central locations. When freight demand and land prices are not as
high, the effect becomes insignificant. Indeed, not only is decentralization linked with large
metro areas but also with very large warehouses.

1 INTRODUCTION
The purpose of this paper is to evaluate at the national level the factors that might explain
warehousing decentralization. I hypothesize that the variance across metro areas in freight
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
33

activity and land price distribution explains the variance in decentralization. Results of
descriptive analysis, hypothesis testing, and econometric models consistently show that
warehousing decentralization is a function of freight activity and land prices. To be specific,
it is closely linked with very large warehouses in large metro areas.
In recent decades, the logistics industry has prioritized throughput: moving large
volumes of products through the supply chain as quickly, cheaply, and reliably as possible.
This reorganization has resulted in a geographically-dispersed system of goods production
at the global scale. At the sub-metropolitan level, larger and automated warehousing and
distribution centers (W&D) have been built on the urban periphery, where land is cheaper
and readily available, hence warehousing decentralization. These spatial shifts have been
attributed to the rebalance of inventory and transportation costs: the gains from lower land
prices, economies of scale, and automation outweigh the increase in transportation costs as
warehouses move farther from the market.
Because efficient supply chains are essential for the smooth functioning of
metropolitan areas, changes in scale and location of W&D should occur everywhere.
However, this is not necessarily true. Recent literature has documented decentralization in
Atlanta, Los Angeles, Paris, Tokyo, and Toronto (Dablanc & Ross, 2012; Dablanc, Ogilvie &
Goodchild, 2014; Dablanc & Rakotonarivo, 2010; Sakai et al. 2015; Woudsma et al., 2015).
In Seattle, warehouses decreased the distance from their geographic center (Dablanc,
Ogilvie & Goodchild, 2014). Furthermore, according to Giuliano and Kang (2016), a case
study of the spatial dynamics of the warehousing industry in four metropolitan areas in
California between 2003 and 2013, not all major metro areas have experienced
decentralization. In San Francisco, Sacramento, and San Diego, warehouses made marginal
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
34

location changes. Los Angeles was the only place with significant changes in location. The
authors explain that the difference may be attributed to the variance in the characteristics
across metro areas, such as metro size, economic structure, and physical geography.
In this paper, I expand the scope of analysis and evaluate at the national level
whether the disparity in metro-level characteristics explains the difference in the extent of
warehousing decentralization. In that regard, this study contributes to the theoretical
understanding and empirical testing of the phenomenon. This paper is organized as
follows. Section 2 reviews recent literature on how/why warehouses have changed
location. Section 3 presents the research approach, measurements, and data. Section 4
presents results of descriptive and econometric analyses. In Section 5, this paper closes
with conclusions and future research suggestions.

2 LITERATURE REVIEW
2.1 Why Should We Care?
Over the last decade, various sources have documented the expansion of freight movement
and linked industry sectors in the U.S. From 2000 to 2011, the increase in foreign trade
(U.S. dollars) was significant (40%), relative to the moderate increase in U.S. population
(10%), employment (3%), and businesses (4%).
1
During roughly the same period,
containerized trade volumes (TEU) increased by 44%, and domestic commodity shipments
(U.S. dollars) increased by 29%.
2, 3
Moreover, between 2003 and 2013, the warehousing

1
Freight Facts and Figures 2013, USDOT Bureau of Transportation Statistics
2
Total vessel calls in U.S. port, terminals and lightering areas report 2002-2012, U.S. Maritime Administration
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
35

sector in terms the number of establishments and employment increased by 15% and 33%,
respectively.
4
These statistics indicate that the volume of goods shipped per capita
increased nationally. Factors attributed to this trend are globalized trade, consumer
demand shifts, just-in-time production, containerization, as well as advances in information,
logistics, and transportation technology, and the restructuring of the logistics industry
(Hesse & Rodrigue, 2004; McKinnon, 2009; Cidell, 2011).
Logistics restructuring has led to a spatial shift of warehousing facilities, which, in
turn, has influenced the geography of freight movement in urban areas (Hesse, 2007). It
has been argued that, if facilities are located farther from the urban center, this change may
contribute to increased freight vehicle miles traveled (VMT) and associated negative
externalities on society (e.g. GHG and criteria emissions, noise, congestion, increased fuel
consumption, infrastructure damage, and environment justice) (Anderson, Allen, & Browne,
2005; Andreoli, et al. 2010; Crainic et al. 2004; Dablanc, 2013; Dablanc et al., 2014; Dablanc
and Rakotonarivo, 2010; Dablanc and Ross, 2012; and USDOT, 2012; Wygonik et al., 2015).
Cost savings from relocating may accrue to logistics businesses, while any external costs
from increased vehicle miles are incurred by society at large (Hesse, 2006; Rodrigue, Slack,
& Comtois, 2001). Two studies based on facility-level surveys documented that facility
decentralization resulted in increased truck VMT (Dablanc and Rakotonarivo, 2010; Sakai
et al., 2015). However, there are many operational aspects to consider at the facility level
to accurately calculate the freight travel distance (Sakai et al., 2015). Some argue that the
gains from operational efficiency might offset the negative externalities when shipment is

3
Of 2012 dollars, Commodity Flow Survey, 2002-2012.
4
Warehouse definition derived from NAICS Sector 493 ‘warehousing and storage,’ Economic statistics from
County Business Patterns 2003-2013.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
36

consolidated through centralized logistics facilities (Kohn and Brodin, 2008). Moreover,
new warehousing facilities are more energy-efficient (Dhooma and Baker, 2012).
Therefore, the negative effects of decentralization remain uncertain.
There have been only a limited number of empirical evaluations of this impact
because shipment data are scarce (Sakai et al., 2015). Rather, many studies have focused
on analyzing the distribution of logistics facilities and the changes in W&D distribution
over time to draw implications on freight VMT (Allen, Browne & Cherrett, 2012; Bowen,
2008; Cidell, 2010; Dablanc & Ross, 2012; Dablanc & Rakotonarivo, 2010; Dablanc, Ogilvie
& Goodchild, 2014; Giuliano and Kang, 2016; Heitz and Dablanc, 2015; Sakai et al., 2015;
Van den Heuvel, et al., 2013; Woudsma et al., 2015). Aljohani and Thomson (2016)
provides a thorough review of how W&D distribution has been quantified. Another set of
literature, mostly based on stakeholder interviews, investigated location factors logistics
operators would consider when they choose a location for a facility (Jakubicek and
Woudsma, 2011; Warffemius, 2007). Additionally, some research has evaluated the
systematic factors that constitute warehousing rents (Buttimer et al., 1997; Sivitanidou,
1996). Most of the past studies could not draw a definitive answer to the VMT question but
rather have suggested several directions to proceed for future studies. Accordingly,
understanding how and why the distribution of warehousing facilities has changed may be
the first step to evaluate whether decentralization is a problem worthy of policy
intervention.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
37

2.2 Logistics Restructuring and Warehousing Decentralization
Until recently, goods were produced and stored well ahead of customer demand and were
infrequently shipped in larger volumes. These goods distribution activities happened in
areas of a large metropolitan population (McKinnon, 1983). Inventory control was
laborious because the process was not fully automated (Bowen, 2008). However, logistics
restructuring has changed these processes (McKinnon, 2009).
The primary goal of the restructuring was high throughput – to expand the capacity
and velocity of goods transportation (Hesse, 2004; Rodrigue, 2008). Cidell (2011) stated,
“Parts and products are not meant to sit on a shelf, but to be in constant motion along the
supply chain until the final product reaches store shelves” (pp. 835). That is, the system
has been restructured such that producers can transport a large volume of goods
frequently and reliably (Bowen, 2008; Cidell, 2011; Rodrigue, 2008; Dablanc & Ross, 2012).
The restructuring has been attributed to several factors. We are living in “a new
distribution economy” that is dependent on how efficiently goods are produced and
distributed via progressively more globalized systems (Hesse and Rodrigue, 2004, pp. 178;
Lavassani, et al. 2008; McKinnon, 2009). As customer demand has increased and
diversified, producers have also reprioritized from supply-push to demand-pull production
systems and compete based on time-saving operations (e.g. Just-In-Time production)
(Bowen, 2008; Lasserre, 2004). Moreover, major importers and big-box retailers have
changed how warehouses are utilized (e.g. from storage to shipment consolidation and
regional distribution) (Bowen, 2008; Christopherson and Belzer, 2009; Dablanc & Ross,
2012). These changes in how/where goods are distributed and sold have largely been
driven by the advances in information and logistics management technology and the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
38

concomitant rise of electronic commerce (Dablanc, et al. 2011; McKinnon, 2009; Bowen,
2008; Lavassani et al., 2008). This restructuring is a complex process that involves the
spatial dispersion of the entire system of production and distribution (Hesse & Rodrigue,
2004; Lavassani et al., 2008; Rodrigue, 2008; Cidell, 2011). All of these above factors have
contributed to logistics restructuring.
Over the past two to three decades, the logistics industry has expanded its facility
capacity near intermodal terminals to maintain high throughput. However, this approach
soon reached its limit due to development density, land constraints, and arterial congestion
(Hesse, 2006; Cidell, 2010). For example, in many cases, major airports, seaports, and
railways are in or near urban cores. To deal with these problems, warehousing and
distribution centers have been relocated to the urban periphery – with its vast amounts of
cheap land, large parcels, direct access to congestion-free highways and rail systems, as
well as low-skilled and low-wage labor, and a supportive regulatory and business
environment for logistics operations (Bowen, 2008; Christopherson and Belzer, 2009;
Cidell 2010 & 2011; Dablanc and Ross, 2012; Hesse, 2002, 2004, & 2007; Hesse & Rodrigue,
2004; McKinnon, 2009; Notteboom and Rodrigue, 2005; Rodrigue, 2006; Slack, 1998). This
logic for facility relocation applies to all major segments of the industry: warehousing,
trucking, freight forwarding, and air-cargo service (Hesse & Rodrigue, 2004).
The rebalancing of logistics costs between inventory and transport has eased the
relocation process (McKinnon, 2009). Most importantly, lower land prices offset the
increased transport costs as these facilities locate farther from their market (McKinnon,
2009). Facility automation, feasible when operated on a large scale, further decreased per-
unit inventory costs and enhanced maximum productivity (Bowen, 2008). Moreover,
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
39

owing to decreased transaction and per-ton-mile transport costs, logistics firms could
make facility location decisions within a much greater distance range (Glaeser & Kohlhase,
2004; Hall, Hesse & Rodrigue, 2006; Rodrigue, 2004). These large, automated warehouses
are “directly responsible” for decentralization (Dablanc & Ross, 2012, pp. 433). However,
there have been very few empirical studies which focused on the large warehouses (Bowen,
2008).

3 RESEARCH APPROACH
3.1 Conceptual Framework – Rationale behind Decentralization
The purpose of this paper is to explore factors associated with warehousing
decentralization in major U.S. metro areas. Figure 6 presents a schematic showing why
warehouses decentralize. The metropolitan resident and business have a freight demand,
which is correlated with the size of a population and industry. This demand is
unobservable until producers and retailers fulfill it using logistics services, for which
warehousing operators provide storage capacity. Warehouses and the logistics industry as
a whole are profit-driven business entities who seek “productivity enhancing location
attributes” (Sivitanidou, 1996, pp. 1262). According to the firm location choice literature,
the location attributes, in conjunction with the warehouse characteristics (type, size, etc.),
constitute the cost structure of selecting a location and influence the location choice
(McFadden, 1974; McFadden, 2001).
5

5
Whether a warehousing facility is owned by a logistics business or is leased, its location fulfills a tenant’s
profit maximization objectives. If a facility at a location does not fit any supply chain strategies (e.g. obsolete
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
40

In this context, I focus on the variation across metropolitan areas and hypothesize
that (1) the amount of freight activity and (2) the distribution of land prices jointly
influence the extent of warehousing location change. First, freight activity is a function of
population size. In very large metro areas, very high freight demand is present. Thus, a
large freight volume is destined to/originating from the area. In this case, large-scale
operation and facility automation may not be only feasible but also more sensible because
it will decrease per-unit inventory costs. If so, considering land price and many other
factors, a warehousing operator will build a new facility at an optimal location, therefore
expanding a metro area’s warehousing supply (Figure 6). As explained in the previous
section, the recent trend is that large warehouses are built on the urban periphery where
cheap land is readily available. Also, the recent logistics restructuring through the
advancement of ICT has facilitated the spatial shifts (Hesse & Rodrigue, 2004; McKinnon,
2009; Cidell, 2011). The addition of these large warehouses over a period will eventually
change their spatial patterns. Across metro areas, population size varies significantly. For
instance, population decreases exponentially with respect to its size rank (Zipf’s law;
Gabaix, 1999). If this is the case, the demand for large warehouses would vary widely,
hence a large variation in decentralization.

technology, small size, increased land rent or more stringent regulation), the facility of this location will not
be utilized and close down.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
41

Figure 6 A schematic of the mechanism of W&D location change

Moreover, freight activity is also correlated with a metro area’s trade gateway
function. Namely, major trade gateways transport a large volume of freight. Thus, the
mechanism of location change is similar to the previous case in which substantial demand
for goods movement results in large, automated facilities being built on the outskirts.
Globalized supply chains have facilitated this process in a way that more goods are
transported at the international scale (Hesse & Rodrigue, 2004). The variance in the size of
freight activity across metro areas is also large in that relatively few metropolitan areas
dominate in globalized supply chains – the ten largest container port systems
accommodated 78% of all US container imports (in TEU, the Maritime Administration,
2015). The warehouses in these trade gateways are more likely to decentralize further.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
42

Second, land prices vary not only across metro areas but also at the sub-
metropolitan level. Land price is correlated with land demand, which can be approximated
by population or employment density (Anas, Arnott, and Small, 1998). As discussed earlier,
population size varies widely across metro areas, hence population density varies as well
(Gabaix, 1999). Furthermore, population is not uniformly distributed within a metro area.
Rather, the densities in the urban core (whether of mono- or poly-centric) are significantly
greater than those in the suburbs or the exurbs (Giuliano and Small, 1991; Giuliano et al.,
2015). As suggested in the urban economics literature, this spatial distribution of densities
can be consistently quantified (Clark, 1951; McDonald, 1989; Anas, Arnott, and Small,
1998). If so, how the distribution of densities, as a proxy for land prices, influences the
extent of decentralization can be systematically tested.
In sum, the variance across metro areas in the freight volume and the land price
distribution explains the variance in decentralization. Here, it is worth noting again that the
fundamental driver of decentralization is the unquantifiable process of logistics
restructuring to transport large volumes of freight quickly, cheaply and reliably. Thus, I
first test how much the variation in freight demand across metro areas, jointly with the
land price distribution, explains variation in decentralization (cross section). Furthermore,
there may be some cases in which a significant increase in freight demand resulted in
logistics restructuring and consequential warehousing decentralization. Examples include
a significant increase in population, a drastic change in consumption patterns, a relocation
of expansive manufacturing complexes, or construction of major intermodal terminals
(world-class seaport, air hub, rail yard, and canal). Thus, I also evaluate whether the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
43

change in the freight demand and land price distribution explains the variation in
decentralization (time series). General models are:

∆𝐷 𝑖 = 𝑓 (𝐹 𝑖 , 𝐿 𝑖 ) Cross-section (1)
∆𝐷 𝑖 = 𝑓 (∆𝐹 𝑖 , ∆𝐿 𝑖 ) Time-series (2)

Where in metro area i;
∆ denotes change over time;
∆D is change in warehousing distribution (decentralization);
F is vector of freight demand;
L is vector of land price distribution.

Figure 7 Illustration of two linear models

In Figure 7, I further specify two linear models. In Model 1, I test the linear
relationship between freight flows, in conjunction with the distribution of land prices, and
the extent of decentralization (Equation 1-1). In Model 2, I replace the freight flows with
the change in the number of large W&Ds, as one of the outcomes of the logistics
restructuring to transport large freight flows. The metro areas with greater freight flows
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
44

are more likely to have large W&Ds built therein, and these facilities are more likely to have
been established in more distant locations, hence more decentralization. Thus, jointly with
the distribution of land prices, I test the linear relationship between the change in the
number of large W&Ds and decentralization (Equation 1-2). Equation 1-3 is a fully specified
model with all variables.

∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 = 𝑓 (𝐹𝑟𝑒𝑖𝑔 ℎ𝑡 𝑓𝑙𝑜𝑤 𝑖 ,𝑡 −1
, 𝐿 𝑖 ,𝑡 −1
) (1-1)
∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 = 𝑓 (∆𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,𝑡 −1 𝑡𝑜 𝑡 , 𝐿 𝑖 ,𝑡 −1
) (1-2)
∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 = 𝑓 (𝐹𝑟𝑒𝑖𝑔 ℎ𝑡 𝑓𝑙𝑜𝑤 𝑖 ,𝑡 −1
, ∆𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,𝑡 −1 𝑡𝑜 𝑡 , 𝐿 𝑖 ,𝑡 −1
) (1-3)

Where in metro area i;
∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 is change in warehousing distribution from t-1 to t;
𝐹𝑟𝑒𝑖𝑔 ℎ𝑡 𝑓𝑙𝑜𝑤 𝑖 ,𝑡 −1
is volume of freight flows at time t-1;
∆𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,𝑡 −1 𝑡 o 𝑡
is net change in the number of large W&Ds from t-1 to t;
𝐿 𝑖 ,𝑡 −1
is vector of the measures of land price distribution at time t-1.

Furthermore, I specify three more models that account for the changes in the freight
flows and land price distribution over time (Equation 2-1, 2-2, 2-3). The structure of these
models is equivalent to the previous models.

∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 = 𝑓 (∆𝐹𝑟𝑒𝑖𝑔 ℎ𝑡 𝑓𝑙𝑜𝑤 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 , ∆𝐿 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 ) (2-1)
∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 = 𝑓 (∆𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,𝑡 −1 𝑡𝑜 𝑡 , ∆𝐿 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 ) (2-2)
∆𝐷 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 = 𝑓 (∆𝐹𝑟𝑒𝑖𝑔 ℎ𝑡 𝑓𝑙𝑜𝑤 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 , ∆𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,𝑡 −1 𝑡𝑜 𝑡 , ∆𝐿 𝑖 ,𝑡 −1 𝑡𝑜 𝑡 ) (2-3)

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
45

However, because of data problems, I do not estimate time series models. I explain
the issues more in detail after I present data in page 64. To test the linear relationship, I
use ordinary least squares (OLS) estimation. The unit of analysis is a metro area, and I use
48 major US metro areas in the models.

3.2 Measurement and Data
3.2.1 Warehousing Decentralization: Changes in Distribution over Time
For data, I use ZIP Code Business Patterns (ZBP) for 2003 and 2013. 2003 is the earliest
year from which the ZBP datasets have become consistent after two rounds of industry
code revision in 1997 and 2002.
6
2013 is the latest available data year when I conducted
this study. The Business Register, the source of employer and establishment information in
the ZBP, maintains records of each known establishment with paid employees located in
the U.S. An “establishment” is defined as “a single physical location at which business is
conducted, or services or industrial operations are performed.”
7
Every business with an
EIN (Employer Identification Number) with at least one employee is included. The ZBP
provides the number of establishments by employee size classes at the 6-digit industry
code level.
8
The spatial unit of the ZBP is the United States Postal Service (USPS) ZIP Codes,
which are derived primarily from the businesses’ physical addresses.

6
The industry code changed from SIC to NAICS in 1997, and the NAICS was revised in 2002. SIC (Standard
Industrial Classification); NAICS (North American Industry Classification System)
7
CBP, Census Bureau (http://www. census. gov/econ/cbp/)
8
Employee size classes are: 1-4; 5-9; 10-19; 20-49; 50-99; 100-249; 250-499; 500-999; and more than 1000
employees
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
46

To identify warehouses, I use the NAICS 493 (warehousing and storage), which
includes facilities that store goods and/or provide logistics services. I define an arbitrary
threshold of 100 employees or more for large warehouses and calculate how many of these
facilities are in each metro area for both 2003 and 2013.
In Giuliano and Kang (2016), the authors define multiple spatial measures to
quantify warehousing distribution consistently. One of these calculates the average
distance to the central business district (CBD) from each warehouse in a metropolitan area.
The decentralization argument has focused on the spatial shift of warehouses from the
urban core to the periphery (Aljohani and Thompson, 2016). Accordingly, the CBD is used
as a benchmark to measure the distribution. In this paper, because of the ZIP Code-level
dataset, the CBD is defined as the centroid of the ZIP Code with the highest employment
density of a metro area. The calculation is based on the Euclidean distance between ZIP
Code centroids. This distribution is calculated by metro area as follows:

𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 =
∑ 𝑑 𝑗 ∗𝑒 𝑗 𝑁 𝑗 =1
𝐸 (3)

Where,
𝑑 𝑗 = distance to the CBD from ZIP Code j (j = 1, 2, …, N);
𝑒 𝑗 = number of W&Ds in ZIP Code j;
E = sum of 𝑒 𝑗 .

Therefore, decentralization in metro area i is calculated as the difference in
distribution from t and t+1:

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
47

𝐷𝑒𝑐𝑒𝑛𝑡𝑟𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 𝑖 ≡ 𝐶 ℎ𝑎𝑛𝑔𝑒𝑠 (∆) 𝑖𝑛 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑖 ,𝑡 𝑡𝑜 𝑡 +1

= 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑖 ,𝑡 +1
− 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑖 ,𝑡 (4)
3.2.2 Spatial Distribution of Land Prices
According to the urban economics literature, the spatial distribution of land prices is
approximated by the negative exponential curve of employment density (Clark, 1951;
McDonald, 1989; Anas, Arnott, and Small, 1998).

𝐷 (𝑥 ) = 𝐷 0
∗ 𝑒 −𝐺 ∗𝑥 +𝑢 (5)

Where,
D(x) = employment density at distance x from the CBD;
D0 = peak employment density at the CBD (x = 0);
x = distance from the CBD;
G = density gradient;
u = error term.

The logarithm transformation of Equation 5 yields a simple linear equation with a
slope, G, and an intercept, log(D0). I use the estimated density gradient, 𝐺 ̂
, and peak density,
log (𝐷 0
̂
), to describe the spatial distribution of land prices. The calculation is based on the
ZBP datasets for 2003.

𝑙𝑜𝑔 (𝐷 (𝑥 )) = 𝑙𝑜𝑔 (𝐷 0
) − 𝐺 ∗ 𝑥 + 𝑢 (as 𝑌 = 𝑎 − 𝑏𝑋 ) (6)

Figure 8 illustrates how the variance in density gradients and peak densities
influences the extent of decentralization. As discussed, employment density is a proxy for
land price, and I will use it as the measurement for illustration. I assume a monocentric
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
48

urban structure. The left side presents two exponential curves of two different metro areas
A and B that have the same gradient (0.08) but different peak density values (7.5 for A and
6.5 for B). Assume a) the optimal land price for a new warehouse is 200 and b)
warehousing operators look for an ideal location from the CBD and outward into the
periphery. The land price on the A line reaches 200 at approximately 28 miles from the
CBD, whereas on the B line it is 15 miles. This difference implies that controlling for the
density gradient, warehouses in a metro area with a higher peak density will decentralize
more to reach the optimal price than those with a lower peak density. Thus, peak density is
positively correlated with decentralization. The right side of Figure 8 again presents two
exponential curves of two metro areas C and D that have an identical peak density (7.0) yet
different gradients (0.12 for C and 0.06 for D). In order to reach the land price of 200,
warehouses on the C line will have to move less than those on the D. This implies that,
controlling for peak density, warehouses in a metro area with a higher density gradient will
decentralize less to reach the optimal level. Thus, density gradient is negatively correlated
with decentralization.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
49

Figure 8 Illustration of density gradient and peak density measures

3.2.3 Freight Flow
As discussed, freight flow is a proxy for the demand for warehouses to operate on a large
scale, and greater freight flows are correlated with more decentralization. To quantify the
flow, I use the Commodity Flow Survey (CFS) 2002. I calculate all domestic inbound and
outbound freight volumes (in million tons), which do not duplicate domestic flows
originating from and destined to the same area. The Bureau of Transportation Statistics
publishes the CFS data every five years. The CFS provides the origins and destinations of
freight flow by value (USD) and weight (tonnage) by mode of transportation at the
metropolitan and state levels. The data source is shipper-based surveys. The CFS 2002
does not include international trade portions, as all CFS flows originate from and are
destined to a domestic region. An alternative data source for international trade is the
Freight Analysis Framework (FAF) datasets – a refined version of the CFS. However, the
FAF 2002 is available only at the state level. Because all imported freight will be
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
50

transported either to the same region or to other domestic regions, all imports are
accounted in the CFS; this also applies to exports. Table 7 summarizes the dependent
variable and all explanatory variables.

Table 7 Definition and data source of variables
Variables Definition Data source
W&D decentralization
Changes in warehouse distribution from
2003 to 2013 quantified as the average
distance to the CBD from warehouses
ZIP Code Business Patterns in
2003 and 2013
Change in large
warehouses
The net change in the number of
warehouses with 100 or more employees
ZIP Code Business Patterns 2003
Density gradient Estimated 𝐺 ̂
in Equation 6 ZIP Code Business Patterns 2003
Peak density Estimated log(𝐷 0
̂
) in Equation 6 ZIP Code Business Patterns 2003
Freight flow All domestic freight flows in million tons Commodity Flow Survey 2002

4 RESULTS
Results are presented in two parts. First, I describe how the extent of warehousing
decentralization differs across metro areas. I also conduct multiple hypothesis tests
regarding the distribution and location change of warehouses. Second, I present the results
of econometric model estimations.

4.1 Descriptive Statistics and Hypothesis Testing
Table 8 and Table 9 list 48 US metropolitan areas with their population and employment
statistics – 42 of which are Combined Statistical Areas (CSAs) and six of which are
Metropolitan Statistical Areas (MSAs). The MSA is a metropolitan statistical area with
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
51

either one or multiple counties that have at least one urban core with a minimum
population of 10,000. A CSA consists of either a single or multiple neighboring MSAs with a
significant level of economic interactions quantified by commuting patterns. The CSAs and
MSAs are delineated by the Office of Management and Budget.
9
In this study, I draw 48
metro areas from the CFS regions, which are consistent with CSA/MSA boundaries. I
exclude those metro areas with suppressed freight flow information. Across metro areas,
population size in 2000 varies considerably from 22 million in New York to 0.9 million in
Tucson. The median population of all 48 metro areas is 2.06 million, and employment 0.94
million.
After a preliminary analysis, I observed a non-linear relationship between
decentralization and population size. Thus, I divide 48 metro areas into two groups: Group
1 Large metro areas (size rank #1-22) and Group 2 Small metro areas (size rank #23-48).
This arbitrary division is based on the scatter plot of decentralization and metro size (see
Figure 9). Later in the regression models, to account for the unobservable heterogeneity
between the large and small metro areas, I use a metro-size dummy interaction variable.

9
http://www.census.gov/population/metro/data/glossary.html
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
52

Table 8 Large metro areas (N=22) and their population (2000) and employment (2003)
Rank Short Name Type Full Name
Pop.
2000
(Million)
Emp.
2003
(Million)
1 New York CSA New York-Newark, NY-NJ-CT-PA 22.24 9.22
2 Los Angeles CSA Los Angeles-Long Beach, CA 16.37 6.43
3 Chicago CSA Chicago-Naperville, IL-IN-WI 9.47 4.16
4 Washington DC CSA
Washington-Baltimore-Arlington, DC-MD-VA-
WV-PA
7.98 3.13
5 San Francisco CSA San Jose-San Francisco-Oakland, CA 7.66 3.32
6 Boston CSA Boston-Worcester-Providence, MA-RI-NH-CT 7.63 3.53
7 Philadelphia CSA Philadelphia-Reading-Camden, PA-NJ-DE-MD 6.69 2.89
8 Dallas CSA Dallas-Fort Worth, TX-OK 5.57 2.56
9 Miami CSA Miami-Fort Lauderdale-Port St. Lucie, FL 5.48 2.06
10 Detroit CSA Detroit-Warren-Ann Arbor, MI 5.46 2.24
11 Houston CSA Houston-The Woodlands, TX 4.88 2.10
12 Atlanta CSA Atlanta-Athens-Clarke-Sandy Springs 4.78 2.25
13 Seattle CSA Seattle-Tacoma, WA 3.78 1.65
14 Cleveland CSA Cleveland-Akron-Canton, OH 3.58 1.54
15 Phoenix MSA Phoenix-Mesa-Scottsdale, AZ 3.25 1.41
16 San Diego MSA San Diego-Carlsbad, CA 2.81 1.12
17 St. Louis CSA St. Louis-St. Charles-Farmington, MO-IL 2.77 1.26
18 Pittsburgh CSA Pittsburgh-New Castle-Weirton, PA-OH-WV 2.75 1.14
19 Denver CSA Denver-Aurora, CO 2.63 1.23
20 Portland CSA Portland-Vancouver-Salem, OR-WA 2.55 1.04
21 Tampa MSA Tampa-St. Petersburg-Clearwater, FL 2.40 0.99
22 Orlando CSA Orlando-Deltona-Daytona Beach, FL 2.19 0.97
Median 4.83 2.08

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
53

Table 9 Small metro areas (N=26) and their population (2000) and employment (2003)
Rank Short Name Type Full Name
Pop.
2000
(Million)
Emp.
2003
(Million)
23 Kansas City CSA Kansas City-Overland Park-Kansas City, MO-KS 2.12 0.99
24 Columbus CSA Columbus-Marion-Zanesville, OH 2.07 0.94
25 Cincinnati CSA Cincinnati-Wilmington-Maysville, OH-KY-IN 2.05 0.94
26 Indianapolis CSA Indianapolis-Carmel-Muncie, IN 2.03 0.95
27 Milwaukee CSA Milwaukee-Racine-Waukesha, WI 1.94 0.93
28 Charlotte CSA Charlotte-Concord, NC-SC 1.87 0.93
29 Salt Lake City CSA Salt Lake City-Provo-Orem, UT 1.85 0.77
30 San Antonio MSA San Antonio-New Braunfels, TX 1.71 0.65
31 Virginia Beach CSA Virginia Beach-Norfolk, VA-NC 1.67 0.62
32 Las Vegas CSA Las Vegas-Henderson, NV-AZ 1.56 0.75
33 New Orleans CSA New Orleans-Metairie-Hammond, LA-MS 1.53 0.58
34 Nashville CSA Nashville-Davidson–Murfreesboro, TN 1.49 0.71
35 Raleigh CSA Raleigh-Durham-Chapel Hill, NC 1.46 0.67
36 Greensboro CSA Greensboro–Winston-Salem–High Point, NC 1.41 0.64
37 Louisville CSA
Louisville/Jefferson County–Elizabethtown–
Madison, KY-IN
1.32 0.58
38 Grand Rapids CSA Grand Rapids-Wyoming-Muskegon, MI 1.31 0.57
39 Buffalo CSA Buffalo-Cheektowaga, NY 1.25 0.50
40 Austin MSA Austin-Round Rock, TX 1.25 0.55
41 Birmingham CSA Birmingham-Hoover-Talladega, AL 1.22 0.49
42 Greenville CSA Greenville-Spartanburg-Anderson, SC 1.22 0.52
43 Rochester CSA Rochester-Batavia-Seneca Falls, NY 1.16 0.46
44 Albany CSA Albany-Schenectady, NY 1.12 0.41
45 Dayton CSA Dayton-Springfield-Sidney, OH 1.09 0.45
46 Richmond MSA Richmond, VA 1.06 0.49
47 Tulsa CSA Tulsa-Muskogee-Bartlesville, OK 1.02 0.41
48 Tucson CSA Tucson-Nogales, AZ 0.88 0.31
Median 1.44 0.60

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
54

4.1.1 W&D decentralization in 48 U.S. Metropolitan Areas
Table 10 presents the summary statistics of decentralization by metro size. The mean and
median decentralization in all metro areas is 1.06 and 1.07 miles, respectively. The
standard deviation is large (2.23), and is even larger (2.73) for Group 2. Of the 48 metro
areas, warehouses in New Orleans decentralized the most (6.50 miles farther from the
CBD), whereas those in Tucson substantially centralized (6.75 miles closer to the CBD).

Table 10 Summary statistics of decentralization by metro size
Variable Group N Mean Median SD Min Max
Decentralization
2003-2013
(miles)
All metro areas 48 1.06 1.07 2.23
-6.75
Tucson
6.50
New Orleans
Group 1
(Rank #1-22)
22 1.10 1.25 1.49
-1.81
St. Louis
3.66
Miami
Group 2
(Rank #23-48)
26 1.02 0.60 2.73
-6.75
Tucson
6.50
New Orleans

Figure 9 presents the scatter plot of metro size in logarithm with base ten
population and decentralization in miles. The population of ten million in log is seven, and
one million in log is six. The labels (#1-48) represent the size rank in Table 8 and 9. The
distribution is clearly distinguished between the two groups. At approximately 2.2 million
in population (1 million employment), the scatter pattern changes. It is noticeable that
population size – a proxy for freight demand – has a positive correlation with
decentralization for Group 1 (Corr. = 0.65). The relationship is not significant for Group 2
(Corr. = 0.28).

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
55

Group 1 (Rank #1-22) in Orange; Group 2 (Rank #23-48) in gray; (N=48)
Figure 9 Scatter plot of W&D decentralization and 2000 population

Below are results of three hypothesis tests. The first hypothesis (H1) is that the 48
metro areas, as a whole, did experience statistically significant warehousing
decentralization. The t-test rejected the null hypothesis that the mean of decentralization
by metro area is equal to zero.
10
The second (H2) is that warehouses in Group 1 did
decentralize significantly more than those in Group 2. However, the t-test did not reject the
null that the mean of the change is equal to each other.
11
The statistics in Table 10 and the
scatter plot in Figure 9 all support this result. The third hypothesis (H3) is tested in each

10
𝐻 0
1
: ∆𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖 𝑜 𝑛 ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 0; all t-tests based on P<0.05
11
𝐻 0
2
: ∆𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝐺𝑟𝑜𝑢𝑝 1
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝐺𝑟𝑜𝑢𝑝 2
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
-8.00
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
8.00
5.80 6.00 6.20 6.40 6.60 6.80 7.00 7.20 7.40 7.60
W&D decentralization 2003-2013 (mile)
Population 2000 (log with base 10)
(10 million)
(1 million)
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
56

metro area that from 2003 to 2013 warehouses did decentralize significantly. The null
hypothesis is by ZIP Code the distribution in 2003 is equal to that in 2013.
12
Results are in
Table 11. Of the 48 samples, a total of seven, not only large metro areas (Los Angeles,
Chicago, Atlanta, Detroit) but also several small ones (Milwaukee, Las Vegas, and New
Orleans), experienced significant W&D decentralization.
13

Table 11 Decentralization by significance and by metro size
Decentralization Group 1 – Large metro areas Group 2 – Small metro areas
Significant Los Angeles, Chicago, Atlanta, Detroit Milwaukee, Las Vegas, New Orleans
Not significant
New York, Washington-DC, Boston, San
Francisco, Dallas, Philadelphia, Houston,
Miami, Seattle, Phoenix, Cleveland,
Denver, St. Louis, Pittsburgh, San Diego,
Portland, Orlando
Tampa, Indianapolis, Charlotte, Kansas
City, Columbus, Cincinnati, Salt Lake City,
San Antonio, Nashville, Raleigh, Austin,
Louisville, Greensboro, Virginia Beach,
Grand Rapids, Richmond, Greenville,
Buffalo, Birmingham, Rochester, Tulsa,
Albany, Dayton, Tucson

4.1.2 Distribution and Decentralization of Large W&Ds
Now I turn the focus to the location change of large warehouses, which hire 100 or more
employees. Employment might not be perfectly correlated with a facility’s floor area. For
instance, a product picking-and-packing facility might hire more employees per unit area
than a regional distribution center. However, because the floor area information is not
available at the national level, I use employment as a proxy. Only a small portion of
warehouses are in this size category – approximately 9% in 2003 and 11% in 2013. In
order to minimize the bias originating from the small sample number, I analyze only 24

12
𝐻 0
3
: 𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑖 ,2003
̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅
= 𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡 𝑖 𝑜𝑛
𝑖 ,2013
̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅
, in metro area i (i=1, 2, …, 48)
13
Detailed statistics of the number, distribution and decentralization of W&Ds are available in Appendix A2.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
57

metro areas with at least ten large warehouses in both 2003 and 2013. In Figure 10, I
present the scatter plot of population size and decentralization of large warehouses. I use a
green marker for larger ones and a gray marker otherwise, so the metro label appears
twice. Overall, in almost all metro areas, large warehouses decentralized more than small
ones. It can also be seen that, with respect to the unit change in population size, the
variance of large warehouse decentralization is greater than the variance of small ones’.

Only those metro areas with at least ten large W&Ds in 2003 and 2013 (N=24)
Figure 10 Scatter plot of decentralization of large/small W&Ds and 2000 population

1
2
3
4
5
6
7
8
10
11
12
13
15
19
20
23
24
25
26
28
29
34
36
42
1
2
3
4
5
6
7
8
10
11
12
13
15
19
20
23
24
25
26
28
29
34
36
42
-8.00
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
8.00
10.00
12.00
5.80 6.00 6.20 6.40 6.60 6.80 7.00 7.20 7.40 7.60
W&D decentralization 2003-2013 (mile)
Population 2000 (log 10)
Small W&Ds Large W&Ds
(10 million)
(1 million)
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
58

Below are results of four more hypothesis tests. The fourth hypothesis (H4) is that
the pattern of large warehouses shows more dispersal than that of small ones in 2003 and
2013 separately.
14
If land-intensive businesses are sensitive to high land price, large and
small facilities will be allocated differently, hence different distribution patterns. If the
demand for large warehouses changes over time, there would also be changes in the
distribution between 2003 and 2013. Table 12 presents the results. In 2003, the null
hypothesis was not rejected in 16 of the 24 metro areas. Whereas in 2013, in more than
half of the samples, the null was rejected, hence large warehouses were significantly
farther from the CBD. Over the period, in Los Angeles, Boston, Philadelphia, Dallas,
Houston, Atlanta, and Charlotte, the distribution of large warehouses has become
distinctively different from that of small ones. In two metro areas, Columbus and Phoenix,
their distribution has become more similar.

Table 12 Comparison of the distribution of large and small W&Ds
Test results In 2003 In 2013
Large W&Ds are
significantly
farther from
the CBD
New York, Chicago, Washington-DC, San
Francisco, Columbus, Nashville,
Greenville (N= 7)
New York, Los Angeles, Chicago,
Washington-DC, San Francisco, Boston,
Philadelphia, Dallas, Houston, Atlanta,
Charlotte, Nashville, Greenville (N= 13)
Not different
Los Angeles, Boston, Philadelphia, Dallas,
Detroit, Houston, Atlanta, Seattle,
Denver, Portland, Kansa City, Cincinnati,
Indianapolis, Charlotte, Salt Lake City,
Greensboro (N= 16)
Detroit, Seattle, Phoenix, Denver,
Portland, Kansa City, Columbus,
Cincinnati, Indianapolis, Salt Lake City,
Greensboro (N = 11)
Large W&Ds are
significantly
closer to the CBD
Phoenix (N= 1) -

14
𝐻 0
4
: 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,𝑡 ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜 𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝑖 ,𝑡 ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
, in metro area i (i=1, 2, …, 24); t =
2003, 2013
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
59

Excluded metro areas are: Miami, Cleveland, San Diego, St. Louis, Pittsburgh, Tampa, Orlando, Milwaukee, San
Antonio, Virginia Beach, Las Vegas, New Orleans, Raleigh, Louisville, Grand Rapids, Buffalo, Austin,
Birmingham, Rochester, Albany, Dayton, Richmond, Tulsa, and Tucson

The fifth hypothesis test is that large warehouses decentralized more than small
warehouses across metro areas (H5). The t-test rejected the null that the mean of large
ones’ decentralization is equal to that of small ones’.
15
Table 13 presents the summary
statistics. Notwithstanding the large standard deviation, the mean of large warehouses’
decentralization is much greater than that of smaller ones’.

Table 13 Summary statistics of decentralization by W&D size
Variable Group N Mean Median SD Min Max
Decentralization
2003-2013
(miles)
Large W&Ds
(Emp. ≥ 100)
24 2.88 3.42 4.24
-6.23
Greensboro
10.62
Los Angeles
Small W&Ds
(Emp. < 100)
24 0.70 0.95 1.45
-2.75
Greenville
3.53
Detroit

The sixth hypothesis (H6-1) is that the extent of decentralization of large
warehouses in large metro areas is greater than that in small metro areas (H6-1).
16
If land
prices are higher in large metro areas, large warehouses will relocate to the outskirts to a
greater extent than they would do in small metro areas. Similarly, I test if small
warehouses are sensitive to land price (H6-2). The t-test rejected the null H6-1, but not the
null H6-2. There is indeed a difference between large and small metro areas in the extent
of large warehouses’ decentralization, which is not the case for small warehouses.

15
𝐻 0
5
: ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

16
𝐻 0
6−1
: ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔 𝑒 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 1
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙 𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 2
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
and
𝐻 0
6−2
: ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 1
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 2
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
60

Seventh, I evaluate whether the location changes of large warehouses reach
significance. The null hypothesis is that the distribution at the ZIP Code level in 2003 is
equal to that in 2013.
17
Results are in Table 14. Of the 24 metro areas, Los Angeles,
Chicago, Boston, and Phoenix have experienced significant distribution changes over the
ten-year period. Large warehouses did not change location significantly in small metro
areas.
18

Table 14 Decentralization of large W&Ds by significance by metro size
Decentralization Group 1 – Large metro areas Group 2 – Small metro areas
Significant Los Angeles, Chicago, Boston, Phoenix -
Not significant
New York, Washington-DC, San
Francisco, Philadelphia, Dallas, Detroit,
Houston, Atlanta, Seattle, Denver,
Portland
Kansa City, Columbus, Cincinnati,
Indianapolis, Charlotte, Salt Lake City,
Nashville, Greensboro, Greenville

4.1.3 Decentralization and Freight Flow
Lastly, I evaluate whether freight flow is correlated with decentralization. This time, I
compare among quartile groups delineated by freight volume. Table 15 presents summary
statistics. The 4
th
quartile has the largest freight volume. Due to the small sample size, no
formal comparison of the mean between the quartile groups has been made. However,
given the large variances of all four groups relative to their means, it is hard to claim the
mean of any one group is greater than the others. Rather, as shown in Figure 11, the non-
linear relationship is again existing – positive correlation where trade volume is high; not
so clear otherwise. The freight volume is in logarithm with base 10.

17
𝐻 0
7
: 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,2003
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖 𝑜 𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,2013
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
, in metro area i (i=1, 2, …, 24)
18
A summary of hypothesis testing is in Appendix A1.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
61

Table 15 Decentralization by freight flow quartile groups
Variable Freight flow N Mean Median SD
Decentralization
2003-2013
(miles)
4
th
quartile 12 1.80 1.78 2.16
3
rd
quartile 12 1.31 1.41 1.31
2
nd
quartile 12 0.77 0.55 1.75
1
st
quartile 12 0.33 0.04 3.23

Figure 11 Scatter plot of W&D decentralization and 2002 freight flow

4.2 Model Results
In this section, I present econometric model results. Table 16 shows the summary
statistics of the explanatory variables. From 2003 to 2013, 247 large warehouses were
added to 48 metro areas – 5 per metro area on average. The change is correlated with
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
-8.00
-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
8.00
1.00 1.20 1.40 1.60 1.80 2.00 2.20 2.40 2.60 2.80 3.00
Average decentralization of W&Ds (mile)
Freight volume (log)
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
62

metro size. Atlanta, followed by Los Angeles, Chicago, Dallas, and Miami, gained the most,
while Austin lost the most. The distribution of freight flows is highly skewed: the top five
trade gateways (Chicago, Houston, Los Angeles, New York, and New Orleans) transported
36% of all freight volume (in million tons). Thus, I use a logarithm form. The density
gradient and peak densities are estimated figures from the density gradient estimation
(equation 5). The density gradient has an inverse correlation (-0.36) with population,
whereas the peak density has a positive correlation (0.52). For example, large metro areas
with polycentric urban centers and small metro areas without a significant urban core have
gentle density gradients (e.g. bottom 5: Seattle, Miami, Las Vegas, Los Angeles, and Virginia
Beach). Moreover, large metro areas, as expected, have urban cores with very high peak
densities (e.g. top 5: Chicago, New York, Los Angeles, San Francisco, and Detroit).
19
In
summary, densities (a proxy for the land price) in large metro areas are significantly high
in their urban cores and decrease at a lesser rate per-unit-distance from the center. In
small metro areas, not only is the peak much lower but it also decreases at a greater rate.
These differences should drive what I documented in the last section – more spatial shifts
in large metro areas.

19
The estimated peak density of Chicago is very slightly greater than that of New York.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
63

Table 16 Summary statistics of explanatory variables
Explanatory variables N Mean Median SD Min Max
Freight flow (million ton) 48 151.53 100.25 146.99
14.82
Tucson
724.00
Chicago
Change in large W&Ds 48 5.15 3 6.80
-3
Austin
32
Atlanta
Density gradient (𝐺 ̂
) 48 0.09 0.09 0.03
-0.01
Seattle
0.18
Austin
Peak density (𝐷 0
̂
, log) 48 7.02 7.22 0.96
3.50
Seattle
8.69
Chicago

Table 17 presents the pairwise correlation between the explanatory variables and
decentralization by size group. Again, the statistics suggest a non-linear relationship in
that the correlation coefficients are significantly different between the two groups,
particularly for the variables: change in large warehouses, density gradient, and peak density.
Freight flow is moderately correlated with decentralization in all groups. Table 18 shows
the pairwise correlation among the explanatory variables. Freight flow is correlated with
the change in large W&Ds and peak density, which is reasonable because all three variables
are partially a function of metro size. All other variables have a moderate level of
correlation.

Table 17 Pairwise correlation of explanatory variables with decentralization
Explanatory variables
All
metro areas
(N=48)
Group 1
Rank #1-22
(N=22)
Group 2
Rank #23-48
(N=26)
Freight flow (million ton) 0.41 0.46 0.56
Change in large W&Ds 0.24 0.62 -0.01
Density gradient (𝐺 ̂
) -0.06 -0.48 0.11
Peak density (𝐷 0
̂
, log) 0.22 0.19 0.32

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
64

Table 18 Pairwise correlation between the explanatory variables
Explanatory variables
(N=48)
Foreign flow
Change in
large W&Ds
Density gradient Peak density
Freight flow (million ton) 1.00
Change in large W&Ds 0.54 1.00
Density gradient (𝐺 ̂
) -0.16 -0.20 1.00
Peak density (𝐷 0
̂
, log) 0.57 0.31 0.35 1.00

Of the models laid out in the research approach, I present results of the cross-
section models only. I do not estimate time series models, because a) the change from
2003 to 2013 in the estimated density gradients and peak densities is marginal (corr. =
0.997 and 0.991, respectively) and b) the geography delineation of the 2012 CFS is neither
consistent nor comparable with that of the 2002 CFS. Hence, I estimate three cross section
models: Model 1 using freight flow and land price measures; Model 2 using change in large
warehouses and land price measures; and Model 3 using all variables. Models are run twice
using decentralization of all warehouses (Model 1-1A, 1-2A, and 1-3A) and that of large
warehouses (Model 1-1L, 1-2L, and 1-3L) as dependent variables separately. For all
warehouse model runs, a metro-size dummy interaction is incorporated to account for the
non-linearity (Small = 1, if Group 2; Small = 0, otherwise). I present stepwise results of the
fully specified models (Model 1-3A and 1-3L). Results of the other specifications (Models
1-1 and 1-2) are not presented because the models are underspecified. In order to
compare the effect size across variables, I present standardized coefficients. The model is:

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
65

Model 3:
∆𝐷 2003−2013
= 𝛽 0
+ 𝛽 1
𝐹𝑙𝑜𝑤 2002
+ 𝛽 2
𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷 2003−2013
+ 𝛽 3
𝐺𝑟𝑎𝑑𝑖𝑒𝑛𝑡 2003
+ 𝛽 4
𝑃𝑒𝑎𝑘 2003

+𝛽 5
𝑆 ∗ 𝐹𝑙𝑜𝑤 2002
+ 𝛽 6
𝑆 ∗ 𝐿𝑎𝑟𝑔𝑒 𝑊 &𝐷 2003−2013
+ 𝛽 7
𝑆 ∗ 𝐺𝑟𝑎𝑑𝑖𝑒𝑛𝑡 2003
+ 𝛽 8
𝑆 ∗ P𝑒𝑎𝑘 2003
+
𝛽 9
𝑆 + 𝜀

Where,
∆D is change in W&D distribution (decentralization);
Flow is freight flow;
Large W&D is change in the number of large warehouses;
Gradient is density gradient;
Peak is peak density;
S is small metro area dummy; and ε is an error term

Results of all warehouse models are shown in Table 19. Given the small sample size
(N=48), the model offers a reasonable level of explanatory power (R
2
= 0.431, Step 3).
Moreover, as expected, the statistical significance of the estimated parameters varies
between large and small metro areas. For Step 1, freight flow for large metro areas (Group
1, when Small = 0) is significant and has expected signs. With one standard deviation (SD)
increase in freight flow in large metro areas, decentralization increases by 0.354 SD. For
Step 2, the inclusion of change in large W&Ds alters the size and significance of freight flow
variables: freight flow becomes insignificant, whereas Small*freight flow becomes
significant. Change in large W&Ds in large metro areas is significant and positive, whereas,
in small metro areas, it is not different from zero (0.272-0.324). As documented in the
previous hypothesis testing and summary statistics, the correlation between freight
demand measures and decentralization differs between large and small metro areas. In
large metro areas, decentralization increases by 0.272 SD with a net increase (1 SD) of
large warehouses. In small metro areas, decentralization also significantly increases as
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
66

freight flow increases.
20
For Step 3, despite the inclusion of land price measures, results of
freight demand measures do not change. The correlation between land price measures and
decentralization is significant and as expected. Generally speaking, a steeper density
gradient results in less decentralization, whereas a higher peak density leads to more.
When the absolute values of the standardized coefficients are compared, the variance in
density gradient influences the variance in decentralization the most (-0.524).
21
Again, the
correlation between land price measures and decentralization differs between large and
small metro areas. In sum, the distribution of land prices with respect to the rate of
decrease by the unit distance from the center is a critical factor in large metro areas. I
surmise that land prices are already relatively high for warehousing operations in most
large metro areas. A recent report from Cushman & Wakefield, a commercial real estate
services company, documented that warehousing development has been the strongest in
primary markets (the Inland Empire near Los Angeles
22
, Chicago, Dallas, Houston, and New
Jersey) and the largest growth in rents has been in markets with constrained land supply
(Cushman & Wakefield, 2015). Thus, decentralization was inevitable. In small metro areas,
land prices are not as high, hence decentralization is primarily a function of freight flow.

20
In the small metro area-only model, which I did not present here, the standardized coefficients of freight
flow are 0.616 in Step 2 and 0.545 in Step 3. Full results are available upon request.
21
In small metro areas, freight flow is the most explanatory variable.
22
The Inland Empire is an area with intense warehousing activity in San Bernardino and Riverside counties
in the Los Angeles region.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
67

Table 19 Results of model estimation: Model 1A and 2A (all warehouses)
W&D
Decentralization
2003-2013
Model 1-3A
Step 1

Model 1-3A
Step 2

Model 1-3A
Step 3

Std. β Sig.

Std. β Sig.

Std. β Sig.
Freight flow 0.354 ** 0.126 0.046
∆Large W&Ds 0.272 * 0.214 **
Density gradient -0.524 **
Peak density 0.259 **
Small*Freight flow 1.500 2.379 * 2.283 *
Small*∆Large W&Ds -0.324 * -0.300 *
Small*Gradient 0.517
Small*Peak density 0.503
Small -1.277 -2.018 * -2.646 *
Constant . . .
R
2
0.293 0.357 0.431
N 48 48 48
(** P < 0.01; * P < 0.05; + P < 0.1)

The results for large warehouse models (N=24) are consistent with the previous
results. Table 20 presents the stepwise results of Models 1-3L. Results of Steps 1 and 2 are
very similar. Regardless of the inclusion of changes in large W&Ds, freight flow remains
significant and positive. In Step 3, when land price measures are included, the change in
large W&Ds reaches significance with a size of effect (0.256) similar to the previous model
(1-3A, Step 3). The effect size of density gradient is consistent with the previous model,
whereas that of peak density increased substantially (e.g. from 0.259 in Model 1-3A Step 3
to 0.581 in Model 1-3L Step 3). This result shows that the decentralization of large
warehouses is certainly influenced by the high land prices in the urban core, which pushes
large facilities to be established on the outskirts.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
68

Table 20 Results of model estimation: Model 1L and 2L (large warehouses)
W&D
Decentralization
2003-2013
Model 1-3L
Step 1

Model 1-3L
Step 2

Model 1-3L
Step 3

Std. β Sig. Std. β Sig. Std. β Sig.
Freight flow 0.402 ** 0.303 + -0.156
∆Large W&Ds 0.165 0.256 +
Density gradient -0.566 *
Peak density 0.581 *
Constant . . .
R
2
0.162 0.179 0.381
N 24 24 24
(** P < 0.01; * P < 0.05; + P < 0.1)

5 CONCLUSIONS AND FUTURE RESEARCH
Several studies have examined warehousing decentralization in urban areas worldwide.
Only a handful of US metro areas have been explored separately, yet few studies have
tested such location changes at the national level. So far, none has empirically tested what
has led to their relocation. The current understanding is that the recent restructuring of
globalized supply chains has resulted in increased demand for larger and automated
warehouses. To fulfill this demand, many facilities have been built on the urban periphery;
hence, the decentralization of warehousing. However, because not every metro area
participates in global supply chains, decentralization has not occurred everywhere.
Dablanc et al. (2014) theorize that decentralization is only a problem of very large
metropolitan areas. In this research, I have hypothesized that the variation in
characteristics across metro areas has resulted in differences in the extent of
decentralization.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
69

I focused on two factors: 1) variation in land price distribution and 2) variation in
freight demand. These two factors are exogenous, yet they both influence decentralization
at the same time. Because there are no priors regarding this phenomenon, I began with
descriptive analyses and hypothesis testing of warehousing distribution and
decentralization in 48 US metro areas. ZIP Code-level datasets of 2003 and 2013 were used.
I drew the following observations. First, over this ten-year period, warehouses did
decentralize at the national level. The extent of decentralization in large metro areas (size
rank #1-22) was not significantly greater than that in smaller ones (size rank #23-48).
Second, significant decentralization occurred in seven metro areas – Los Angeles, Chicago,
Atlanta, Detroit, Milwaukee, Las Vegas, and New Orleans. Third, the correlation between
decentralization and metro size, a proxy for the land price, was non-linear. With the rank
order from large to small metro areas, the extent of decentralization decreased linearly
until the population reached approximately 2.2 million. Beyond this point, the relationship
was not significant. Fourth, the correlation between freight flow and decentralization was
similarly non-linear. Fifth, nationally, large warehouses (100 or more employees)
decentralized more than small ones (fewer than 100 employees) did. Also, large
warehouses decentralized to a greater degree in large metro areas than they did in small
metro areas. However, this pattern was not documented in the case of small warehouses.
Lastly, large warehouses decentralized significantly in Los Angeles, Chicago, Boston, and
Phoenix.
I further evaluated the linear relationship between decentralization and two
explanatory measures – freight demand (freight flow and change in large warehouses) and
land price (density gradient and peak density) – at the metropolitan level with stepwise OLS
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
70

models. Decentralization of large warehouses was also evaluated separately. I summarize
the results as follows. Most importantly, descriptive scatter plots, correlation statistics, and
the significance of estimated parameters with dummy interactions, all of these factors,
suggest that the effect of freight demand and land price on decentralization is indeed
significant and nonlinear. Furthermore, density gradient across large metro areas,
controlling for all other factors, had the greatest impact on decentralization. Peak density
and changes in large W&Ds had almost half of this effect size. When the decentralization of
large warehouses was considered, density gradient and peak density were equally
influential. Lastly, decentralization in small metro areas was mainly a function of freight
flow.
The results of the hypothesis testing and linear models provide robust evidence for
the theory that freight and land demand influences decentralization. Large freight demand
increases the feasibility of scale operation and automation, which will substantially
increase the facility size and land consumption. High land prices, or the spatial
concentration of land demand, push land-intensive businesses away from central locations.
When these demands are not sufficiently high, the effect becomes insignificant. Indeed,
decentralization is, as Dablanc et al. (2014) stated, linked with very large metro areas.
However, it is also associated with very large warehouses.
Conversely, the results of the decentralization in small metro areas are vague. There
may be some unobserved effects other than freight flow, and I suspect two location-specific
factors: (1) land use/zoning regulations and (2) proximity to intermodal terminals (e.g.
airports, seaports, and railways). Depending on the facility function, logistics operators
will prioritize different location factors, because different warehouses play different roles
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
71

in a supply chain. For example, a regional distribution center of a global freight distributor
would prioritize access to cargo service airports, railways, and highways much more than
access to the urban center, whereas a fulfillment center of an online shopping company
would prioritize instant access to their local market. Thus, employment might not be the
best proxy for facility size, even though the size division of warehouses was made based on
the currently available information. The facility location choice logic will likely differ,
hence their location patterns.
These findings cannot provide a definitive answer to the inquiry on whether
warehousing decentralization will lead to more truck miles traveled. Controlling for the
operational side of the logistics industry, large warehouses’ relocation to the periphery
should result in more travel. However, this assumption hardly holds, because the
relocation has been fundamentally based on restructuring of the logistics operation. As
discussed in Kohn and Brodin (2008), the advancement in logistics technology should have
substantially changed how a warehousing facility is utilized and how freight is transported.
If so, an evaluation at the sub-metropolitan level of the introduction of state-of-the-art
warehouses and the changes in the geography of urban freight movement may be the
natural choice for future research.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
72

CHAPTER 4
WAREHOUSE LOCATION CHOICE: A CASE STUDY IN LOS ANGELES, CA

ABSTRACT
During the last decade, the logistics industry has drastically altered how goods are
transported. It also has reprioritized from storage to throughput in order to move large
volumes of goods more frequently and reliably. Concurrently, the warehousing industry
has significantly expanded and has built large and automated facilities on the periphery of
metropolitan areas. Researchers have been interested in this spatial shift because it
implies more truck travel and associated negative externalities. Decentralization is
attributed to inventory and transport cost tradeoffs, in which the gains of low land prices
and lower per-unit inventory costs from economies of scale outweigh the increase in
transport costs that result from locating further from urban markets. As the firm location
literature suggests, the characteristics of a firm’s demand (size of a warehouse) and those
of the chosen location (land price, access to market, labor and trade nodes) influence the
firm’s profit structure and eventually the probability that a location is selected. I
hypothesize that (a) the logic of location choice varies with respect to the facility size and
(b) the logic has changed over time. I evaluate the location choice of 5,364 warehousing
facilities built between 1951 and 2016 in the Greater Los Angeles area. I use the
framework of firm location choice and estimate a discrete choice model with facility and
location attributes as independent variables. Results suggest significant differences in the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
73

effect of location choice factors over facility size and over time. For warehouses built
before 1980, the most influential factors affecting location choice are local market, labor,
seaport/intermodal terminal proximity. In contrast, lower land price and
airport/intermodal terminal proximity have the greatest effects for warehouses built after
2000. Thus, we observe recent large-scale developments appearing in San Bernardino and
Riverside counties, much further from the urban core of Los Angeles.

1 INTRODUCTION
The purpose of this research is to understand how and why warehouses have decentralized
over time from central urban areas to the periphery by examining the location choice of
warehouse owners/developers (Dablanc and Ross, 2012). The location of a warehouse, as
part of a supply chain, is strategically chosen based on “productivity enhancing location
attributes” (Sivitanidou, 1996, pp. 1262). A change in where warehouses are being built
suggests that profit maximizing location attributes have changed. For instance, as a new
facility is relocated farther from the urban center, the gains of lower land prices and lower
per-unit inventory costs from economies of scale and automation offset the increase in
transport costs. In this way, logistics operators may internalize the cost savings of facility
relocation, but any increased negative impacts from more truck travel will be incurred by
the society as a whole.
This paper evaluates trends in the location choice of existing warehousing facilities
that were built between 1951 and 2016 in the Greater Los Angeles region. I use the
conceptual framework of firm location choice and estimate a discrete choice model with
facility and location attributes as independent variables. Results suggest significant
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
74

differences in warehouse location and location choice factors over time. We observe recent
large-scale developments appearing in San Bernardino and Riverside counties, much
further from the urban core. Warehouses built after 2000 have prioritized lower land price
and airport/intermodal terminal proximity over local market access, a significant shift
from location choice factors for warehouses built before 1980.
This paper is organized as follows: First, I review the recent literature on
warehousing decentralization and location choice determinants. Section 2 presents the
conceptual approach and econometric models. Section 3 presents data, and Section 4
shows the results of econometric models. In Section 5, this paper closes with conclusions
and discussion.

1.1 Warehousing Decentralization
Over the last two to three decades, the logistics industry has been restructured to transport
large volumes of goods through globalized systems of supply chains and just-in-time
operations (Cidell, 2011; Dablanc et al., 2014; Rodrigue, 2008). This restructuring has been
attributed to advances in information, communication, and logistics technologies; decrease
in freight transport costs; and outsourcing of the logistics functions (third party logistics;
3PLs), through which production and transport facilities of supply chains have been
systematically integrated as well as geographically fragmented (Hesse and Rodrigue, 2004;
Glaeser and Kohlhase, 2004). The scale and location of these logistics facilities have been
strategically optimized for maximum productivity. This first resulted in the increase in the
use of centralized, large-scale, hub distribution centers (Bowen, 2008; Cidell, 2011;
Rodrigue, 2008; Dablanc & Ross, 2012), which prioritize direct access to globalized
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
75

networks of supply chains and consumption markets (Dablanc et al., 2014; Movahedi et al.,
2009). This optimization also resulted in the recent development of the hub distribution
centers in peripheral locations, which provided low cost land, access to congestion-free
highways and railyards, and a supportive business environment for logistics operations
(Bowen, 2008; Christopherson and Belzer, 2009; Cidell 2010 & 2011; Dablanc and Ross,
2012; Hesse, 2002, 2004, & 2007; Hesse & Rodrigue, 2004; McKinnon, 2009; Notteboom
and Rodrigue, 2005; Rodrigue, 2006; Slack, 1998). On the contrary, traditional industrial
zones and locations near major trade nodes (e.g. airports and seaports), in many cases of
large metro areas, are close to central urban areas, which may not be desirable for large-
scale logistics operations because of high land prices, congestion, and stringent land use
regulations. Thus, decentralization was a natural choice.
Decentralization has been examined in several metropolitan areas around the world,
because of concerns that decentralization may result in increased freight travel distance
and associated negative externalities (e.g. increased fuel consumption, greenhouse gas and
criteria emissions, congestion, noise, infrastructure damage, and environmental justice)
(Allen, Browne & Cherrett, 2012; Bowen, 2008; Cidell, 2010; Dablanc & Ross, 2012;
Dablanc & Rakotonarivo, 2010; Dablanc et al., 2014; Giuliano and Kang, 2016; Heitz and
Dablanc, 2015; Sakai et al., 2015; Van den Heuvel, et al., 2013; Woudsma et al., 2016).
Because of the lack of shipment data, most studies have drawn implications on the travel
distance from the changes in the spatial distribution of logistics facilities, calculated based
on the Euclidean distance to each facility from a fixed location, such as the geographic
centroid of the logistics facilities or the central business district of a metro area. Results of
the spatial analysis are mixed. Decentralization over the last decade has been documented
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
76

in multiple metro areas with large populations and a high volume of freight trade (e.g.
Atlanta, Los Angeles, Paris, Tokyo, and Toronto), but in several other metro areas (e.g.
Seattle, Sacramento, and San Diego), there was no significant change in spatial distribution
(Dablanc and Ross, 2012; Dablanc et al., 2014; Sakai et al., 2015; Woudsma et al., 2016).
The change in spatial distribution of logistics facilities within a metro area at the
county level has also been measured. Results are mixed; the level of concentration has not
only increased in suburban counties with rich infrastructures of airports and highways but
also in the central counties with growing populations and intensive freight sector
businesses (Bowen, 2008; Cidell, 2010). So far, only two studies have documented
increased freight vehicle miles traveled (VMT), because of W&D decentralization (Dablanc
and Rakotonarivo, 2010; Sakai et al., 2015). The two studies are based on facility-level
shipment surveys in Paris and Tokyo. However, owing to a variety of operational aspects
to consider at the facility level, the changes in freight travel distance have not been
accurately estimated (Sakai et al., 2015).

1.2 Location Determinants for Warehousing Facilities
Firms’ location decisions have been widely studied in the fields of economics, urban
planning, and operations research. I provide a brief overview of the firm location literature
in which transport and land costs play a significant role when firms make location
decisions. Initially, Weber (1929) proposed that a firm will choose the optimal location
where transport costs between resource suppliers and consumer markets are minimized.
Later, Alonso (1967) introduced factor substitution and modified the theory that the
optimal location is the one that maximizes profits. He also added the concept of
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
77

willingness-to-pay of firms for a location (bid-rent) that firms compete for locations with
different levels of market and labor access. At the same time, the development in
econometric modeling of disaggregate firm location choices allowed researchers to
quantify the effect of a firm’s characteristics (who makes location decisions) and that of the
location choice alternatives separately (McFadden, 1974; 2001; Arauzo, 2010). Recently,
new economic geography literature considered the interaction between agglomeration
economies and transport costs that similar firms tend to agglomerate to decrease costs by
resource sharing and knowledge spillover (Krugman, 1991). In sum, a firm’s optimal
location is where profit per unit output is maximized, considering the tradeoffs between
transport costs to market and labor as well as land costs, the presence of similar firms, and
other intermediate inputs.
Relative to commercial or industrial firms, the location decisions of the logistics
industry have received far less attention (Dablanc et al., 2014). However, the general
location choice rationale is applicable to logistics facilities as well in that location choice
decisions include productivity maximizing location attributes (Sivitanidou, 1996). One of
the ways to evaluate the location factors is to assess the systematic components of
warehousing rent: location and building characteristics. Two examples are a study in Los
Angeles, CA that evaluated location factors for warehousing facilities (land supply,
submarket occupancy, access to local markets, transport nodes, and labor) and a study in
Dallas, TX that analyzed logistics building characteristics (ceiling heights, building age, and
number of doors and docks) (Sivitanidou, 1996; Buttimer, 1997). With respect to the
facilities of the logistics and linked industry sectors (e.g. wholesale and transportation), the
importance of access to transportation infrastructure has been stressed in several
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
78

empirical studies in Chicago, IL, Calgary in Canada, Strathclyde in Scotland, and some
regions of France (Kawamura, 2001; Woudsma et al., 2008; Button et al., 1995; and Masson
and Petiot, 2015, respectively).
The effect of agglomeration economies on the clustering of logistics firms is unclear.
Masson and Petiot (2015) in a French case study documented that neither urbanization
economies (the effect of total economic activity) nor localization economies of linked
industry businesses (the effect of transportation sector activity) provided productivity
gains of warehousing firms. However, the clustering of warehousing facilities has been
observed in several studies. Accordingly, the spatial autocorrelation has reached a
significant level in a wholesale firm location study using the 2004 goods movement survey
in the Tokyo metro area in Japan (Nguyen and Sano, 2010) and a study of logistics parcel
development from 1996 to 2000 in Calgary, Canada (Woudsma et al., 2008). I suspect that
the clustering of logistics facilities might not be because of agglomeration benefits but
because of land use control measures.
Another difference in logistics facilities’ location patterns is that they require less
access to local markets than retail firms and are comparatively less-capital intensive than
manufacturing facilities, which are also part of the supply chain (Jakubicek and Woudsma,
2011; Warffemius, 2007). Furthermore, not only has the outsourcing of logistics functions
(third-party logistics; 3PLs) shortened the planning horizon of a supply chain from decades
to 3-5 years, but it also has changed the ownership structure from owning to leasing a
facility (Supply Chain Brain, 2010). Because of these factors, logistics firms could respond
swiftly and flexibly to internal and external demands to relocate, restructure, or build new
facilities (Jakubicek and Woudsma, 2011). Several studies based on stakeholder interviews
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
79

confirmed that logistics operators seek location attributes consistent with the previous
discussion: low land costs, proximity to highways and airports, access to labor, access to
markets, and physical and regulatory environments (Jakubicek and Woudsma, 2011;
ITrans Consulting, 2004; and Holl, 2004). In summary, location choice schemes for logistics
facilities are overall consistent with other industry sectors in that the goal is profit
maximization. The differences are 1) access to transportation infrastructure is critical,
while access to consumption markets is not as important as retail firms would demand; 2)
benefits of agglomeration are not clear; 3) ownership structure is flexible; and 4) the
outsourcing of logistics functions has shortened the planning horizon.
The research gap is large. First, there are only a small number of empirical studies
on the location choice of logistics facilities. As discussed, their results are consistent with
the theoretical understanding. However, empirical testing of how location choices might
differ by facility type is scarce. This might be partly attributed to data scarcity on logistics
facilities’ physical characteristics (e.g. building floor area, ceiling height, and number of
docks) and operations (e.g. volume of freight movement and origins/destinations of
commodities). As a proxy for the size of a facility, two studies used the number of
employees to separate large warehouses from others (Bowen, 2008; Andreoli et al., 2010).
Furthermore, when decentralization is concerned, no study has evaluated changes in
facility location, in conjunction with changes in location choice determinants. In other
words, facility location has been quantified in more than two time periods in several urban
areas to see if the location change over time was statistically significant. However, no study
has tested the change in location and associated location attributes to examine whether
decentralizing facilities have sought a specific location attribute or if any tradeoffs among
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
80

location attributes have occurred. With this paper, I aim to understand if warehousing
facility characteristics have changed over time; if location varies by facility type; if location
choice has changed over time; and if the change in location is correlated with a specific
location attributes.

2 RESEARCH APPROACH
2.1 Conceptual Framework
The literature on warehouse location and decentralization suggests that the processes of
logistics restructuring have led to the establishment of large distribution centers in the
periphery of urban areas. The consolidation of inventory via large-scale distribution
centers, rather than utilizing dispersed conventional warehouses, optimizes supply flows,
enhances economies of scale, increases throughput, and consequently reduces inventory
and costs (Hesse, 2004; Hesse, 2007; McKinnon, 2009). Large facility size and freight flows
require a specific location, in which its location characteristics (e.g. low land rent, land
availability, and highway access) satisfy logistics objectives. In sum, through logistics
restructuring, the characteristics of warehousing facilities changed, specific location
requirements arose, and hence facility location has changed to fulfill the new operation
goals (Hesse, 2007).
To empirically evaluate this transition in facility size and location, I adopt the
conceptual framework of firm location choice. The characteristics of a warehousing facility
that a logistics firm or developer demands, in conjunction with those of a chosen location,
constitute an unobservable structure of business costs, which, in turn, influence the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
81

probability that the firm establishes a warehousing facility on the location (McFadden,
1974; Carlton, 1979, 1983; Guimarães et al., 2004; Arauzo et al., 2010). Likewise, if the
characteristics of a facility are different, the facility’s location choice will also differ. In the
case of W&D decentralization, we observed both the change in facility characteristics
(larger facility size; consolidated inventory) and the change in location decision (urban
outskirts). These shifts have occurred as cost tradeoffs, in which the cost benefits of lower
land prices, scale economies, and facility centralization outweigh the increase in transport
costs (McKinnon, 1998; McKinnon, 2009; Hesse, 2004; Hesse, 2007).
Here I elaborate on the relationship between the change in facility characteristics
and the change in location choice. I first consider increase in facility size. Conventionally,
warehousing facilities have been concentrated in the hinterland of the port complexes and
old industrial zones in the central urban area because of high demand for warehousing
capacity; relatively low costs of transporting intermediate/final goods from trade nodes;
and available land use and zoning for warehousing businesses. However, despite the path
dependence (i.e. sunk resources, such as infrastructure, business environment, and
established land use/zoning policies), which makes it difficult for such businesses to
relocate, growing metropolitan areas and subsequently increasing land values have slowly
priced out the land intensive warehousing industry. Moreover, with a significant increase
in land consumption and freight flows through logistics restructuring, the location for large
distribution centers has changed to the urban periphery, in which relatively low-cost land
is available (Hesse, 2007).
The second factor to consider is time. Over the last several decades, advances in the
ICT and logistics technology have been substantial, and they have been the main driver of
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
82

logistics restructuring. Namely, warehouses built recently will differ from those built in the
1980s in terms of building design, stock layout, and inventory management. Because
technology progresses with time, I use the built year of a facility to distinguish facilities
equipped with recent technologies from the conventional, past ones. As discussed, with
respect to the time periods, the cost structure will differ, hence location decisions will vary.
Third, I consider location characteristics. The literature suggests that important
location attributes for a warehouse location decision are land prices and access to labor,
consumer markets, and trade nodes. As discussed in the urban economics literature,
population and employment densities are used as a proxy for land price (Clark, 1951;
McDonald, 1989). Labor force access is quantified as the sum of population within a
distance threshold with an inverse-distance weight. The distribution of land prices and
labor force access is a function of the distribution of population or employment. Access to
consumer markets and trade nodes is quantified as a proximity measure with respect to
travel time. As consumer markets, I use the definition evaluated in Giuliano et al. (2007).
As trade nodes, I include seaport, airport, intermodal terminal and highway. Namely, a
location entails a unique combination of location attributes.
In summary, restructuring of the logistics industry has resulted in the changes in the
facility characteristics, rebalance of logistics costs, as well as facility location. Thus, I
hypothesize that the characteristics of a firm’s demand for a facility (size) and the
characteristics of a location (land price, proximity to the nearest local market, proximity to
the nearest trade node, and labor force access) jointly influence the cost structure and
ultimately the probability that a given location is chosen and that the logic of location
choice has changed over time. A general probability function is:
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
83

𝑝 𝑛𝑗
= Pr(𝑦 𝑛 = 𝑗 ) = 𝐹 (𝑋 𝑗 , 𝑊 𝑛 )

Where,
p is the probability of facility n to be established in location j, which is a function of the location
characteristics (X) and the facility characteristics (W).

I estimate the location choice in multiple time periods separately and formally
compare estimated parameters across the periods. I expect that parameters of the location
attributes differ significantly by facility size and over time.

2.2 Discrete Choice Model
I use the framework of firm location choice and estimate a discrete choice model (DCM)
with facility and location attributes as independent variables. The assumptions of the DCM
are that a logistics firm makes a warehousing facility’s (n = 1, …, N) location decision among
a choice set of J sites. The choice of a site, j = 1, …, J, entails an unobservable profit, ∏
𝑛𝑗
,
meanwhile the choice of site j over site i is made if/only if ∏
𝑛𝑗
> ∏
𝑛𝑖
(j i). The profit (∏) is
decomposable into a systematic component (𝜋 𝑛𝑗
) and a random component (𝜀 𝑛𝑗
) – an error
term assumed to be i.i.d. The systematic component is a linear function,

𝜋 𝑛𝑗
= 𝜋 (𝑥 𝑗 , 𝑤 𝑛 ) = 𝜃 ′𝑍 𝑛𝑗

where,
𝑥 𝑗 is the characteristics of a site j; 𝑤 𝑛 is the characteristics of a facility n; 𝜋 𝑛𝑗
is assumed to be linear in
parameters (𝜃 ′𝑍 𝑛𝑗
), where θ is a parameter vector to be estimated and 𝑍 𝑛𝑗
= {𝑥 𝑗 , 𝑤 𝑛 }.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
84

Thus, the probability of locating a facility, n, at a site location, j, is

𝑃 𝑛 (𝑗 ) = 𝑒𝑥𝑝 𝜃 ′𝑍 𝑛𝑗
∑ 𝑒𝑥𝑝 𝜃 ′𝑍 𝑛𝑗
𝑗 ⁄

where 𝜃 is estimated by maximum likelihood.
In sum, the discrete choice model estimates, controlling for all other factors, how the
differences in the variables in the systematic component of profit 𝜋 𝑛𝑗
influence the
probability (𝑃 𝑛 (𝑗 )) that the location decision at j of a facility n is made (McFadden, 1974).
One of the principal assumptions incorporated in the DCM is Independence of
Irrelevant Alternatives (IIA), which “imposes a uniform pattern of substitution between
locations ‘in that one cannot postulate a pattern of differential substitutability and
complementarity between alternatives’” (Arauzo et al. 2010, pp. 697; McFadden, 1974, pp.
112). In other words, it assumes that the inclusion or exclusion of a choice within the set
does not influence the relative probability linked to the explanatory variables in the
remaining choices within the set. IIA is a strong assumption due to unmeasured factors,
which are correlated with the characteristics of location choice sets and therefore influence
firm profits and ultimately location choices (Bartik, 1985, pp. 16). On the other hand,
Carlton (1983) theorizes that because the geographic distance among choice sets in typical
firm location studies (e.g. municipality, counties, and states) is large, the IIA assumption is
not completely implausible. However, no empirical studies using multinomial/conditional
logit models could verify this claim (Arauzo et al., 2010). There are a few approaches to
mitigate this problem. One can use the nested logit model, in which the choice sets are
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
85

divided into non-overlapping ‘nests’ that prevent varying substitutability or
complementarity between alternatives across different nests (Hensher and Greene, 2002).
Another approach is to aggregate individual location choices to a designated number of
discrete location choice sets in a way that heterogeneity between the aggregated sets is far
greater than that within a set (Arauzo, 2004).
Another common problem in the DCM is that the choice set is too large. In the case
of sub-metropolitan firm location choice, where the unit of analysis is a census tract, the
number of a complete choice set could become in the order of hundreds. Not only does it
require infeasible data processing but it also increases the likelihood that IIA is violated
because the unobserved attributes of locations in the same neighborhood are likely to be
similar (McFadden, 1978). The methodology that I use to mitigate this problem is the two-
stage discrete choice approach, which has been frequently used in travel destination choice
studies (McFadden, 1978; Manski, 1977; Swait and Ben-Akiva, 1987; Ben-Akiva and
Boccara, 1995; Pellegrini et al., 1997; Scott and He, 2012). In the first stage, from the
universal choice set (G), a subset of choice alternatives (C) is formulated (probability that a
firm n generates a choice set of C: P
𝑛 (𝐶 )). Subsequently, in the second stage, conditional on
the formulated choice set (C), an actual choice of an alternative j is made (probability that a
firm makes a choice at j: P
𝑛 (𝑗 |𝐶 )). The general model is:

𝑃 𝑛 (𝑗 ) = ∑ 𝑃 𝑛 (𝑗 |𝐶 )𝑃 𝑛 (𝐶 )
𝐶 ∈𝐺

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
86

The actual choice alternatives (C) are unknown; we only observe the chosen alternative (j).
With the IIA assumption, parameters can be consistently estimated using only a subset (C)
of the alternatives from the universal choice set (G) (McFadden, 1978; Pozsgay and Bhat,
2002; Scott and He, 2012). Thus, for each of the chosen alternative, I assign nine randomly
selected choice alternatives to formulate a choice set of ten alternatives. For estimation, I
use a multinomial logit model.
In summary, I estimate a multinomial logit model of warehouse location choice by
multiple time periods as a function of location and facility attributes. Location attributes
include land prices as well as measure of labor force, local market, and trade node access.
A facility attribute (building size) is included as an interaction term to identify varying
effects of the location attributes by facility size on location choice. To evaluate changes in
the logic of location choice over time, I formally compare estimated parameters across the
models of different time periods (Allison, 1999).

3 DATA
This paper is a case study of the recent warehousing location choice in the Los Angeles, CA
combined statistical area (CSA).
23
Consisting of five counties (Los Angeles, Orange,
Riverside, San Bernardino, and Ventura), it is the second largest CSA in the US with 18
million population and 7 million employment in 34,000 mi
2
in 2010. First, the region is
known for its polycentric urban structure with dozens of employment clusters along the

23
Formally entitled, the Los Angeles-Long Beach, CA Combined Statistical Area (CSA) consists of three
Metropolitan Statistical Areas (MSAs) – Los Angeles-Long Beach-Anaheim, CA MSA; Riverside-San
Bernardino-Ontario, CA MSA; Oxnard-Thousand Oaks-Ventura, CA MSA.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
87

Wilshire corridor and major highway networks (Giuliano and Small, 1991; Agarwal et al.,
2012). The two central counties (Los Angeles and Orange counties combined) have the
highest average population density in the US (2,650 population/mile
2
).
24
On the other
hand, the two peripheral counties (Riverside and San Bernardino counties, commonly
called “the Inland Empire,” an area with intense warehousing activity of the recent decades)
have a vast supply of relatively low-cost land (27,000 mi
2
), in which rich transportation
infrastructure (airports, railroads, rail-to-truck intermodal terminals, and highways)
supports the region’s productive freight transportation activity.
Second, the Los Angeles CSA is one of the largest international trade gateways in the
US. Its seaport complex (Port of Los Angeles and Port of Long Beach) handles the largest
portion (37%) of all container shipment (TEU) in the US (Strocko et al., 2013), and the air
cargo volume of its three cargo service airports (Los Angeles International, LAX; Ontario
International, ONT; and Long Beach Airport, LGB) is ranked seventh in the US (FAA, 2015).
There are also seven rail-to-truck intermodal terminals for the inter-continental rail system
operated by Union Pacific and BNSF. Due to these intense freight functions, this region has
had an active, expanding warehousing sector. The expansion from 2003 and 2013 has been
substantial (29% increase in establishment and 43% increase in employment), and facility
distribution decentralized significantly (Giuliano and Kang, 2016). In sum, these
characteristics make the Los Angeles CSA an interesting and appropriate place to test
location choice factors.

24
All statistics are from the 2010 US Census and 2010 LEHD.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
88

3.1 Warehouse Data and Sample Distribution
The warehousing location data are from CoStar, a real estate database which includes
commercial and industrial real estate listings. The database provides facility characteristics,
such as rentable building area (RBA), year of construction, number of loading docks, and
number of floors. It also provides each facility’s address, by which the exact location and
the location’s characteristics can be identified. Because facility and location characteristics
are available at the facility level, researchers can evaluate their spatial distribution and
location choice factors. In this paper, I use 5,364 listings of logistics facilities, which are
classified as warehouses, truck terminals, distribution centers, or cold storage facilities.
Similar facilities coded under manufacturing or retail trade sectors are not included. The
minimum rentable building area (RBA) is set as 30,000 ft
2
. The built year ranges from
1951 to 2016. The listings data were collected in early 2016.
The CoStar database has an important limitation. The primary purpose of the
database is to provide real estate agents with up-to-date property listings. Hence, the
dataset used in this paper includes only those facilities active at the time of data retrieval in
early 2016. CoStar does not keep records of facilities that are currently unavailable. For
example, if a facility is demolished and has thus left the market, there is no record of it in
following years. Prior year CoStar data is not available, hence there is no way to construct
an historical inventory of warehousing.
Because of this limitation, I conduct a cross-sectional evaluation of the location
choices of existing warehousing facilities (early 2016). Warehouses are profit driven
entities of continuous business decision making, and thus, we may assume that the current
location approximates their optimal location. If a facility no longer fits the logistics strategy
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
89

of the firm, the firm will move to another more suitable facility. Therefore, the location
attributes of 2016 will be most relevant to the current distribution of warehousing facilities.
Figure 12 presents the spatial distribution of the facilities in the Los Angeles region.
Warehouses are relatively concentrated in the areas adjacent to the Port of Los Angeles
(POLA) and Port of Long Beach (POLB) complex, old industrial areas near the central
business district of Los Angeles, the Inland Empire areas (Ontario-Riverside-San
Bernardino), and multiple distant locations along the highway network (Oxnard, Santa
Clarita, Moreno Valley, and Palm Springs).

Figure 12 Spatial distribution of 5,364 warehouses and trade nodes

As an independent variable, I use an interaction dummy of facility size. The size, as
rentable building area (RBA), ranges from 30,000 to 1,800,000 square feet; I divide into
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
90

three categories: small (30-100k), medium (100-300k), and large (over 300k). The
division is arbitrarily made based on the sample distribution (mean: 125,000 ft
2
; median
70,000 ft
2
). I also divide the sample into three time periods using the facility built year:
stage 1 (1951-1980), stage 2 (1981-2000), and stage 3 (2001-2016). I also make this
division arbitrarily based on the sample distribution (mean 1985; median 1986). The
sample distribution is presented in Table 21. With respect to the size, most facilities (65%)
are under 100,000 ft
2
, and 9% are larger than 300,000 ft
2
. With respect to the built year,
approximately 79% were built before 2000. As expected, there is some positive correlation
between the size and built year. The smallest warehouses were mostly built in the earlier
periods, while the largest were mostly built after 2000.
Figures 13 and 14 present the percent distribution of warehouses by built year and
building area, respectively, with respect to the Euclidean distance from the central business
district of Los Angeles. The curve reaches up to 130 miles from the CBD. In both figures,
three locations with a high level of concentration are documented at approximately 20
mile-, 40 mile-, and 60 mile-distance from the CBD. They correspond to the clustering of
warehouses shown in Figure 13: Downtown LA-Norwalk-West Covina (hinterland of the
port complex and old industrial areas) around the 20-mile band; b) Ontario (the Inland
Empire) around the 40-mile band; and San Bernardino-Riverside-Moreno Valley around
the 60-mile band. Figure 14 shows that percent distributions among the warehouses built
in the earlier, mid, and later periods differ significantly, and it supports the hypothesis that
warehouses have been built further away from the CBD over time in Los Angeles. Figure
14 shows that the percent distribution of the largest warehouses is distinguished from all
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
91

others, which also supports the hypothesis that large warehouses have been established on
the urban periphery.

Table 21 Distribution and share of W&D entries by built year and size
Built year
Size (ft
2
) 1951-1980 1981-2000 2001-2016 Sum
30-100k
1,604
(29.9%)
1,311
(24.4%)
578
(10.8%)
3,493
(65.1%)
100-300k
528
(9.8%)
554
(10.3%)
309
(5.8%)
1,391
(25.9%)
Over 300k
61
(1.1%)
157
(2.9%)
262
(4.9%)
480
(8.9%)
Sum
2,193
(40.9%)
2,022
(37.7%)
1,149
(21.4%)
5,364
(100%)

Figure 13 Percent distribution of warehouses from the Los Angeles CBD by built year

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
92

Figure 14 Percent distribution of warehouses from the Los Angeles CBD by facility size

3.2 Choice Set Design and Location Attributes
There are 3,920 census tracts in the Los Angeles CSA, but only 660 have at least one
warehousing facility. From this universal choice set (G) of 660 census tracts, I draw nine
random choice alternatives per chosen alternative to form a choice set of ten (5,364 x 10
alternatives = 53,640 observations). I use a typical number of choice set alternatives (10),
because the size of a choice set does not influence estimation results (McConnell and Tseng,
2000; Narella and Bhat, 2004). Because there is no prior research regarding whether to
constrain choices for warehousing location decision in the choice set generation, I do not
constrain choices but draw random alternatives from the universal choice set (G).
The location attributes used in this analysis are: land prices (population and
employment densities), proximity to the nearest local market (as an employment
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
93

subcenter), proximity to the nearest trade node (airport, seaport, intermodal terminal, and
highway exit), and labor force access. First, I use population and employment densities in
2010, as a proxy for land prices (Clark, 1951; McDonald, 1989). Second, the proximity to
the nearest local market is calculated based on the travel time to the nearest employment
subcenter, which is to account for Los Angeles’ polycentric urban structure. As described
in Giuliano et al. (2007), the 20/20 employment subcenter definition is used: (a) the
employment density of each census tract in each subcenter exceeds the density threshold
(20 jobs per acre); (b) neighboring census tracts that share a common edge form a
subcenter; and (c) the sum of employment of each subcenter exceeds the size threshold
(20,000 jobs). Using the 2010 LEHD dataset at the census tract level, twenty subcenters
were identified, in which 1.3 million jobs (18.6%) are concentrated in 0.1% of the total land
area.
25
The travel time (PM peak) from the centroid of each census tract to that of the
nearest employment subcenter is calculated from the SCAG 2012 Regional Transportation
Plan loaded network database using Network Analyst in ArcGIS 10.4.1.
26
Third, the
proximity to the nearest trade node is the travel time (PM peak) from each census tract
centroid to the nearest trade node. I used three airports (LAX, ONT, LGB), two seaports
(POLA/POLB, as one complex), and seven rail-to-truck intermodal terminals operated by
UP and BNSF; and calculated the travel time by mode. I also included the Euclidean
distance to major highway ramps (interstate highway, US highway, and state highway).
Figure 15 presents the distribution of employment subcenters and all trade nodes.

25
Longitudinal Employer and Household Survey (LEHD), published annually by the Census Bureau
26
The Southern California Associate of Governments (SCAG) serves six counties in Southern California (Los
Angeles, Orange, San Bernardino, Riverside, Ventura, and Imperial) as the metropolitan planning organization
(MPO).
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
94

Figure 15 Distribution of employment subcenters and trade nodes (close-up of the region)

Fourth, labor force access is calculated as the sum of population with an inverse
travel-time weight. The travel time was limited to less than 30 minutes between two
census tracts’ centroids. The mathematical formula to calculate labor force access is:

𝐿𝑎𝑏𝑜𝑟 𝑓𝑜𝑟𝑐𝑒 𝑎𝑐𝑐𝑒𝑠𝑠 , 𝐿 𝑖 = ∑ 𝑃𝑂𝑃 𝑗 ×𝑡 𝑖𝑗
−0.6769
𝑛 𝑗 =1

where,
POPj = population in census tract j;
tij = travel time between census tract i and j (t < 30 minutes).

Table 22 presents summary statistics of location attributes. To account for the
skewed distribution, all variables are used in natural logarithm.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
95

Table 22 Summary statistics of location attributes (N=660)
Variable Mean Median SD
Population density (ppl/mile
2
) 6,865 5,647 5,886
Employment density (job/mile
2
) 3,912 2,725 5,443
Labor force access (workforce) 838,892 882,430 436,037
Proximity to employment subcenter (min) 15.4 11.8 14.1
Proximity to airport (min) 21.9 18.2 16.2
Proximity to seaport (min) 40.3 37.4 22.6
Proximity to intermodal terminal (min) 20.7 15.8 16.4
Proximity to highway (mile) 1.0 0.7 1.2

Table 23 presents the summary of variable definitions.

Table 23 Summary of variable definitions
Variable Definition
Dependent variable

𝑃 𝑛 (𝑌 = 𝑗 ) Probability that a firm n selects an alternative j for its warehouse location
Independent variables
Facility attributes
Rentable building area
Small size 1 if RBA < 100k ft
2
, 0 otherwise (reference)
Medium size 1 if 300k ft
2
> RBA ≥ 100k ft
2
, 0 otherwise
Large size 1 if RBA ≥ 300k ft
2
, 0 otherwise
Location attributes
Land prices
Population density Population per mile
2

Employment density Employment per mile
2

Proximity measures
To local market Driving time to the nearest employment subcenter (Giuliano et al., 2007)
To airport Driving time to the nearest airport
To seaport Driving time to the nearest seaport
To intermodal terminal Driving time to the nearest intermodal terminal
To highway ramps Distance to the nearest highway ramp
Labor force access
Sum of population with an inverse distance weight within a 30-minute
driving distance; ∑ 𝑃𝑂𝑃 𝑗 ×𝑡 𝑖𝑗
−0.6769 𝑛 𝑗 =1

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
96

4 RESULTS
I estimate models of warehouse location choice for three time periods (1951-1980; 1981-
2000; 2001-2016). The first set of models excludes interaction terms (Model 1-1, 1-2, and
1-3). The second set of models includes interaction terms for facility size (Model 2-1, 2-2,
and 2-3). The interaction term identifies the varying preferences of warehouse location by
facility size. The negative intercept of all models is because of the model design (Scott and
He, 2012). The number of the chosen alternative (1) is far smaller than that of the
reference group (9 random alternatives), hence it is much less likely to be chosen. Results
for Models 1-1, 1-2, and 1-3 are given in Table 24; and Models 2-1, 2-2, and 2-3 in Table 25.
Also, included in the table are tests of whether the difference in coefficients across time
periods is significant. If the chi-square statistic (χ
2
) is greater than 3.841, the difference is
significant at P < 0.05.
I evaluate whether estimated parameters are mostly consistent with expectation
(Table 24). In Model 1-1, land price attributes work significantly in opposite directions:
population as a push factor (-) and employment as a pull factor (+). When the land price
coefficients are compared over time against those of Models 1-2 and 1-3, the effect of a
push factor has increased over time, whereas that of a pull factor has diminished. It
supports the prior discussion that increasing land values have pushed land intensive
businesses away from populous areas. In addition, the association between the general
employment density and warehouse location has been weakened, possibly because more
warehouses have been established outside of traditional industrial zones near the urban
core. The labor force access coefficient is significant, positive, and stable over time.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
97

Coefficients of all proximity measures except highway access are significant. The
trends differ from each other. A negative proximity measure represents that fewer
warehouses are likely as distance increases. Local market proximity has been positively
correlated throughout the periods (the farther, the more warehouses). Particularly,
warehouses built in the mid period (Model 1-2) are the most likely to be more distant from
an employment subcenter (a proxy for a local market). In Model 1-1, airport proximity is a
deterrent for warehouse location choice (+, the farther, the more warehouses). The sign of
airport proximity changes in Model 1-2 to negative and remain negative and significant in
Model 1-3 (the closer, the more warehouses). The opposite pattern is documented for
seaport proximity: from negative in Model 1-1 to positive in Models 1-2 and 1-3 (the
farther, the more warehouses). Intermodal proximity has been consistently negative and
significant throughout the periods, but the recently built warehouses are the most likely to
be near intermodal terminals. Highway proximity is either insignificant or positive. In sum,
when only proximity measures are compared, W&Ds in 1951-1980 prioritized seaport and
intermodal terminals; W&D in 1981-2000 prioritized airport; and W&Ds in 2001-2016
prioritized airport and intermodal terminals.
Results of Models 2-1, 2-2 and 2-3 (Table 25) with the inclusion of interaction
dummies of building size are very consistent with the previous results. To be specific, the
coefficients on the model 1 variables (without interaction) do not change with or without
interaction dummies. They also show heterogeneous preferences on warehouse location
choice by facility size. The reference is the small warehouses (< 100k ft
2
). With respect to
the population density, employment density, and labor force access, location choice is not
differentiated by facility size over the periods 1951-1980 and 1981-2000. Compared to the
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
98

reference, medium warehouses built in 2001-2016 are significantly more likely to be
influenced by the push effect of population. Contrarily, the magnitude of employment
density’s pull effect has diminished over time. Particularly, the pull effect for large
warehouses in 2001-2016 is almost canceled out (0.502 – 0.533 ≅ 0). Labor force access
has become more important for medium warehouses in 2001-2016.
Local market proximity is one of the deterrents for medium (throughout the periods)
and large warehouses (1981-2000). Possible explanations are that a regional goods
distributor would not require direct access to a local market; and that high land prices
around employment subcenters are not suitable for land intensive businesses. Seaport
proximity is also one powerful deterrent for large warehouses throughout the periods, and
the magnitude of effect increased substantially over time. It may not be because a seaport
is no longer important for the warehousing industry but because the POLA/POLB complex
is right next to central urban areas. The overall decentralizing pattern of warehouses from
the urban core may have been captured here. On the contrary, airport proximity and
intermodal proximity, especially for the large warehouses built 1981-2016, are two very
important location determinants. Again, the signs of airport proximity changes between
the first and second periods. The size of effect is the largest in 1981-2000. The
development captured in Figure 13 around the 40-mile distance band (near the Ontario
International Airport) might be captured here. Intermodal proximity is important
throughout the periods particularly for large warehouses. The size of effect is the largest in
2001-2016. As presented in Figure 15, intermodal facilities are both in traditional
industrial zones near the urban core and San Bernardino on the periphery. Hence, the
development of warehouses in both 1951-1980 and 2001-2016 periods are captured here
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
99

in the urban core and in San Bernardino, respectively. The effect of highway proximity has
not been properly captured, possibly due to lack of variation – the ubiquity of highway
infrastructure throughout the Los Angeles region.
In summary, estimated parameters of multinomial models excluding or including
interaction dummies are very consistent. Land price measures work in two directions: as a
push factor, the effect of population density has become more intensified, especially for
large warehouses, whereas, as a pull factor, the effect of employment density has
diminished substantially over time. Labor force access has been an important and stable
location factor over time. Local market proximity and seaport proximity are two significant
deterrents for warehouses built in 1981-2016. On the contrary, airport proximity and
intermodal terminal proximity are two important location determinants particularly for
large warehouses built in 1981-2016. It is notable that the difference in location choice
attributes between 1951-1980 and 1981-2000 is far greater than that between 1981-2000
and 2001-2016. Overall, the transition in the effect of location attributes on warehouse
location choice is as expected that large warehouses in 2001-2016 have been established in
locations with lower land prices, lower access to local markets, yet better access to airport
and intermodal terminals.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
100

Table 24 Estimation results of Models 1-1, 1-2, and 1-3

Model 1-1
W&D built in 1951-1980
Model 1-2
W&D built in 1981-2000
Difference
Between
Model 1-3
W&D built in 2001-2016
Difference
Between

Coef.
Std.
Err.
Sig. Coef.
Std.
Err.
Sig.
1-1 & 1-2
χ
2

Coef.
Std.
Err.
Sig.
1-2 & 1-3
χ
2

Population density -0.322 0.013 ** -0.405 0.014 ** 18.874 * -0.414 0.019 ** 0.145
Employment density 0.986 0.035 ** 0.675 0.030 ** 45.516 * 0.326 0.033 ** 61.237 *
Labor force access 0.847 0.086 ** 0.882 0.088 ** 0.081 0.728 0.110 ** 1.195
Prox. Local market 0.834 0.065 ** 1.971 0.082 ** 118.072 * 1.294 0.101 ** 27.080 *
Prox. Airport 0.834 0.073 ** -0.704 0.063 ** 254.404 * -0.585 0.082 ** 1.324
Prox. Seaport -0.459 0.060 ** 1.090 0.084 ** 225.169 * 1.988 0.126 ** 35.165 *
Prox. Intermodal -0.469 0.054 ** -0.114 0.062 + 18.643 * -0.613 0.069 ** 28.937 *
Prox. Highway -0.008 0.026

0.050 0.029 + 2.218 0.054 0.040

0.007
Constant -20.788 1.410 ** -23.409 1.593 ** 1.518 -18.987 2.011 ** 2.971
Log likelihood, Null model -7,129.1

-6,573.2

-3,735.2

Log likelihood, Full model -5,806.1

-5,327.1

-3,127.1

Pseudo-R
2
0.186

0.190

0.163

N 2,193

2,022

1,149

P (** < 0.01; * < 0.05; + < 0.1); χ
2
at P < 0.05 (1 df) = 3.841

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
101

Table 25 Estimation results of Models 2-1, 2-2, and 2-3

Model 2-1
W&D built in 1951-1980
Model 2-2
W&D built in 1981-2000
Difference
Between
Model 2-3
W&D built in 2001-2016
Difference
Between

Coef.
Std.
Err.
Sig. Coef.
Std.
Err.
Sig.
2-1 & 2-2
χ
2

Coef.
Std.
Err.
Sig.
2-2 & 2-3
χ
2

Population density -0.316 0.016 ** -0.398 0.018 ** 11.593 * -0.387 0.026 ** 0.121
Employment density 0.963 0.041 ** 0.674 0.037 ** 27.384 * 0.502 0.049 ** 7.847 *
Labor force access 0.895 0.089 ** 0.892 0.090 ** 0.001 0.725 0.118 ** 1.266
Prox. Local market 0.744 0.073 ** 1.749 0.094 ** 71.304 * 1.256 0.130 ** 9.444 *
Prox. Airport 0.726 0.080 ** -0.497 0.074 ** 125.946 * -0.430 0.106 ** 0.269
Prox. Seaport -0.466 0.070 ** 1.031 0.098 ** 154.510 * 1.604 0.156 ** 9.674 *
Prox. Intermodal -0.286 0.064 ** -0.061 0.074

5.289 * -0.278 0.102 ** 2.965
Prox. Highway -0.011 0.030

0.062 0.036 + 2.427 0.110 0.056 + 0.520
With interaction terms

Population ⨯ medium -0.027 0.032

-0.013 0.031

0.099 -0.082 0.044 + 1.643
Population ⨯ large -0.071 0.078

-0.077 0.056

0.004 -0.046 0.050

0.171
Employment ⨯ medium 0.064 0.083

0.034 0.071

0.075 -0.165 0.083 * 3.319
Employment ⨯ large 0.081 0.210

-0.034 0.123

0.223 -0.533 0.085 ** 11.139 *
Labor force ⨯ medium -0.065 0.065

-0.018 0.061

0.278 0.184 0.081 * 3.968 *
Labor force ⨯ large -0.144 0.160

-0.008 0.108

0.496 0.149 0.091

1.236
Market ⨯ medium 0.354 0.154 * 0.565 0.171 ** 0.841 0.570 0.232 * 0.000
Market ⨯ large 0.092 0.410

1.242 0.356 ** 4.486 * -0.221 0.243

11.521 *
Airport ⨯ medium 0.505 0.178 ** -0.434 0.130 ** 18.148 * -0.226 0.180 0.878
Airport ⨯ large 0.520 0.513 -1.235 0.248 ** 9.487 * -0.677 0.199 ** 3.080
Seaport ⨯ medium -0.098 0.131

-0.117 0.156

0.009 0.083 0.224

0.537
Seaport ⨯ large 0.888 0.373 * 1.271 0.320 ** 0.607 2.168 0.287 ** 4.355 *
Intermodal ⨯ medium -0.625 0.117 ** -0.072 0.129

10.083 * -0.344 0.157 * 1.792
Intermodal ⨯ large -0.760 0.319 * -0.555 0.224 * 0.277 -0.903 0.169 ** 1.538
Highway ⨯ medium 0.027 0.063

0.045 0.067

0.038 -0.098 0.092

1.579
Highway ⨯ large -0.101 0.162

-0.274 0.114 * 0.763 -0.113 0.106

1.070
Constant -21.212 1.426 ** -23.610 1.597 ** 1.254 -20.828 2.080 ** 1.125
Log likelihood, Null model -7,129.1

-6,573.2

-3,735.2

Log likelihood, Full model -5,777.7

-5,283.7

-3,038.3

Pseudo-R
2
0.190

0.196

0.187

N 2,193

2,022

1,149

P (** < 0.01; * < 0.05; + < 0.1); χ
2
at P < 0.05 (1 df) = 3.841
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
102

5 CONCLUSIONS AND DISCUSSION
I evaluated the change in the location determinants of warehouses in the Greater Los
Angeles region to understand how and why warehouses have decentralized over the last
several decades. I used the conceptual framework of firm location choice that the facility
and location characteristics jointly influence the cost structure of a logistics firm, and
consequently, the probability that the firm selects one of the location choice alternatives for
its new warehouse facility. With this structure, I tested the hypothesis that the logic of
location choice varies with respect to the facility size and that the logic has changed over
time. This hypothesis is based on the current understanding that logistics restructuring
has demanded a substantial increase in the warehousing capacity and throughput, which
has led to numerous large-scale distribution centers established on the periphery. I
focused on the variation in the facility and location attributes to explain the variation in the
probability of warehouse location choice. A unique set of nine randomly drawn location
choice alternatives has been assigned to each of the 5,364 chosen alternatives. Two
multinomial logit models (without/with building size interaction terms) in three time
periods (1951-1980; 1981-2000; 2001-2016) were estimated.
Model results consistently show that both the building size (interaction dummies)
and time (over three models) variables are significant factors to explain the variation in
warehouse location choice. The location choice of the warehouses built in 1951-1980 is
different from that of the facilities built in 2001-2016, and at the same time, the location
choice of small warehouses is distinguished from that of large distribution centers. As
Table 21 suggests, there has been a drastic change in the composition of large/small
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
103

warehouses over time. This shift was captured in the changes in coefficient signs. The
development of large warehouses in 2001-2016, relative to all other warehouses, has taken
place in locations with relatively lower land prices, lower market access, yet higher airport
and intermodal terminal proximity. This finding suggests a potential decrease in facility
and inventory costs from the lower land prices, economies of scale, and inventory
consolidation. Savings in transport costs are unclear, because warehouses moved away
from local markets to trade nodes. The increase or decrease in transport costs depends on
the function of a facility within goods supply chains, which is unknown.
Results from descriptive analysis also provide empirical evidence of spatial shifts in
the warehouse location choice. Most warehouses built in the early period (1951-1980) are
located around the 20-mile band from the CBD – traditional industrial zones around the
central urban areas and the vicinity of the port complex, whereas the recent development
has been around the 40- and 60-mile bands from the CBD – the Inland Empire areas near
Ontario, San Bernardino, and Riverside. Particularly, the distribution of the largest
warehouses (over 300k ft
2
) was very distinctive from all others.
These results raise two important research questions for future research: 1) what is
the potential impact of the concentration of large-scale warehousing facilities in an inland
area of the region; 2) will present or future logistics restructuring result in different
location patterns? First, over the last decade, there has been a substantial increase in the
number of large-scale distribution centers in the San Bernardino-Riverside area. For
example, 333 facilities (6.2% share of total) in SB-Riverside have supplied a total of 97
million ft
2
of warehouse floor area (14.5% share of total), and 85% of the floor area (209
facilities, 83 million ft
2
) have been constructed since 2001. The average size is
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
104

approximately 400,000 ft
2
, more than three times the average size of all warehouses (125
thousand ft
2
) of the region. Given that warehouses are major truck travel attractors, it is
important that potential negative externalities on the community are carefully examined.
Recent developments in warehouse management systems and the rise in e-
commerce and online shopping may change the location determinants of future
warehousing facilities. So far, the development of the state-of-the-art distribution facilities
has sought the location determinants documented in this study. In January, 2017, Amazon,
a leading online shopping company, promised an expansion of their supply chain capacity
with an air cargo hub in Kentucky for fast, reliable delivery speeds (Amazon, 2017). Also,
in the Greater Los Angeles region, the company has built multiple fulfillment centers, which
have apparently unique facility and location characteristics: very large-scale, high-tech
facility with direct access to the local market and an air cargo hub. The expansion of
facilities and changes in logistics practices might have entirely changed the geography of
urban freight movement. Therefore, an evaluation of whether these changes are a problem
worthy of planning/policy intervention is necessary.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
105

CHAPTER 5
CONCLUSIONS

The objectives of this dissertation were to examine how and why warehousing and
distribution centers (W&D) have decentralized from central urban areas to the periphery.
With three independent yet interrelated essays, I examined the spatial shift with three
different methodologies at three different geographic levels. In this chapter, I summarize
the findings.
In Essay 1, I proposed multiple spatial measures and compared the extent of
decentralization in four metro areas in California. When measured with respect to
establishments, only Los Angeles consistently showed decentralization between 2003 and
2013. When measured with respect to employment, all metro areas across most of the
measures showed decentralization. Each metro area showed a unique pattern of
decentralization, and I used three factors (metro size, economic structure, and physical
geography) to explain the variation.
Density, a proxy for demand, is correlated with metro size, and high density in large
metro areas implies high land prices. As large metropolitan areas continue to grow, the
more land-intensive businesses seek cheaper land away from the center. Also, large metro
areas are the hubs of global commerce. Hence, for those W&Ds serving non-local markets,
an alternative location with rich transport infrastructure yet with relatively lower land
prices is a good substitute. Furthermore, in San Francisco and San Diego, the physical
constraints (the Bay, the coast, borders with Mexico, and hilly terrain) have limited where
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
106

development can occur. In Los Angeles and Sacramento, where the physical geography is
not as restrictive, W&D development has been more of a function of land prices.
In Essay 2, I examined W&D decentralization between 2003 and 2013 in 48 US
metropolitan areas. I also evaluated the hypothesis that the variation in characteristics
across metro areas has resulted in differences in the extent of decentralization. Main
results of descriptive analysis suggest that 1) warehouses at the national level did
decentralize, but the correlation between decentralization and metro size (a proxy for land
price) as well as freight flow was non-linear; 2) large warehouses decentralized more than
small ones; and 3) large warehouses decentralized more in large metro areas than they did
in small metro areas, but this pattern was not observed in the case of small warehouses.
I further evaluated the linear relationship between decentralization and two
explanatory measures: variation in land price distribution (quantified by density gradient
and peak density) and variation in freight demand (quantified by freight flow and change in
the number of large warehouses). Model results suggest that the effect of freight demand
and land price on decentralization is significant and non-linear. Density gradient across
large metro areas had the greatest impact, whereas decentralization in small metro areas
was mainly a function of freight flow. In sum, decentralization is linked not only with very
large metro areas but also with very large warehouses.
In Essay 3, I examined the change in W&D location determinants to understand how
and why W&Ds have decentralized over the last several decades. I evaluated facility and
location characteristics of 5,364 W&Ds in the Greater Los Angeles region to explain the
variation in the probability of warehouse location choice. I compared estimated
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
107

parameters of multinomial logit models of three built year periods (1951-1980; 1981-2000;
2001-2016).
Model results suggest that there have been significant changes in location
determinants with respect to building size and built year. More recently built and large
warehouses, compared to all others, have prioritized lower land prices and higher airport
and intermodal terminal proximity over local market proximity. This suggests potential
savings in facility and inventory costs from lower land prices, economies of scale, and
inventory consolidation. Savings in transport costs are unclear, because W&Ds moved
away from local markets to trade nodes. The increase or decrease in transport costs
depends on the function of a facility within goods supply chains, which is unknown.
With these three essays, I identified three important planning implications. I found
that the current trend of W&D decentralization has led to the concentration of very large-
scale W&Ds in a small area of urban outskirts in the Los Angeles region. It is consistent
with the literature that W&Ds have been concentrating in places with rich transportation
infrastructure, growing population, and intense freight sector businesses (Bowen, 2008;
Cidell, 2010). However, there are also several critical factors that have not been properly
examined in freight and planning research, such as land use/zoning policies and
governmental measures to provide economic incentives (or dis-incentives) for
warehousing businesses. Indeed, how planning practitioners deal with the warehousing
industry, as a means to promote local economic development, varies widely even within
the same Los Angeles metropolitan area (Dablanc, 2015). Moreover, the effect of other
forms of external economies (e.g. agglomeration economies) have not yet been clearly
identified.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
108

I found a substantial growth in W&D scale. Will it continue to increase? I expect
W&D development in two directions. Logistics restructuring over the last several decades
has focused on throughput, reliability, and speed. On the one hand, I expect the trend of
W&D expansion, inventory consolidation, and spatial decentralization will continue as long
as population and consumption continue to grow. The distribution industry will continue
to optimize logistics costs by expanding throughput at the global scale and utilizing further
sophisticated strategies of supply chain management. On the other hand, I also expect
another shift in warehouse location because the recent growth in e-commerce and online
shopping has put tremendous emphasis on delivery speed (e.g. Just in Time (JIT) delivery)
(Dablanc and Ross, 2012). The logistics objectives (e.g. distribution network optimization,
distribution facility allocation, and sizing of warehouses) for JIT/instant delivery will differ
from those of regional distribution centers, hence the scale and spatial distribution of
facilities for JIT will differ. Therefore, it is important for planners and policy makers to
clearly understand and prepare necessary policies for the future demand of W&D space,
location, and operations.
I found W&D decentralization may be or may not have so many negative
consequences. Decentralization may be bad because facility consolidation and spatial
decentralization further away from central urban markets might increase freight VMT, as
argued in the literature. Another set of literature argues that inventory optimization and
system restructuring will offset negative impacts. However, the evidence from my research
only shows the distribution of W&Ds has changed further from urban markets and closer to
transportation nodes. If so, VMT increase or decrease depends on whether a facility serves
local or regional markets. Further research is necessary to evaluate this hypothesis.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
109

Decentralization may also be bad because W&Ds are likely to settle in marginalized
communities. Affluent communities will zone out locally undesirable land uses (LULUs),
meanwhile they will induce more W&D demand as they consume more goods. Accordingly,
W&D decentralization is an issue with implications for environmental justice. On the
contrary, decentralization may not be so bad. After W&Ds have been zoned out in central
urban areas, more expensive locations can be repurposed for other land uses, such as
housing, office, and mixed use. The W&D businesses in municipalities with scarce jobs are
utilized as a means for local economic development. Particularly, with the development of
W&Ds with sophisticated technologies, some W&Ds have become a feasible means to
provide jobs with decent wage and opportunities of vocational training in supply chain
management (Dablanc, 2015). In sum, W&D decentralization is an issue that merits further
research in many directions of planning scholarship and practice.

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
110

APPENDIX – CHAPTER 2 MATHEMATICAL FORMULAS OF SPATIAL MEASURES
A1 Measure 1. Decentralization
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝐶𝐵𝐷 =
∑ 𝑑 𝑗 ∗ 𝑒 𝑗 𝑁 𝑗 =1
𝐸

Where,
𝑑 𝑗 = distance to CBD from each W&D in ZIP Code (j) (n; j = 1, 2, …, N)
𝑒 𝑗 = the number of W&D establishments or employment in ZIP Code (j)
E = sum of 𝑒 𝑗

A2 Measure 2. Relative Decentralization
𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝑎𝑙𝑙 𝑋 =
∑ [
𝑁 𝑗 =1
∑ (𝑑 𝑖𝑗
×𝑥 𝑖 )
𝑛 𝑖 𝑋 ] ∗ 𝑒 𝑗 𝐸

Where,
𝑑 𝑖𝑗
= distance to ZIP Code (i) from each W&D in ZIP Code (j) or distance to census tract (i) from
each W&D in ZIP Code (j) (i = 1, 2, ..., n; j = 1, 2, …, N)
𝑥 𝑖 = employment in ZIP Code (i), or population in Census Tract (i)
X = sum of 𝑥 𝑖
𝑒 𝑗 = the number of W&D establishments or W&D employment in ZIP Code (j)
E = sum of 𝑒 𝑗

A3 Measure 3. Concentration
𝐺𝑖𝑛𝑖 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 = 2 ∫ [𝑥 − 𝐿 (𝑥 )]
1
0
𝑑𝑥

Where,
𝐿 (𝑥 ) is Lorenz curve of the number of W&Ds in each ZIP Code sorted in ascending order
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
111

APPENDIX – CHAPTER 3
A1 Summary of Hypothesis Tests
Table A. 1 Testing results of null hypotheses #1-7
Results Null hypothesis
Rejected 𝐻 0
1
: ∆𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 0
Accepted 𝐻 0
2
: ∆𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝐺𝑟𝑜𝑢𝑝 1
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝐺𝑟𝑜𝑢𝑝 2
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Rejected in 7
metro areas
𝐻 0
3
: 𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑖 ,2003
̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅
= 𝑊 &𝐷 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑖 ,2013
̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅ ̅̅

Rejected in 8
metro areas
𝐻 0
4−1
: 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,2003
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 𝐷𝑖𝑠 𝑡 𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝑖 ,2003
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Rejected in 13
metro areas
𝐻 0
4−2
: 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,2013
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝑖 ,2013
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Rejected 𝐻 0
5
: ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Rejected 𝐻 0
6−1
: ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 1
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 2
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Accepted 𝐻 0
6−2
: ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 1
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= ∆𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑚𝑎𝑙𝑙 𝑊 &𝐷𝑠
𝐺𝑟𝑜𝑢𝑝 2
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Rejected in 4
metro areas in
Group 1; none in
Group 2
𝐻 0
7
: 𝐷𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,2003
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
= 𝐷𝑖𝑠 𝑡 𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑙𝑎𝑟𝑔𝑒 𝑊 &𝐷𝑠
𝑖 ,2013
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

Where, 𝑥 ̅=
1
𝑛 ∑ 𝑥 𝑖 𝑛 𝑖 =1
(i=metro area 1, 2, …, 48)
All t-tests based on P<0.05

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
112

A2 The number, distribution, and decentralization of W&Ds by metro area
Table A. 2 Number of W&Ds and changes thereof, 2003-2013, Group 1
Rank Name
N of W&Ds
in 2003
N of W&Ds
in 2013
Change in
N of W&Ds
N of large
W&Ds
in 2003
N of large
W&Ds
In 2013
Change in
N of large
W&Ds
1 New York 938 914 -24 100 107 7
2 Los Angeles 775 1001 226 83 108 25
3 Chicago 465 530 65 48 70 22
4 Washington 285 318 33 34 38 4
5 San Francisco 305 349 44 23 23 0
6 Boston 290 294 4 26 29 3
7 Philadelphia 288 340 52 27 34 7
8 Dallas 338 402 64 40 52 12
9 Miami 193 235 42 8 19 11
10 Detroit 196 210 14 14 23 9
11 Houston 221 298 77 19 30 11
12 Atlanta 301 412 111 34 66 32
13 Seattle 162 227 65 15 20 5
14 Cleveland 148 150 2 9 9 0
15 Phoenix 129 168 39 14 23 9
16 San Diego 84 86 2 4 4 0
17 St. Louis 148 144 -4 9 8 -1
18 Pittsburgh 92 98 6 8 10 2
19 Denver 118 147 29 10 10 0
20 Portland 160 163 3 10 13 3
21 Tampa 63 79 16 3 3 0
22 Orlando 75 91 16 5 9 4
*Large warehouses are with 100 employees or more

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
113

Table A. 3 Number of W&Ds and changes thereof, 2003-2013, Group 2
Rank Name
N of W&Ds
in 2003
N of W&Ds
in 2013
Change in
N of W&Ds
N of large
W&Ds
in 2003
N of large
W&Ds
In 2013
Change in
N of large
W&Ds
23 Kansas City 159 153 -6 12 23 11
24 Columbus 208 195 -13 31 41 10
25 Cincinnati 112 122 10 16 18 2
26 Indianapolis 121 171 50 25 35 10
27 Milwaukee 101 98 -3 9 11 2
28 Charlotte 124 145 21 18 25 7
29 Salt Lake City 88 117 29 10 13 3
30 San Antonio 47 67 20 5 11 6
31 Virginia Beach 90 98 8 4 6 2
32 Las Vegas 51 80 29 3 7 4
33 New Orleans 77 83 6 7 6 -1
34 Nashville 116 121 5 21 25 4
35 Raleigh 76 77 1 8 8 0
36 Greensboro 88 88 0 12 14 2
37 Louisville 81 89 8 7 13 6
38 Grand Rapids 62 72 10 4 7 3
39 Buffalo 57 57 0 3 2 -1
40 Austin 38 50 12 4 1 -3
41 Birmingham 47 51 4 5 6 1
42 Greenville 101 97 -4 13 17 4
43 Rochester 45 48 3 2 2 0
44 Albany 54 48 -6 8 6 -2
45 Dayton 54 49 -5 9 12 3
46 Richmond 58 87 29 4 9 5
47 Tulsa 39 37 -2 1 2 1
48 Tucson 33 55 22 0 3 3

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
114

Table A. 4 Distribution of W&Ds and changes thereof, 2003-2013, Group 1
Rank Name
W&D dist.
in 2003
(mile)
W&D dist.
in 2013
(mile)
∆ dist.
2003-
2013
(mile)
Large
W&D dist.
in 2003
(mile)
Large
W&D dist.
in 2013
(mile)
Large
W&D
∆ dist.
2003-
2013
(mile)
1 New York 25.96 27.99 2.03 33.05 36.91 3.86
2 Los Angeles 25.12 28.69 3.57 25.33 35.95 10.62
3 Chicago 23.83 26.61 2.78 28.27 35.03 6.76
4 Washington 32.23 34.18 1.95 45.45 53.61 8.16
5 San Francisco 34.84 35.60 0.76 42.69 47.69 5.00
6 Boston 28.03 29.65 1.62 25.41 34.43 9.02
7 Philadelphia 21.06 22.31 1.25 24.09 26.88 2.79
8 Dallas 19.14 18.63 -0.51 20.76 22.86 2.10
9 Miami 25.25 28.92 3.66 18.93 44.07 25.13
10 Detroit 19.59 22.71 3.12 18.86 18.86 -0.01
11 Houston 14.72 16.25 1.53 16.62 20.22 3.60
12 Atlanta 19.69 22.09 2.40 22.06 26.00 3.94
13 Seattle 18.34 19.91 1.57 21.45 26.08 4.62
14 Cleveland 23.30 23.33 0.04 20.10 17.76 -2.35
15 Phoenix 11.19 10.77 -0.42 8.03 12.83 4.80
16 San Diego 13.47 12.89 -0.58 4.06 7.30 3.24
17 St. Louis 14.86 13.05 -1.81 11.12 21.82 10.69
18 Pittsburgh 15.91 17.17 1.26 18.28 24.17 5.89
19 Denver 10.63 10.21 -0.41 8.66 9.14 0.48
20 Portland 15.18 15.93 0.75 17.11 22.64 5.53
21 Tampa 10.51 10.35 -0.16 14.65 20.86 6.21
22 Orlando 12.24 12.01 -0.23 11.42 10.01 -1.41
*green highlight, if significant at P<0.05; gray highlight, if excluded because of small N

Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
115

Table A. 5 Distribution of W&Ds and changes thereof, 2003-2013, Group 2
Rank Name
W&D dist.
in 2003
(mile)
W&D dist.
in 2013
(mile)
∆ dist.
2003-
2013
(mile)
Large
W&D dist.
in 2003
(mile)
Large
W&D dist.
in 2013
(mile)
Large
W&D
∆ dist.
2003-
2013
(mile)
23 Kansas City 12.37 15.14 2.77 12.71 19.18 6.48
24 Columbus 13.62 13.92 0.29 18.15 13.10 -5.05
25 Cincinnati 14.60 14.33 -0.27 17.40 13.01 -4.39
26 Indianapolis 17.96 17.96 0.00 16.32 19.57 3.25
27 Milwaukee 14.75 19.25 4.50 21.46 21.77 0.31
28 Charlotte 14.23 16.14 1.91 18.78 21.02 2.24
29 Salt Lake City 14.36 14.55 0.19 13.55 12.63 -0.92
30 San Antonio 8.63 11.39 2.76 13.01 10.23 -2.78
31 Virginia Beach 18.47 17.15 -1.32 20.12 23.28 3.17
32 Las Vegas 6.67 12.76 6.09 7.11 7.13 0.02
33 New Orleans 12.08 18.57 6.50 21.79 20.68 -1.11
34 Nashville 17.58 18.87 1.29 23.10 25.90 2.80
35 Raleigh 19.36 19.45 0.09 21.77 30.73 8.96
36 Greensboro 19.62 17.34 -2.27 26.03 19.80 -6.23
37 Louisville 10.88 10.68 -0.20 10.02 12.52 2.49
38 Grand Rapids 13.77 15.24 1.48 15.44 12.86 -2.59
39 Buffalo 10.12 10.12 0.00 7.18 9.40 2.23
40 Austin 10.05 11.32 1.27 15.12 29.26 14.14
41 Birmingham 12.80 15.25 2.46 38.46 20.72 -17.74
42 Greenville 19.90 17.91 -1.99 26.12 25.78 -0.34
43 Rochester 16.08 15.73 -0.35 10.93 4.29 -6.64
44 Albany 17.32 17.60 0.28 25.87 33.44 7.57
45 Dayton 13.41 17.84 4.43 14.99 19.01 4.02
46 Richmond 11.70 12.60 0.90 15.31 19.67 4.35
47 Tulsa 12.18 14.68 2.51 10.28 20.41 10.12
48 Tucson 35.21 28.46 -6.75 - 25.87 25.87
*green highlight, if significant at P<0.05; gray highlight, if excluded because of small N
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
116

BIBLIOGRAPHY
Agarwal, A., Giuliano, G., & Redfearn, C. L. (2012). Strangers in our midst: the usefulness of
exploring polycentricity. The Annals of Regional Science, 48(2), 433-450.
Aljohani, K., & Thompson, R. G. (2016). Impacts of logistics sprawl on the urban
environment and logistics: Taxonomy and review of literature. Journal of Transport
Geography, 57, 255-263.
Allen, J., Browne, M., & Cherrett, T. (2012). Investigating relationships between road freight
transport, facility location, logistics management and urban form. Journal of
Transport Geography, 24, 45-57.
Alonso, W. (1967). A reformulation of classical location theory and its relation to rent
theory. Papers in Regional Science, 19(1), 23-44.
Amazon. (2017). Press Release. Amazon to create 2,000 jobs at new air cargo hub in
Kentucky. Available at:
http://www.businesswire.com/news/home/20170131006306/en/
Anas, A., Arnott, R., & Small, K. A. (1998). Urban spatial structure. Journal of Economic
Literature, 36(3), 1426-1464.
Anderson, S., Allen, J., & Browne, M. (2005). Urban logistics––how can it meet policy makers’
sustainability objectives? Journal of transport geography, 13(1), 71-81.
Andreoli, D., Goodchild, A., & Vitasek, K. (2010). The rise of mega distribution centers and
the impact on logistical uncertainty. Transportation Letters, 2(2), 75-88.
Arauzo Carod, J. M., & Manjón Antolí n, M. C. (2004). Firm size and geographical aggregation:
an empirical appraisal in industrial location. Small Business Economics, 22(3), 299-
312.
Arauzo, J-M., Liviano-Solis, D., and Manjon-Antolin, M. (2010). Empirical studies in
industrial location: an assessment of their methods and results. Journal of Regional
Science, 50(3), 685-711.
Bartik, T. J. (1985). Business location decisions in the United States: Estimates of the effects
of unionization, taxes, and other characteristics of states. Journal of Business &
Economic Statistics, 3(1), 14-22.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
117

Ben-Akiva, M., & Boccara, B. (1995). Discrete choice models with latent choice sets.
International Journal of Research in Marketing, 12(1), 9-24.
Bowen Jr., J., (2008). Moving places: the geography of warehousing in the US. Journal of
Transport Geography, 16, 379–387.
Buttimer Jr, R. J., Rutherford, R. C., & Witten, R. (1997). Industrial warehouse rent
determinants in the Dallas/Fort Worth Area. Journal of Real Estate Research, 13(1),
47-56.
Carlton, D. (1979). Why New Firms Locate Where They Do: An Econometric Model, in W.
Wheaton (ed.), Interregional Movements and Regional Growth. Washington: The
Urban Institute.
Carlton, D. W. (1983). The location and employment choices of new firms: An econometric
model with discrete and continuous endogenous variables. The Review of Economics
and Statistics, 440-449.
Castells, M. (2011). The rise of the network society: The information age: Economy, society,
and culture (Vol. 1). John Wiley & Sons.
Christopherson, S., & Belzer, M. (2009). The next move: metropolitan regions and the
transformation of the freight transport and distribution system. In: Pindus, N., Wial,
H., Wolman, H. (Eds.), Urban and Regional Policy and its Effects, vol. 2. Brookings,
Washington, DC, pp. 194–222.
Cidell, J. (2010). Concentration and decentralization: The new geography of freight
distribution in US metropolitan areas. Journal of Transport Geography, 18(3), 363-
371.
Cidell, J. (2011). Distribution centers among the rooftops: the global logistics network
meets the suburban spatial imaginary. International Journal of Urban and Regional
Research, 35(4), 832-851.
Clark, C. (1951). Urban population densities. Journal of the Royal Statistical Society. Series A
(General), 114(4), 490-496.
Crainic, T. G., Ricciardi, N., & Storchi, G. (2004). Advanced freight transportation systems for
congested urban areas. Transportation Research Part C: Emerging Technologies,
12(2), 119-137.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
118

Cushman & Wakefield. (2015). North American Industrial Real Estate Forecast 2015-2017.
A Cushman & Wakefield Research Publication. Available at:
http://www.cushmanwakefield.com/en/research-and-insight/2015/2015-17-
industrial-forecast/
Dablanc, L., (2013). City logistics. In: Rodrigue, J.P., Notteboom, T., Shaw, J. (Eds.), The SAGE
Handbook of Transport Studies. SAGE, London, England.
Dablanc, L. (2015). How Do Planning Practitioners Address Freight Transport and Logistics
Sprawl? Case Study in Los Angeles. Presented at the 6
th
METRANS International
Urban Freight Conference, Long Beach, CA.
Dablanc, L., Diziain, D., & Levifve, H. (2011). Urban freight consultations in the Paris region.
European Transport Research Review, 3(1), 47-57.
Dablanc, L., and Gardrat, M. (2015). The new Paris urban freight survey – method, main
results, and potential use for urban freight elsewhere. Presented at the 6th
METRANS International Urban Freight Conference, Long Beach, CA.
Dablanc, L. and Ross, C., 2012. Atlanta: A Mega Logistics Center in the Piedmont Atlantic
Megaregion (PAM), Journal of Transport Geography, 24, 432-442.
Dablanc, L., Ogilvie, S., & Goodchild, A. (2014). Logistics Sprawl: Differential Warehousing
Development Patterns in Los Angeles, California, and Seattle, Washington.
Transportation Research Record: Journal of the Transportation Research Board,
(2410), 105-112.
Dablanc, L., & Rakotonarivo, D. (2010). The impacts of logistics sprawl: How does the
location of parcel transport terminals affect the energy efficiency of goods’
movements in Paris and what can we do about it? Procedia-Social and Behavioral
Sciences, 2(3), 6087-6096.
Dhooma, J., & Baker, P. (2012). An exploratory framework for energy conservation in
existing warehouses. International Journal of Logistics Research and Applications,
15(1), 37-51.
FAA (Federal Aviation Administration), Passenger boarding and all-cargo data for U.S.
airports. Accessible at:
http://www.faa.gov/airports/planning_capacity/passenger_allcargo_stats/passeng
er/
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
119

Gabaix, X. (1999). Zipf's law for cities: an explanation. The Quarterly Journal of Economics,
114(3), 739-767.
Giuliano, G., Hou, Y., Kang, S., and Shin, E. (2015). Accessibility, Location and Employment
Center Growth. METRANS Project 11-6, METRANS Transportation Center, Sol Price
School of Public Policy, University of Southern California, Los Angeles, CA. Available
at: https://www.metrans.org/research/11-06-accessibility-location-and-
employment-center-growth
Giuliano, G., & Kang, S. (2016). Spatial Dynamics of the Logistics Industry and Implications
for Freight Flows. A Research Report from the National Center for Sustainable
Transportation. USC-CT-TO-004. Available at: https://ncst.ucdavis.edu/project/usc-
ct-to-004/
Giuliano, G., Kang, S., Yuan, Q., & Hutson, N. (2015). The Freight Landscape: Using
Secondary Data Sources to Describe Metropolitan Freight Flows. METRANS UTC 1-
1B Final Report, METRANS Transportation Center, Sol Price School of Public Policy,
University of Southern California, Los Angeles, CA. Available at:
https://www.metrans.org/research/usc-1-1b-freight-landscape-using-secondary-
data-sources-describe-metropolitan-freight-flows
Giuliano, G., Redfearn, C., Agarwal, A., Li, C., & Zhuang, D. (2007). Employment
concentrations in Los Angeles, 1980–2000. Environment and Planning A, 39(12),
2935-2957.
Giuliano, G., & Small, K. A. (1991). Subcenters in the Los Angeles region. Regional Science
and Urban Economics, 21(2), 163-182.
Glaeser, E. and Kohlhase, J. (2004) Cities, regions and the decline of transport costs. Papers
in Regional Science, 83: 197-228.
Goodchild, A., and Dubie, M. (2016). Logistics sprawl in Chicago, Illinois. Presented at the
14th World Conference on Transport Research, Shanghai, China.
Guimaraes, P., Figueiredo, O., & Woodward, D. (2004). Industrial location modeling:
extending the random utility framework. Journal of Regional Science, 44(1), 1-20.
Hall, P., Hesse, M., & Rodrigue, J-P. (2006). Reexploring the interface between economic and
transport geography. Environment and Planning A, 38(8), 1401-1408.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
120

Heitz, A., & Dablanc, L. (2015). Logistics Spatial Patterns in Paris: Rise of Paris Basin as
Logistics Megaregion. Transportation Research Record: Journal of the Transportation
Research Board, (2477), 76-84.
Hensher, D. A., & Greene, W. H. (2002). Specification and estimation of the nested logit
model: alternative normalisations. Transportation Research Part B: Methodological,
36(1), 1-17.
Hesse, M., & Rodrigue, J. P. (2004). The transport geography of logistics and freight
distribution. Journal of Transport Geography, 12(3), 171-184.
Hesse, M. (2002). Location matters. Access. Transportation Research at the University of
California, 1(21), 22–26.
Hesse, M. (2004). Land for logistics: locational dynamics, real estate markets and political
regulation of regional distribution complexes. Tijdschrift voor Economische en
Sociale Geographie, 95(2), 162–173.
Hesse, M. (2006). Global chain, local pain: Regional implications of global distribution
networks in the German north range. Growth and Change, 37(4), 570-596.
Hesse, M. (2007). The System of Flows and the Restructuring of Space Elements of a
Geography of Distribution (Das System der Ströme und die Re-Strukturierung des
Raumes. Elemente einer Geographie der Distribution). Erdkunde, 1-12.
Hesse, M., & Rodrigue, J. P. (2004). The transport geography of logistics and freight
distribution. Journal of Transport Geography, 12(3), 171-184.
iTrans Consulting. (2004). Goods Movement in Central Ontario: Trends and Issues. Toronto:
Ministry of Transportation of Ontario. Accessible at:
http://www.ontla.on.ca/library/repository/mon/9000/249139.pdf
Jakubicek, P. & Woudsma, C. (2011). Proximity, land, labor and planning? Logistics industry
perspectives on facility location. Transportation Letters, 3:3, 161-173.
Knowles, R. D. and Hall. P. (1998). Transport Deregulation and Privatization. In: Hoyle, B. S.,
R. D. Knowles (Eds.). Modern Transport Geography. Wiley, New York.
Kohn, C., & Brodin, M. H. (2008). Centralised distribution systems and the environment:
how increased transport work can decrease the environmental impact of logistics.
International Journal of Logistics: Research and Applications, 11(3), 229-245.
Krugman, P. (1991). Geography and Trade. Cambridge, MA: MIT Press.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
121

Lasserre, F. (2004). Logistics and the Internet: transportation and location issues are
crucial in the logistics chain. Journal of Transport Geography, 12(1), 73-84.
Lavassani, K., Movahedi, B., & Kumar, V. (2009). Transition to B2B e-marketplace enabled
supply chain: readiness assessment and success factors. The International Journal of
Technology, Knowledge and Society, 5(3), 75-88.
Manski, C.F., 1977. Structure of random utility models. Theory and Decision, 8, 229–254.
McConnell, K.E. and W.Tseng (2000), Some Preliminary Evidence on Sampling of
Alternatives with the Random Parameters Logit. Marine Resource Economics, 14,
317–332.
McDonald, J. F. (1989). Econometric studies of urban population density: A survey. Journal
of Urban Economics, 26(3), 361-385.
McFadden, D. (1974). Conditional Logit Analysis of Qualitative Choice Behaviour, in P.
Zarembka (ed.) Frontiers in Econometrics. Academic Press, pp. 105–142.
McFadden, D. (1978). Modeling the choice of residential location. Transportation Research
Record, 673.
McFadden, D. (2001). Economic Choices, American Economic Review, 91, 351–378.
McKinnon, A. (1983). The development of warehousing in England. Geoforum 14(4), 389-
399.
McKinnon, A. (2009). The present and future land requirements of logistical activities. Land
Use Policy, 26, 293-301.
Movahedi B., Lavassani, K. & Kumar, V. (2009). Transition to B2B e-Marketplace Enabled
Supply Chain: Readiness Assessment and Success Factors, The International Journal
of Technology, Knowledge and Society, 5(3): 75-88.
Nerella, S., & Bhat, C. (2004). Numerical analysis of effect of sampling of alternatives in
discrete choice models. Transportation Research Record: Journal of the
Transportation Research Board, (1894), 11-19.
Nguyen, Y., & Sano, K. (2010). Location choice model for logistic firms with consideration of
spatial effects. Transportation Research Record: Journal of the Transportation
Research Board, (2168), 17-23.
Notteboom, T. E., & Rodrigue, J-P. (2005). Port regionalization: towards a new phase in port
development. Maritime Policy & Management, 32(3), 297-313.
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
122

Pellegrini, P. A., Fotheringham, A. S., & Lin, G. (1997). An empirical evaluation of parameter
sensitivity to choice set definition in shopping destination choice models. Papers in
Regional Science, 76(2), 257-284.
Pozsgay, M., & Bhat, C. (2001). Destination choice modeling for home-based recreational
trips: analysis and implications for land use, transportation, and air quality planning.
Transportation Research Record: Journal of the Transportation Research Board, 1777,
47-54.
Rodrigue, J.-P. (2004). Freight, gateways and mega-urban regions: The logistical integration
of the Bostwash Corridor. Tijdschrift voor economische en sociale geografie, 95, 147-
161.
Rodrigue, J.-P. (2006). Transportation and the Geographical and Functional Integration of
Global Production Networks. Growth and Change, 37(4), 510-525.
Rodrigue, J.-P. (2008) The Thruport concept and transmodal rail freight distribution in
North America. Journal of Transport Geography 16(4), 233–46.
Rodrigue, J.-P., Slack, B., & Comtois, C. (2001). Green Logistics (The Paradoxes of). In: A. M.
Brewer, K. J. Button and D. A. Hensher (eds) (2001) The Handbook of Logistics and
Supply-Chain Management, Handbooks in Transport #2, London: Pergamon/ Elsevier.
Sakai, T., Kawamura, K., & Hyodo, T. (2015). Locational dynamics of logistics facilities:
Evidence from Tokyo. Journal of Transport Geography, 46, 10-19.
Scott, D. M., & He, S. Y. (2012). Modeling constrained destination choice for shopping: a GIS-
based, time-geographic approach. Journal of Transport Geography, 23, 60-71.
Sivitanidou, R. (1996). Warehouse and distribution facilities and community attributes: an
empirical study. Environment and Planning A, 28(7), 1261-1278.
Slack, B. (1998). Intermodal Transportation. In: Hoyle, B., Knowles, R. (Eds.), Modern
Transport Geography, second ed. Wiley, London, pp. 263–289.
Strocko, E., M. Sprung, L. Nguyen, C. Rick and J. Sedor. (2013). Freight Facts and Figures
2013. Publication FHWA-HOP-14-004. FHWA, U.S. Department of Transportation.
Supply Chain Brain. (2010). Trends in Industrial Real Estate: Interview with Prologis CEO
Walt Rakowich. Retrieved January 20, 2010, from
http://www.supplychainbrain.com/content/videos/cscmp-09/the-state-of-
industrialreal-estate/
Unraveling Decentralization of Warehousing and Distribution Centers Kang (2017)
123

Swait, J.D., & Ben-Akiva, M. (1987). Incorporating random constraints in discrete models of
choice set generation. Transportation Research Part B: Methodological 21, 91–102.
US Department of Transport: Federal Highway Administration (FHWA), 2012. FHWA
Freight and Land Use Handbook. Retrieved from
http://ops.fhwa.dot.gov/publications/fhwahop12006/fhwahop12006.pdf.
Van den Heuvel, F., de Langen, P., van Donselaar, K., and Fransoo, J. (2013). Spatial
concentration and location dynamics in logistics: the case of a Dutch province.
Journal of Transport Geography, 28, 39-48.
Weber, A., (1929). Theory of the Location of Industry. (trans, C. Friedrich), University of
Chicago Press.
Woudsma, C., Jensen, J. F., Kanaroglou, P., & Maoh, H. (2008). Logistics land use and the city:
A spatial–temporal modeling approach. Transportation Research Part E: Logistics
and Transportation Review, 44(2), 277-297.
Woudsma, C.G., and Jakubicek, P. (2016). Logistics land use patterns in metropolitan
Canada. Presented at 14
th
World Conference on Transport Research, Shanghai,
China.
Woudsma, C., Jakubicek, P., & Dablanc, L. (2016). Logistics sprawl in North America:
methodological issues and a case study in Toronto. Transportation Research
Procedia, 12, 474-488.
Wygonik, E., Bassok, A., Goodchild, A., McCormack, E., & Carlson, D. (2015). Smart growth
and goods movement: emerging research agendas. Journal of Urbanism:
International Research on Placemaking and Urban Sustainability, 8(2), 115-132.

Abstract (if available)

Abstract This dissertation examines how and why warehousing and distribution centers have decentralized from central urban areas to the periphery. The research verifies theoretical discourses on the factors that explain warehouse location change and generates consistent and robust empirical evidence with descriptive analysis and estimation of econometric models. Through three independent yet interrelated empirical studies, this dissertation investigates how the restructuring of the logistics industry has reshaped the spatial distribution of warehousing facilities at the sub-metropolitan level. Findings suggest that freight demand and land prices are two main factors for decentralization. To transport large volumes of freight, the logistics industry has built large-scale warehousing facilities on urban outskirts where land is readily available at relatively lower costs. This process of decentralization involves tradeoffs of logistics costs between land and transport costs.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Location of warehouses and environmental justice: Three essays

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Urban air pollution and environmental justice: three essays

PDF

Essays on the economics of cities

PDF

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

PDF

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

PDF

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

PDF

Productive frictions and urbanism in transition: planning lessons from traffic flows and urban street life in Ho Chi Minh City, Vietnam

PDF

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

PDF

Evaluating city logistics using two-level location routing modeling and SCPM simulation

PDF

Who learns where: understanding the equity implications of charter school reform in the District of Columbia

PDF

The built environment, tour complexity, and active travel

PDF

Intelligent urban freight transportation

PDF

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

PDF

Urban consumer amenities and their accessibility

PDF

Property and labor formalization in the age of the sharing economy: Airbnb, housing affordability, and entrepreneurship in Havana

PDF

Community development agreements: addressing inequality through urban development projects

PDF

Household mobility and neighborhood impacts

PDF

Environmental justice in real estate, public services, and policy

PDF

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

Asset Metadata

Creator Kang, Sanggyun (author)

Core Title Unraveling decentralization of warehousing and distribution centers: three essays

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Publication Date 07/24/2017

Defense Date 06/16/2017

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

Tag distribution centers,firm location,freight transportation,logistics sprawl,OAI-PMH Harvest,transportation planning,urban freight movement,urban logistics,warehouse,warehouse location choice,warehousing decentralization

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Giuliano, Genevieve (committee chair), Boarnet, Marlon (committee member), Ioannou, Petros (committee member)

Creator Email sanggyuk@usc.edu,sggkang@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-413570

Unique identifier UC11214524

Identifier etd-KangSanggy-5622.pdf (filename),usctheses-c40-413570 (legacy record id)

Legacy Identifier etd-KangSanggy-5622-1.pdf

Dmrecord 413570

Document Type Dissertation

Format application/pdf (imt)

Rights Kang, Sanggyun

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